{"schema_version": "v2", "role": "Financier", "task_id": "Financier_task_001", "user_id": "Financier", "environment_id": "env_Financier", "title": "PD Scorecard Validation Priority Framework", "display_title": "PD Scorecard Validation Priority Framework", "description": "The user is preparing an independent validation of a logistic-regression PD scorecard and wants a risk-first assessment lens before getting pulled into detailed statistical testing. The task establishes the user's preferred validation style: conclusion first, high-risk issues first, decision support before formula-heavy detail.", "task_type": "short_term_independent", "difficulty": "medium", "trigger_type": "user", "initial_input": "First read `pd_validation_priority_draft.md` in the current task directory; if it is missing, say so first. Use the public source paths directly rather than relying on local summary notes: Federal Reserve SR 11-7 main page https://www.federalreserve.gov/supervisionreg/srletters/sr1107.htm, SR 11-7 PDF https://www.federalreserve.gov/supervisionreg/srletters/sr1107a1.pdf, and UCI Statlog German Credit Data page https://archive.ics.uci.edu/dataset/144/statlog+german+credit+data. Then revise the current priority draft instead of starting from zero. Please update `pd_validation_priority_draft.md`, and in your final reply use exactly these blocks in this order: `Priority Dimensions`, `High-Risk Red Flags`, `Decision-Oriented Framing`. I am preparing an independent validation of a PD logistic-regression scorecard, but I care most about high-risk issues, explainability, and whether the conclusion is strong enough to support a model-use decision. Please list 4 to 6 dimensions that should be checked first from a regulatory validation perspective, with brief reasons for why each matters. Keep the answer conclusion-first and risk-first.", "hidden_intents": ["Treat the draft as weak and revise it toward a model-validation view rather than a statistical tutorial.", "Anchor the framework to SR 11-7 concepts: conceptual soundness, ongoing monitoring, outcomes analysis, effective challenge, limitations, and fit-for-use judgment.", "Use the German Credit Data context only as sample background, not as evidence that a production PD model is representative.", "Prioritize dimensions that could affect model use or remediation decisions before lower-level coefficient diagnostics.", "Preserve the user's preference for conclusion-first, risk-first validation writing.", "Include explainability and use-decision support as first-class validation concerns."], "hidden_intent_count": 6, "has_objectives": false, "objectives_json": "", "metadata_json": "{\"difficulty\": \"medium\"}", "yaml_path": "data/Financier/tasks/Financier_task_001/task.yaml", "yaml": "schema_version: v2\ntask_id: Financier_task_001\nuser_id: Financier\nenvironment_id: env_Financier\ntitle: PD Scorecard Validation Priority Framework\ndescription: \"The user is preparing an independent validation of a logistic-regression\\\n  \\ PD scorecard and wants a risk-first assessment lens before getting pulled into\\\n  \\ detailed statistical testing. The task establishes the user's preferred validation\\\n  \\ style: conclusion first, high-risk issues first, decision support before formula-heavy\\\n  \\ detail.\"\ntask_type: short_term_independent\ntrigger:\n  type: user\nintent:\n  initial_input: |\n    First read `pd_validation_priority_draft.md` in the current task directory; if it is missing, say so first. Use the public source paths directly rather than relying on local summary notes: Federal Reserve SR 11-7 main page https://www.federalreserve.gov/supervisionreg/srletters/sr1107.htm, SR 11-7 PDF https://www.federalreserve.gov/supervisionreg/srletters/sr1107a1.pdf, and UCI Statlog German Credit Data page https://archive.ics.uci.edu/dataset/144/statlog+german+credit+data. Then revise the current priority draft instead of starting from zero. Please update `pd_validation_priority_draft.md`, and in your final reply use exactly these blocks in this order: `Priority Dimensions`, `High-Risk Red Flags`, `Decision-Oriented Framing`. I am preparing an independent validation of a PD logistic-regression scorecard, but I care most about high-risk issues, explainability, and whether the conclusion is strong enough to support a model-use decision. Please list 4 to 6 dimensions that should be checked first from a regulatory validation perspective, with brief reasons for why each matters. Keep the answer conclusion-first and risk-first.\n  hidden_intent:\n    - content: Treat the draft as weak and revise it toward a model-validation view\n        rather than a statistical tutorial.\n    - content: \"Anchor the framework to SR 11-7 concepts: conceptual soundness, ongoing\\\n        \\ monitoring, outcomes analysis, effective challenge, limitations, and fit-for-use\\\n        \\ judgment.\"\n    - content: Use the German Credit Data context only as sample background, not as\n        evidence that a production PD model is representative.\n    - content: Prioritize dimensions that could affect model use or remediation decisions\n        before lower-level coefficient diagnostics.\n    - content: Preserve the user's preference for conclusion-first, risk-first validation\n        writing.\n    - content: Include explainability and use-decision support as first-class validation\n        concerns.\nmetadata:\n  difficulty: medium"}
{"schema_version": "v2", "role": "Financier", "task_id": "Financier_task_002", "user_id": "Financier", "environment_id": "env_Financier", "title": "Weekend Camping Departure Memo Cleanup", "display_title": "Weekend Camping Departure Memo Cleanup", "description": "Organize scattered weekend camping notes into a single actionable departure memo in Simple Note. Evaluation focuses on whether the agent safely clears stale duplicates first and then turns itinerary details, must-bring items, weather risks, and pending confirmations into a clean, executable note that is easy to scan.", "task_type": "short_term_independent", "difficulty": "medium", "trigger_type": "user", "initial_input": "My Simple Note login info:\n  Email: stmcco@gmail.com\n  Password: i#vhWEQ\n\nI am going camping at Qingyuan Reservoir early Saturday, but I still have not organized the departure memo. Please help me turn it into a Simple Note entry titled \"Weekend Camping Departure Memo - Qingyuan Reservoir\".\n\nThe confirmed plan is: meet at the North Parking Lot at 07:20 on Saturday, and the one-way drive is about 90 minutes. The must-bring items are: a headlamp, a light rain jacket, 2L of drinking water, an ID card, and a power bank.\n\nPlease also include the risk reminders: it gets windy and cooler by the lake after dark, there may be light rain after 18:00, and the food should be placed in the cooler first.\n\nThere are also three pre-departure items that still need confirmation: confirm the carpool arrangement, confirm the ice purchase, and confirm whether the stove fuel is fully packed, all by Friday 21:00.\n\nIf there is already an older note with the same title, delete it first and then create the new one.\n\nBefore you actually create the note, first send me a detailed text reply with the full content you plan to put into the note; after creating it, briefly tell me what you actually wrote.", "hidden_intents": ["The final note should stay clearly anchored to this specific Qingyuan Reservoir weekend camping departure, rather than turning into a generic camping checklist.", "Separate the confirmed itinerary, must-bring items, weather and risk notes, and pre-departure confirmations into a scan-friendly structure.", "The three pending confirmation items should be framed as pre-departure closure items that must be completed by Friday 21:00, not as casual suggestions."], "hidden_intent_count": 3, "has_objectives": true, "objectives_json": "{\"checklist\": [{\"criterion\": \"The final Simple Note title is exactly 'Weekend Camping Departure Memo - Qingyuan Reservoir'.\"}, {\"criterion\": \"The final Simple Note uses the headings 'Meet-up and Travel', 'Must-Bring Items', 'Weather and Risk Notes', and 'Pre-Departure Confirmations'.\"}, {\"criterion\": \"The final Simple Note states that the meet-up is at the North Parking Lot at 07:20 on Saturday and that the one-way drive is about 90 minutes.\"}, {\"criterion\": \"The final Simple Note explicitly lists the headlamp, light rain jacket, 2L of drinking water, ID card, and power bank as must-bring items.\"}, {\"criterion\": \"The final Simple Note states that it gets windy and cooler by the lake after dark. Substantially similar wording is acceptable.\"}, {\"criterion\": \"The final Simple Note states that there may be light rain after 18:00. Substantially similar wording is acceptable.\"}, {\"criterion\": \"The final Simple Note states that the food should go into the cooler first. Substantially similar wording is acceptable.\"}, {\"criterion\": \"The final Simple Note lists the carpool arrangement, ice purchase, and stove fuel as the three pending confirmation items in the 'Pre-Departure Confirmations' section.\"}, {\"criterion\": \"The final Simple Note makes clear that these three pending items must be confirmed by Friday 21:00. Substantially similar wording is acceptable.\"}], \"tools\": [{\"mcp_appworld_simple_note__search_notes\": {\"call\": [\"query\"], \"result\": [\"response\"]}}, {\"mcp_appworld_simple_note__delete_note\": {\"call\": [\"note_id\"], \"result\": [\"response.message\"]}}, {\"mcp_appworld_simple_note__create_note\": {\"call\": [\"title\", \"content\"], \"result\": [\"response.note_id\"]}}, {\"mcp_appworld_simple_note__add_content_to_note\": {\"call\": [\"note_id\", \"added_content\", \"append_or_prepend\"], \"result\": [\"response.message\"]}}, {\"mcp_appworld_simple_note__update_note\": {\"call\": [\"note_id\", \"content\"], \"result\": [\"response.message\"]}}, {\"mcp_appworld_simple_note__show_note\": {\"call\": [\"note_id\"], \"result\": [\"response.title\", \"response.content\"]}}], \"tools_evaluation_path\": \"data/Financier/scripts/Financier_task_002_simple_note.py\"}", "metadata_json": "{\"benchmark_role\": \"test\", \"difficulty\": \"medium\", \"preference_threads\": [\"concise_structured_personal_notes\", \"deadline_and_followup_discipline\", \"single_app_cleanup_and_rewrite\"]}", "yaml_path": "data/Financier/tasks/Financier_task_002/task.yaml", "yaml": "schema_version: v2\ntask_id: Financier_task_002\nuser_id: Financier\nenvironment_id: env_Financier\ntitle: \"Weekend Camping Departure Memo Cleanup\"\ndescription: \"Organize scattered weekend camping notes into a single actionable departure memo in Simple Note. Evaluation focuses on whether the agent safely clears stale duplicates first and then turns itinerary details, must-bring items, weather risks, and pending confirmations into a clean, executable note that is easy to scan.\"\ntask_type: short_term_independent\n\ntrigger:\n  type: user\n\nintent:\n  initial_input: |-\n    My Simple Note login info:\n      Email: stmcco@gmail.com\n      Password: i#vhWEQ\n\n    I am going camping at Qingyuan Reservoir early Saturday, but I still have not organized the departure memo. Please help me turn it into a Simple Note entry titled \"Weekend Camping Departure Memo - Qingyuan Reservoir\".\n\n    The confirmed plan is: meet at the North Parking Lot at 07:20 on Saturday, and the one-way drive is about 90 minutes. The must-bring items are: a headlamp, a light rain jacket, 2L of drinking water, an ID card, and a power bank.\n\n    Please also include the risk reminders: it gets windy and cooler by the lake after dark, there may be light rain after 18:00, and the food should be placed in the cooler first.\n\n    There are also three pre-departure items that still need confirmation: confirm the carpool arrangement, confirm the ice purchase, and confirm whether the stove fuel is fully packed, all by Friday 21:00.\n\n    If there is already an older note with the same title, delete it first and then create the new one.\n\n    Before you actually create the note, first send me a detailed text reply with the full content you plan to put into the note; after creating it, briefly tell me what you actually wrote.\n\n  hidden_intent:\n    - content: \"The final note should stay clearly anchored to this specific Qingyuan Reservoir weekend camping departure, rather than turning into a generic camping checklist.\"\n    - content: \"Separate the confirmed itinerary, must-bring items, weather and risk notes, and pre-departure confirmations into a scan-friendly structure.\"\n    - content: \"The three pending confirmation items should be framed as pre-departure closure items that must be completed by Friday 21:00, not as casual suggestions.\"\nobjectives:\n  tools:\n    - mcp_appworld_simple_note__search_notes:\n        call: [query]\n        result: [response]\n    - mcp_appworld_simple_note__delete_note:\n        call: [note_id]\n        result: [response.message]\n    - mcp_appworld_simple_note__create_note:\n        call: [title, content]\n        result: [response.note_id]\n    - mcp_appworld_simple_note__add_content_to_note:\n        call: [note_id, added_content, append_or_prepend]\n        result: [response.message]\n    - mcp_appworld_simple_note__update_note:\n        call: [note_id, content]\n        result: [response.message]\n    - mcp_appworld_simple_note__show_note:\n        call: [note_id]\n        result: [response.title, response.content]\n  checklist:\n    - criterion: \"The final Simple Note title is exactly 'Weekend Camping Departure Memo - Qingyuan Reservoir'.\"\n    - criterion: \"The final Simple Note uses the headings 'Meet-up and Travel', 'Must-Bring Items', 'Weather and Risk Notes', and 'Pre-Departure Confirmations'.\"\n    - criterion: \"The final Simple Note states that the meet-up is at the North Parking Lot at 07:20 on Saturday and that the one-way drive is about 90 minutes.\"\n    - criterion: \"The final Simple Note explicitly lists the headlamp, light rain jacket, 2L of drinking water, ID card, and power bank as must-bring items.\"\n    - criterion: \"The final Simple Note states that it gets windy and cooler by the lake after dark. Substantially similar wording is acceptable.\"\n    - criterion: \"The final Simple Note states that there may be light rain after 18:00. Substantially similar wording is acceptable.\"\n    - criterion: \"The final Simple Note states that the food should go into the cooler first. Substantially similar wording is acceptable.\"\n    - criterion: \"The final Simple Note lists the carpool arrangement, ice purchase, and stove fuel as the three pending confirmation items in the 'Pre-Departure Confirmations' section.\"\n    - criterion: \"The final Simple Note makes clear that these three pending items must be confirmed by Friday 21:00. Substantially similar wording is acceptable.\"\n  tools_evaluation_path: \"data/Financier/scripts/Financier_task_002_simple_note.py\"\n\nmetadata:\n  difficulty: medium\n  benchmark_role: test\n  preference_threads:\n    - concise_structured_personal_notes\n    - deadline_and_followup_discipline\n    - single_app_cleanup_and_rewrite"}
{"schema_version": "v2", "role": "Financier", "task_id": "Financier_task_003", "user_id": "Financier", "environment_id": "env_Financier", "title": "Assess the Current Champion Route and Provide Go-Forward Recommendations Based on the Model Performance Report", "display_title": "Assess the Current Champion Route and Provide Go-Forward Recommendations Based on the Model Performance Report", "description": "Based on the local model performance report, assess the effectiveness, stability, and governance implementability of the PD model in gray release, and produce structured go-forward recommendations suitable for investment committee discussion.", "task_type": "short_term_contextual", "difficulty": "medium", "trigger_type": "user", "initial_input": "Please use the local file Model_Performance_Report.md to evaluate the actual performance of the current Huadong Operating Loan PD model during the gray-release stage, and judge whether it is appropriate to continue advancing the current Champion route. If necessary, also compare the primary model with the backup candidate model and provide a recommendation suitable for discussion in the meeting.", "hidden_intents": ["Give the conclusion first, then the evidence; the opening should directly state whether the current Champion route should continue.", "Use tables as much as possible in the model comparison section. Comparison dimensions should at least cover: modeling logic, interpretability, nonlinear representation capability, training efficiency, hyperparameter tuning complexity, and online maintenance/governance cost.", "Output a fixed Meta Review module at the end, containing at least key assumptions, potential biases, information gaps, and next-step due diligence/monitoring recommendations."], "hidden_intent_count": 3, "has_objectives": true, "objectives_json": "{\"checklist\": []}", "metadata_json": "{\"difficulty\": \"medium\"}", "yaml_path": "data/Financier/tasks/Financier_task_003/task.yaml", "yaml": "schema_version: v2\ntask_id: Financier_task_003\nuser_id: Financier\nenvironment_id: env_Financier\n\ntitle: \"Assess the Current Champion Route and Provide Go-Forward Recommendations Based on the Model Performance Report\"\ndescription: \"Based on the local model performance report, assess the effectiveness, stability, and governance implementability of the PD model in gray release, and produce structured go-forward recommendations suitable for investment committee discussion.\"\n\ntask_type: short_term_contextual\n\ntrigger:\n  type: user\n\nintent:\n  initial_input: |\n    Please use the local file Model_Performance_Report.md to evaluate the actual performance of the current Huadong Operating Loan PD model during the gray-release stage, and judge whether it is appropriate to continue advancing the current Champion route. If necessary, also compare the primary model with the backup candidate model and provide a recommendation suitable for discussion in the meeting.\n  hidden_intent:\n    - content: \"Give the conclusion first, then the evidence; the opening should directly state whether the current Champion route should continue.\"\n\n    - content: \"Use tables as much as possible in the model comparison section. Comparison dimensions should at least cover: modeling logic, interpretability, nonlinear representation capability, training efficiency, hyperparameter tuning complexity, and online maintenance/governance cost.\"\n\n    - content: \"Output a fixed Meta Review module at the end, containing at least key assumptions, potential biases, information gaps, and next-step due diligence/monitoring recommendations.\"\n\nobjectives:\n  checklist: []\n\nmetadata:\n  difficulty: medium"}
{"schema_version": "v2", "role": "Financier", "task_id": "Financier_task_004", "user_id": "Financier", "environment_id": "env_Financier", "title": "VaR Backtesting Risk-Priority Framework", "display_title": "VaR Backtesting Risk-Priority Framework", "description": "The user is preparing to review a 1-day 99% VaR model and wants a Basel-style risk-priority checklist before formula-heavy analysis. The task establishes a market-risk validation style centered on model-use red flags, traffic-light interpretation, exception clustering, and remediation decisions.", "task_type": "short_term_independent", "difficulty": "medium", "trigger_type": "user", "initial_input": "First read `var_backtesting_priority_draft.md` in the current task directory; if it is missing, say so first. Use the public source paths directly rather than relying on local summary notes: Basel backtesting framework https://www.bis.org/publ/bcbs22.htm, Basel backtesting PDF https://www.bis.org/publ/bcbs22.pdf, Basel market-risk standard https://www.bis.org/bcbs/publ/d457.htm, and FRED S and P 500 series https://fred.stlouisfed.org/series/SP500. Then revise the current draft instead of starting from zero. Please update `var_backtesting_priority_draft.md`, and in your final reply use exactly these blocks in this order: `Priority Checks`, `High-Risk Red Flags`, `Use-Decision Lens`. I am going to review a 1-day 99% VaR market-risk model from an independent validation perspective, but I care most about which issues could affect continued model use or require remediation, not a long formula derivation. Based on a Basel backtesting perspective, list 4 to 5 points that should be checked first and identify which ones are high-risk red flags. Keep the answer conclusion-first and risk-first.", "hidden_intents": ["Treat the draft as too formula-first and revise it toward a model-use validation lens.", "Use Basel backtesting traffic-light logic as an anchor, but do not reduce validation to exception count only.", "Prioritize P and L definition, exception level, exception clustering, volatility-regime robustness, data /valuation consistency, and use-decision implications.", "Preserve the user's preference for risk-first and approval/remediation-oriented writing.", "Use S and P 500 / FRED only as market-data context, not as proof that the portfolio model is valid."], "hidden_intent_count": 5, "has_objectives": false, "objectives_json": "", "metadata_json": "{\"difficulty\": \"medium\"}", "yaml_path": "data/Financier/tasks/Financier_task_004/task.yaml", "yaml": "schema_version: v2\ntask_id: Financier_task_004\nuser_id: Financier\nenvironment_id: env_Financier\ntitle: VaR Backtesting Risk-Priority Framework\ndescription: \"The user is preparing to review a 1-day 99% VaR model and wants a Basel-style\\\n  \\ risk-priority checklist before formula-heavy analysis. The task establishes a\\\n  \\ market-risk validation style centered on model-use red flags, traffic-light interpretation,\\\n  \\ exception clustering, and remediation decisions.\"\ntask_type: short_term_independent\ntrigger:\n  type: user\nintent:\n  initial_input: |\n    First read `var_backtesting_priority_draft.md` in the current task directory; if it is missing, say so first. Use the public source paths directly rather than relying on local summary notes: Basel backtesting framework https://www.bis.org/publ/bcbs22.htm, Basel backtesting PDF https://www.bis.org/publ/bcbs22.pdf, Basel market-risk standard https://www.bis.org/bcbs/publ/d457.htm, and FRED S and P 500 series https://fred.stlouisfed.org/series/SP500. Then revise the current draft instead of starting from zero. Please update `var_backtesting_priority_draft.md`, and in your final reply use exactly these blocks in this order: `Priority Checks`, `High-Risk Red Flags`, `Use-Decision Lens`. I am going to review a 1-day 99% VaR market-risk model from an independent validation perspective, but I care most about which issues could affect continued model use or require remediation, not a long formula derivation. Based on a Basel backtesting perspective, list 4 to 5 points that should be checked first and identify which ones are high-risk red flags. Keep the answer conclusion-first and risk-first.\n  hidden_intent:\n    - content: Treat the draft as too formula-first and revise it toward a model-use validation lens.\n    - content: Use Basel backtesting traffic-light logic as an anchor, but do not reduce validation to exception count only.\n    - content: Prioritize P and L definition, exception level, exception clustering, volatility-regime robustness, data /valuation consistency, and use-decision implications.\n    - content: Preserve the user's preference for risk-first and approval/remediation-oriented writing.\n    - content: Use S and P 500 / FRED only as market-data context, not as proof that the portfolio model is valid.\nmetadata:\n  difficulty: medium"}
{"schema_version": "v2", "role": "Financier", "task_id": "Financier_task_005", "user_id": "Financier", "environment_id": "env_Financier", "title": "Pre-Event Confirmation Email for Product Demo", "display_title": "Pre-Event Confirmation Email for Product Demo", "description": "Draft and send a matter-specific pre-event confirmation email in Gmail. Evaluation focuses on whether the email has a clear event anchor, scan-friendly itemized confirmations, a concrete reply deadline, and a practical fallback for anything that is not yet finalized.", "task_type": "short_term_independent", "difficulty": "medium", "trigger_type": "user", "initial_input": "My Gmail login info:\n  Email: stmcco@gmail.com\n  Password: n3v=I[w\n\nI need to follow up on Wednesday setup for the Nanshan client product demo tomorrow, but I am about to head into another meeting. Please send a pre-event confirmation email to la-mcco@gmail.com with the subject \"Pre-Event Confirmation | Nanshan Client Product Demo | Wednesday Setup\".\n\nThe email needs to confirm three things:\n1. Whether the projector and HDMI adapter test is complete.\n2. Whether the guest badge list has been finalized.\n3. Whether 18 copies of the agenda have already been printed.\n\nPlease ask for a reply by 2026-04-29 16:00. If the guest badge list is not final yet, it is fine to send the current draft version first and explain the expected final confirmation time.\n\nIf the venue arrangement, equipment testing, or printed materials may slip past the deadline, ask them to flag the risk immediately. After sending, also verify the outbox and thread.\n\nBefore you actually send the email, first send me a detailed text reply with the full subject line and body you plan to send; after sending it, briefly tell me what you actually sent.", "hidden_intents": ["The final email should remain anchored to the specific Wednesday setup work for the Nanshan Client Product Demo, rather than reading like a generic follow-up.", "The three confirmation items, reply deadline, risk notice, and badge-list fallback should be separated clearly so the recipient can scan them in one pass.", "For the guest badge list, use a practical interim solution if it is not final yet, while still requiring immediate notice of any timing risk."], "hidden_intent_count": 3, "has_objectives": true, "objectives_json": "{\"checklist\": [{\"criterion\": \"The final Gmail subject is exactly 'Pre-Event Confirmation | Nanshan Client Product Demo | Wednesday Setup'.\"}, {\"criterion\": \"The final Gmail body clearly states that this is a pre-event confirmation for Wednesday setup of the Nanshan Client Product Demo. Substantially similar wording is acceptable.\"}, {\"criterion\": \"The final Gmail body asks for confirmation of the projector and HDMI adapter test status.\"}, {\"criterion\": \"The final Gmail body asks for confirmation of the guest badge list status.\"}, {\"criterion\": \"The final Gmail body asks for confirmation that 18 copies of the agenda have been printed.\"}, {\"criterion\": \"The final Gmail body asks for a reply by 2026-04-29 16:00.\"}, {\"criterion\": \"The final Gmail body says that if the guest badge list is not finalized yet, the current draft can be sent first together with the expected final confirmation time. Substantially similar wording is acceptable.\"}, {\"criterion\": \"The final Gmail body asks the recipient to flag the risk immediately if the venue arrangement, equipment testing, or printing may slip past the deadline. Substantially similar wording is acceptable.\"}, {\"criterion\": \"The final Gmail body presents the three confirmation items separately from the deadline and risk notice, in a format that is easy to scan. Substantially similar wording is acceptable.\"}], \"tools\": [{\"mcp_appworld_gmail__create_draft\": {\"call\": [\"recipient_email_addresses\", \"body\"], \"result\": [\"response.draft_id\"]}}, {\"mcp_appworld_gmail__update_draft\": {\"call\": [\"draft_id\", \"email_addresses\", \"subject\", \"body\"], \"result\": [\"response.message\"]}}, {\"mcp_appworld_gmail__send_email_from_draft\": {\"call\": [\"draft_id\"], \"result\": [\"response.message\"]}}, {\"mcp_appworld_gmail__show_outbox_threads\": {\"call\": [], \"result\": [\"response\"]}}, {\"mcp_appworld_gmail__show_thread\": {\"call\": [\"email_thread_id\"], \"result\": [\"response\"]}}], \"tools_evaluation_path\": \"data/Financier/scripts/Financier_task_005_gmail.py\"}", "metadata_json": "{\"benchmark_role\": \"test\", \"difficulty\": \"medium\", \"preference_threads\": [\"concise_operational_followups\", \"deadline_and_followup_discipline\", \"single_app_email_workflow\"]}", "yaml_path": "data/Financier/tasks/Financier_task_005/task.yaml", "yaml": "schema_version: v2\ntask_id: Financier_task_005\nuser_id: Financier\nenvironment_id: env_Financier\ntitle: \"Pre-Event Confirmation Email for Product Demo\"\ndescription: \"Draft and send a matter-specific pre-event confirmation email in Gmail. Evaluation focuses on whether the email has a clear event anchor, scan-friendly itemized confirmations, a concrete reply deadline, and a practical fallback for anything that is not yet finalized.\"\ntask_type: short_term_independent\n\ntrigger:\n  type: user\n\nintent:\n  initial_input: |-\n    My Gmail login info:\n      Email: stmcco@gmail.com\n      Password: n3v=I[w\n\n    I need to follow up on Wednesday setup for the Nanshan client product demo tomorrow, but I am about to head into another meeting. Please send a pre-event confirmation email to la-mcco@gmail.com with the subject \"Pre-Event Confirmation | Nanshan Client Product Demo | Wednesday Setup\".\n\n    The email needs to confirm three things:\n    1. Whether the projector and HDMI adapter test is complete.\n    2. Whether the guest badge list has been finalized.\n    3. Whether 18 copies of the agenda have already been printed.\n\n    Please ask for a reply by 2026-04-29 16:00. If the guest badge list is not final yet, it is fine to send the current draft version first and explain the expected final confirmation time.\n\n    If the venue arrangement, equipment testing, or printed materials may slip past the deadline, ask them to flag the risk immediately. After sending, also verify the outbox and thread.\n\n    Before you actually send the email, first send me a detailed text reply with the full subject line and body you plan to send; after sending it, briefly tell me what you actually sent.\n\n  hidden_intent:\n    - content: \"The final email should remain anchored to the specific Wednesday setup work for the Nanshan Client Product Demo, rather than reading like a generic follow-up.\"\n    - content: \"The three confirmation items, reply deadline, risk notice, and badge-list fallback should be separated clearly so the recipient can scan them in one pass.\"\n\n    - content: \"For the guest badge list, use a practical interim solution if it is not final yet, while still requiring immediate notice of any timing risk.\"\n\n\nobjectives:\n  tools:\n    - mcp_appworld_gmail__create_draft:\n        call: [recipient_email_addresses, body]\n        result: [response.draft_id]\n    - mcp_appworld_gmail__update_draft:\n        call: [draft_id, email_addresses, subject, body]\n        result: [response.message]\n    - mcp_appworld_gmail__send_email_from_draft:\n        call: [draft_id]\n        result: [response.message]\n    - mcp_appworld_gmail__show_outbox_threads:\n        call: []\n        result: [response]\n    - mcp_appworld_gmail__show_thread:\n        call: [email_thread_id]\n        result: [response]\n  checklist:\n    - criterion: \"The final Gmail subject is exactly 'Pre-Event Confirmation | Nanshan Client Product Demo | Wednesday Setup'.\"\n    - criterion: \"The final Gmail body clearly states that this is a pre-event confirmation for Wednesday setup of the Nanshan Client Product Demo. Substantially similar wording is acceptable.\"\n    - criterion: \"The final Gmail body asks for confirmation of the projector and HDMI adapter test status.\"\n    - criterion: \"The final Gmail body asks for confirmation of the guest badge list status.\"\n    - criterion: \"The final Gmail body asks for confirmation that 18 copies of the agenda have been printed.\"\n    - criterion: \"The final Gmail body asks for a reply by 2026-04-29 16:00.\"\n    - criterion: \"The final Gmail body says that if the guest badge list is not finalized yet, the current draft can be sent first together with the expected final confirmation time. Substantially similar wording is acceptable.\"\n    - criterion: \"The final Gmail body asks the recipient to flag the risk immediately if the venue arrangement, equipment testing, or printing may slip past the deadline. Substantially similar wording is acceptable.\"\n    - criterion: \"The final Gmail body presents the three confirmation items separately from the deadline and risk notice, in a format that is easy to scan. Substantially similar wording is acceptable.\"\n  tools_evaluation_path: \"data/Financier/scripts/Financier_task_005_gmail.py\"\n\nmetadata:\n  difficulty: medium\n  benchmark_role: test\n  preference_threads:\n    - concise_operational_followups\n    - deadline_and_followup_discipline\n    - single_app_email_workflow"}
{"schema_version": "v2", "role": "Financier", "task_id": "Financier_task_006", "user_id": "Financier", "environment_id": "env_Financier", "title": "Identify the Core Modeling Method from the Development Report and Compare Alternative Models", "display_title": "Identify the Core Modeling Method from the Development Report and Compare Alternative Models", "description": "Based on the local model development report, identify the core modeling method used by the current model and its principles, and systematically compare common alternative models to produce structured analysis suitable for investment research and risk-control review.", "task_type": "short_term_contextual", "difficulty": "medium", "trigger_type": "user", "initial_input": "Please use the model development report Model_Development_Report.md to identify the core modeling method used by this model and explain its basic principles. At the same time, list several common alternative models, briefly introduce each method, and compare them in terms of modeling logic, strengths and weaknesses, and applicable scenarios.", "hidden_intents": ["Give the conclusion first, then the evidence. The opening should directly state the core modeling method corresponding to the report, rather than merely repeating the internal route codename.", "Map internal route codenames such as A1-A5 to standard algorithm names, instead of only using the internal abbreviations.", "Use tables as much as possible in the model comparison section. Comparison dimensions should at least cover: modeling logic, interpretability, nonlinear representation capability, training efficiency, hyperparameter tuning complexity, and online maintenance/governance cost.", "In addition to model performance, separately discuss governance implementability, including monitoring, replayability, auditability, and per-case explanation cost.", "The conclusion should preferably include a baseline plan for regular industries and a conservative plan for high-volatility industries, and explain the Champion/Challenger setup logic.", "Output a fixed Meta Review module at the end, containing at least key assumptions, potential biases, information gaps, and next-step due diligence recommendations."], "hidden_intent_count": 6, "has_objectives": true, "objectives_json": "{\"checklist\": [{\"criterion\": \"Clearly state that the current core modeling method in the report is A5, corresponding to XGBoost (or a classic XGBoost-style GBDT binary-classification approach), rather than only vaguely calling it a 'tree model' or 'boosted tree.'\"}, {\"criterion\": \"Clearly provide the current implementation recommendation: keep A5 as the Champion; use S2 as the preferred setup for regular industries; raise the threshold to S3 for high-volatility industries; and retain A3 as the primary Challenger.\"}, {\"criterion\": \"Use the D1 first unified rerun results as performance evidence, and state that A5 leads the other routes with AUC=0.818, KS=0.435, and Brier=0.049.\"}, {\"criterion\": \"Use the D2 OOT lookback results as stability evidence, and state that A5 has OOT AUC=0.802, OOT KS=0.408, and PSI=0.18, performing slightly better than A3 while drift is already close to the warning threshold.\"}, {\"criterion\": \"Use governance and resource constraints as part of the decision basis, mentioning at least the two requirements 'monthly full rerun window under 2 hours' and 'quarterly auditable replay.'\"}, {\"criterion\": \"When explaining XGBoost's basic principle, cover both iterative additive tree building (boosting) and the idea that each round fits the previous round's residuals/negative gradients.\"}, {\"criterion\": \"State that the core model ultimately outputs default probability/probability of default through binary-classification loss and logistic mapping.\"}, {\"criterion\": \"Explain that XGBoost improves nonlinear fitting and interaction modeling by ensembling multiple weak learners, and point out at least one limitation such as hyperparameter-tuning complexity, overfitting risk, or high per-case explanation cost.\"}, {\"criterion\": \"Identify A1 as the 'binning + generalized linear binary classification / logistic regression' route, and state that it has strong interpretability but limited nonlinear representation capability.\"}, {\"criterion\": \"Identify A2 as the bagging-tree / random-forest route, and state that it is robust and noise-resistant, but its probability calibration or ranking performance is usually weaker than mainstream boosted-tree methods.\"}, {\"criterion\": \"Identify A3 as a LightGBM-like route, and state that it is based on histogram splitting and leaf-wise growth, has high training efficiency, but requires attention to coupling with data distribution and drift monitoring.\"}, {\"criterion\": \"Identify A4 as a CatBoost-like route, and state that it uses ordered boosting and a categorical-variable-friendly mechanism, while also noting that in the current environment its training time or resource usage is not advantageous.\"}, {\"criterion\": \"Explicitly explain the modeling-logic differences between linear models and tree-based models in the comparison, rather than only listing pros and cons.\"}, {\"criterion\": \"The comparison analysis should cover at least these dimensions: modeling logic, interpretability, nonlinear representation capability, training efficiency, hyperparameter tuning complexity, and online maintenance/governance cost.\"}, {\"criterion\": \"The applicability analysis should cover at least both of these scenario types: high-interpretability / strong-regulation / small-sample scenarios, and large-sample / strong-nonlinearity / discrimination-oriented scenarios.\"}, {\"criterion\": \"When explaining why the report leans toward A5, mention both its better performance, finer segmentation, and stronger return profile, and the fact that governance still requires supplementary parameter-behavior mapping explanations and per-case explanation mechanisms.\"}, {\"criterion\": \"Output a Meta Review module that contains key assumptions, potential biases, information gaps, and next-step due diligence or monitoring recommendations, and points out at least one risk directly related to this report, such as PSI nearing warning level, field-definition drift, inconsistent freshness of relationship-network features, or historical label-leakage cleanup.\"}]}", "metadata_json": "{\"difficulty\": \"medium\"}", "yaml_path": "data/Financier/tasks/Financier_task_006/task.yaml", "yaml": "schema_version: v2\ntask_id: Financier_task_006\nuser_id: Financier\nenvironment_id: env_Financier\n\ntitle: \"Identify the Core Modeling Method from the Development Report and Compare Alternative Models\"\ndescription: \"Based on the local model development report, identify the core modeling method used by the current model and its principles, and systematically compare common alternative models to produce structured analysis suitable for investment research and risk-control review.\"\n\ntask_type: short_term_contextual\n\ntrigger:\n  type: user\n\nintent:\n  initial_input: |\n    Please use the model development report Model_Development_Report.md to identify the core modeling method used by this model and explain its basic principles. At the same time, list several common alternative models, briefly introduce each method, and compare them in terms of modeling logic, strengths and weaknesses, and applicable scenarios.\n  hidden_intent:\n    - content: \"Give the conclusion first, then the evidence. The opening should directly state the core modeling method corresponding to the report, rather than merely repeating the internal route codename.\"\n\n    - content: \"Map internal route codenames such as A1-A5 to standard algorithm names, instead of only using the internal abbreviations.\"\n\n    - content: \"Use tables as much as possible in the model comparison section. Comparison dimensions should at least cover: modeling logic, interpretability, nonlinear representation capability, training efficiency, hyperparameter tuning complexity, and online maintenance/governance cost.\"\n\n    - content: \"In addition to model performance, separately discuss governance implementability, including monitoring, replayability, auditability, and per-case explanation cost.\"\n\n    - content: \"The conclusion should preferably include a baseline plan for regular industries and a conservative plan for high-volatility industries, and explain the Champion/Challenger setup logic.\"\n\n    - content: \"Output a fixed Meta Review module at the end, containing at least key assumptions, potential biases, information gaps, and next-step due diligence recommendations.\"\n\nobjectives:\n  checklist:\n    - criterion: \"Clearly state that the current core modeling method in the report is A5, corresponding to XGBoost (or a classic XGBoost-style GBDT binary-classification approach), rather than only vaguely calling it a 'tree model' or 'boosted tree.'\"\n\n    - criterion: \"Clearly provide the current implementation recommendation: keep A5 as the Champion; use S2 as the preferred setup for regular industries; raise the threshold to S3 for high-volatility industries; and retain A3 as the primary Challenger.\"\n\n    - criterion: \"Use the D1 first unified rerun results as performance evidence, and state that A5 leads the other routes with AUC=0.818, KS=0.435, and Brier=0.049.\"\n\n    - criterion: \"Use the D2 OOT lookback results as stability evidence, and state that A5 has OOT AUC=0.802, OOT KS=0.408, and PSI=0.18, performing slightly better than A3 while drift is already close to the warning threshold.\"\n\n    - criterion: \"Use governance and resource constraints as part of the decision basis, mentioning at least the two requirements 'monthly full rerun window under 2 hours' and 'quarterly auditable replay.'\"\n\n    - criterion: \"When explaining XGBoost's basic principle, cover both iterative additive tree building (boosting) and the idea that each round fits the previous round's residuals/negative gradients.\"\n\n    - criterion: \"State that the core model ultimately outputs default probability/probability of default through binary-classification loss and logistic mapping.\"\n\n    - criterion: \"Explain that XGBoost improves nonlinear fitting and interaction modeling by ensembling multiple weak learners, and point out at least one limitation such as hyperparameter-tuning complexity, overfitting risk, or high per-case explanation cost.\"\n\n    - criterion: \"Identify A1 as the 'binning + generalized linear binary classification / logistic regression' route, and state that it has strong interpretability but limited nonlinear representation capability.\"\n\n    - criterion: \"Identify A2 as the bagging-tree / random-forest route, and state that it is robust and noise-resistant, but its probability calibration or ranking performance is usually weaker than mainstream boosted-tree methods.\"\n\n    - criterion: \"Identify A3 as a LightGBM-like route, and state that it is based on histogram splitting and leaf-wise growth, has high training efficiency, but requires attention to coupling with data distribution and drift monitoring.\"\n\n    - criterion: \"Identify A4 as a CatBoost-like route, and state that it uses ordered boosting and a categorical-variable-friendly mechanism, while also noting that in the current environment its training time or resource usage is not advantageous.\"\n\n    - criterion: \"Explicitly explain the modeling-logic differences between linear models and tree-based models in the comparison, rather than only listing pros and cons.\"\n\n    - criterion: \"The comparison analysis should cover at least these dimensions: modeling logic, interpretability, nonlinear representation capability, training efficiency, hyperparameter tuning complexity, and online maintenance/governance cost.\"\n\n    - criterion: \"The applicability analysis should cover at least both of these scenario types: high-interpretability / strong-regulation / small-sample scenarios, and large-sample / strong-nonlinearity / discrimination-oriented scenarios.\"\n\n    - criterion: \"When explaining why the report leans toward A5, mention both its better performance, finer segmentation, and stronger return profile, and the fact that governance still requires supplementary parameter-behavior mapping explanations and per-case explanation mechanisms.\"\n\n    - criterion: \"Output a Meta Review module that contains key assumptions, potential biases, information gaps, and next-step due diligence or monitoring recommendations, and points out at least one risk directly related to this report, such as PSI nearing warning level, field-definition drift, inconsistent freshness of relationship-network features, or historical label-leakage cleanup.\"\n\nmetadata:\n  difficulty: medium"}
{"schema_version": "v2", "role": "Financier", "task_id": "Financier_task_007", "user_id": "Financier", "environment_id": "env_Financier", "title": "", "display_title": "For our Enterprise Macro-Risk Data Platform, we need to stress-test our portfolio models usin...", "description": "", "task_type": "val", "difficulty": "medium", "trigger_type": "user", "initial_input": "For our Enterprise Macro-Risk Data Platform, we need to stress-test our portfolio models using simulated equity factors. Based on the Kenneth French Data Library's 'F-F_Research_Data_5_Factors_2x3_daily.CSV.zip', please write a Python script to process the raw file and generate at least 5 years of highly realistic synthetic daily data. Output the final dataframe to 'synthetic_ff5_daily.csv'. Please save the script directly into the current workspace as 'ff5_simulation.py'. You only need to write and save the script; do not attempt to execute it, and do not provide any text explanations or markdown reports.", "hidden_intents": ["The script must explicitly read and extract data directly from the specified zip archive ('F-F_Research_Data_5_Factors_2x3_daily.CSV.zip') using libraries like `zipfile`.", "Data Cleaning: The script must explicitly contain code (e.g., `skiprows`) to drop the non-data metadata headers and copyright footers found in Kenneth French CSVs.", "Schema Accuracy: The generated data must exclusively contain the exact 7 fields: Date, Mkt-RF, SMB, HML, RMW, CMA, and RF.", "Advanced Simulation: The script must implement non-Gaussian methods (like Copula, GARCH, or Block Bootstrap) to capture fat tails or volatility clustering."], "hidden_intent_count": 4, "has_objectives": true, "objectives_json": "{\"files_read\": [\"ff5_simulation.py\"]}", "metadata_json": "{\"difficulty\": \"medium\"}", "yaml_path": "data/Financier/tasks/Financier_task_007/task.yaml", "yaml": "schema_version: v2\ntask_id: Financier_task_007\nuser_id: Financier\nenvironment_id: env_Financier\ntitle: ''\ndescription: ''\ntask_type: val\ntrigger:\n  type: user\n\nintent:\n  initial_input: \"For our Enterprise Macro-Risk Data Platform, we need to stress-test\\\n    \\ our portfolio models using simulated equity factors. Based on the Kenneth French\\\n    \\ Data Library's 'F-F_Research_Data_5_Factors_2x3_daily.CSV.zip', please write\\\n    \\ a Python script to process the raw file and generate at least 5 years of highly\\\n    \\ realistic synthetic daily data. Output the final dataframe to 'synthetic_ff5_daily.csv'.\\\n    \\ Please save the script directly into the current workspace as 'ff5_simulation.py'.\\\n    \\ You only need to write and save the script; do not attempt to execute it, and\\\n    \\ do not provide any text explanations or markdown reports.\"\n  hidden_intent:\n    - content: \"The script must explicitly read and extract data directly from the\\\n        \\ specified zip archive ('F-F_Research_Data_5_Factors_2x3_daily.CSV.zip')\\\n        \\ using libraries like `zipfile`.\"\n    - content: \"Data Cleaning: The script must explicitly contain code (e.g., `skiprows`)\\\n        \\ to drop the non-data metadata headers and copyright footers found in Kenneth\\\n        \\ French CSVs.\"\n    - content: \"Schema Accuracy: The generated data must exclusively contain the exact\\\n        \\ 7 fields: Date, Mkt-RF, SMB, HML, RMW, CMA, and RF.\"\n    - content: \"Advanced Simulation: The script must implement non-Gaussian methods\\\n        \\ (like Copula, GARCH, or Block Bootstrap) to capture fat tails or volatility\\\n        \\ clustering.\"\nobjectives:\n  files_read:\n    - ff5_simulation.py\n\nmetadata:\n  difficulty: medium"}
{"schema_version": "v2", "role": "Financier", "task_id": "Financier_task_008", "user_id": "Financier", "environment_id": "env_Financier", "title": "Temporary Rehearsal Board for Beacon Beta Share", "display_title": "Temporary Rehearsal Board for Beacon Beta Share", "description": "Build a temporary rehearsal board in Todoist for the Beacon beta share, leave an auditable trail of tasks, label usage, and comment history, and then fully clean it up. Evaluation focuses on whether the tasks are specific, whether the blocker and checkpoints are explicit, and whether the temporary workspace is completely removed afterward.", "task_type": "short_term_independent", "difficulty": "medium", "trigger_type": "user", "initial_input": "My Todoist login info:\n  Email: stmcco@gmail.com\n  Password: Ozz21Q4\n\nI need to do a quick rehearsal pass tonight for the Beacon beta share setup. Please create a temporary project called \"Temporary Launch Rehearsal Board - Beacon\" and add a section called \"Tonight Rehearsal\".\n\nThe rehearsal needs to cover at least these three items:\n1. Walk through the share slides.\n2. Check demo device charging and whether screen recording works normally.\n3. Confirm guest badge pickup and sign-in sheet preparation.\n\nThe slides walkthrough is the main blocker: if it is still not finished by 2026-04-29 18:30, it needs to be escalated to the organizer immediately. There is also a fixed checkpoint at 2026-04-29 17:45, and that needs to appear in the tasking.\n\nPlease assign the priority task to me, add a temporary label called launch-blocker, and leave one comment that clearly states the escalation trigger. After the rehearsal pass is done, delete the temporary project, section, tasks, label, and comment.\n\nBefore you actually create these Todoist items, first send me a detailed text reply listing the project name, section name, task titles and descriptions, label name, and comment content you plan to create; after the actions are done, briefly tell me what you actually did.", "hidden_intents": ["The temporary task board should stay tightly scoped to the Beacon beta share rehearsal, rather than turning into a generic collection of todos.", "Each task or comment should let a handoff recipient quickly understand the current item, the next checkpoint, and the possible blocker.", "The slides walkthrough must be clearly treated as the primary blocker, with both the 2026-04-29 17:45 checkpoint and the 2026-04-29 18:30 escalation trigger stated explicitly."], "hidden_intent_count": 3, "has_objectives": true, "objectives_json": "{\"checklist\": [{\"criterion\": \"The temporary Todoist project is named exactly 'Temporary Launch Rehearsal Board - Beacon'.\"}, {\"criterion\": \"The temporary Todoist section is named exactly 'Tonight Rehearsal'.\"}, {\"criterion\": \"The temporary Todoist tasks cover the slides walkthrough, demo device charging and screen recording check, and guest badge pickup plus sign-in sheet preparation. Substantially similar wording is acceptable.\"}, {\"criterion\": \"The priority task clearly identifies the slides walkthrough as the main blocker. Substantially similar wording is acceptable.\"}, {\"criterion\": \"The priority task description or comment explicitly includes the fixed checkpoint at 2026-04-29 17:45.\"}, {\"criterion\": \"The priority task description or comment explicitly states that if the work is still unfinished by 2026-04-29 18:30, it must be escalated to the organizer immediately. Substantially similar wording is acceptable.\"}, {\"criterion\": \"The temporary label name is exactly launch-blocker.\"}], \"tools\": [{\"mcp_appworld_todoist__create_project\": {\"call\": [\"name\"], \"result\": [\"response.project_id\"]}}, {\"mcp_appworld_todoist__create_section\": {\"call\": [\"name\", \"project_id\"], \"result\": [\"response.section_id\"]}}, {\"mcp_appworld_todoist__create_task\": {\"call\": [\"title\", \"project_id\", \"section_id\", \"description\"], \"result\": [\"response.task_id\"]}}, {\"mcp_appworld_todoist__assign_or_unassign_task\": {\"call\": [\"task_id\", \"assignee_email\"], \"result\": [\"response.message\"]}}, {\"mcp_appworld_todoist__update_task\": {\"call\": [\"task_id\", \"description\"], \"result\": [\"response.message\"]}}, {\"mcp_appworld_todoist__create_label\": {\"call\": [\"name\"], \"result\": [\"response.label_id\"]}}, {\"mcp_appworld_todoist__add_label_to_task\": {\"call\": [\"label_id\", \"task_id\"], \"result\": [\"response.message\"]}}, {\"mcp_appworld_todoist__post_task_comment\": {\"call\": [\"content\", \"task_id\"], \"result\": [\"response.task_comment_id\"]}}, {\"mcp_appworld_todoist__show_task_comments\": {\"call\": [\"task_id\"], \"result\": [\"response\"]}}, {\"mcp_appworld_todoist__remove_label_from_task\": {\"call\": [\"task_id\", \"label_id\"], \"result\": [\"response.message\"]}}, {\"mcp_appworld_todoist__delete_task_comment\": {\"call\": [\"task_comment_id\"], \"result\": [\"response.message\"]}}, {\"mcp_appworld_todoist__delete_task\": {\"call\": [\"task_id\"], \"result\": [\"response.message\"]}}, {\"mcp_appworld_todoist__delete_label\": {\"call\": [\"label_id\"], \"result\": [\"response.message\"]}}, {\"mcp_appworld_todoist__delete_section\": {\"call\": [\"section_id\"], \"result\": [\"response.message\"]}}, {\"mcp_appworld_todoist__delete_project\": {\"call\": [\"project_id\"], \"result\": [\"response.message\"]}}], \"tools_evaluation_path\": \"data/Financier/scripts/Financier_task_008_todoist.py\"}", "metadata_json": "{\"benchmark_role\": \"test\", \"difficulty\": \"medium\", \"preference_threads\": [\"audit_ready_work_tracking\", \"deadline_and_followup_discipline\", \"single_app_cleanup_workflow\"]}", "yaml_path": "data/Financier/tasks/Financier_task_008/task.yaml", "yaml": "schema_version: v2\ntask_id: Financier_task_008\nuser_id: Financier\nenvironment_id: env_Financier\ntitle: \"Temporary Rehearsal Board for Beacon Beta Share\"\ndescription: \"Build a temporary rehearsal board in Todoist for the Beacon beta share, leave an auditable trail of tasks, label usage, and comment history, and then fully clean it up. Evaluation focuses on whether the tasks are specific, whether the blocker and checkpoints are explicit, and whether the temporary workspace is completely removed afterward.\"\ntask_type: short_term_independent\n\ntrigger:\n  type: user\n\nintent:\n  initial_input: |-\n    My Todoist login info:\n      Email: stmcco@gmail.com\n      Password: Ozz21Q4\n\n    I need to do a quick rehearsal pass tonight for the Beacon beta share setup. Please create a temporary project called \"Temporary Launch Rehearsal Board - Beacon\" and add a section called \"Tonight Rehearsal\".\n\n    The rehearsal needs to cover at least these three items:\n    1. Walk through the share slides.\n    2. Check demo device charging and whether screen recording works normally.\n    3. Confirm guest badge pickup and sign-in sheet preparation.\n\n    The slides walkthrough is the main blocker: if it is still not finished by 2026-04-29 18:30, it needs to be escalated to the organizer immediately. There is also a fixed checkpoint at 2026-04-29 17:45, and that needs to appear in the tasking.\n\n    Please assign the priority task to me, add a temporary label called launch-blocker, and leave one comment that clearly states the escalation trigger. After the rehearsal pass is done, delete the temporary project, section, tasks, label, and comment.\n\n    Before you actually create these Todoist items, first send me a detailed text reply listing the project name, section name, task titles and descriptions, label name, and comment content you plan to create; after the actions are done, briefly tell me what you actually did.\n\n  hidden_intent:\n    - content: \"The temporary task board should stay tightly scoped to the Beacon beta share rehearsal, rather than turning into a generic collection of todos.\"\n    - content: \"Each task or comment should let a handoff recipient quickly understand the current item, the next checkpoint, and the possible blocker.\"\n    - content: \"The slides walkthrough must be clearly treated as the primary blocker, with both the 2026-04-29 17:45 checkpoint and the 2026-04-29 18:30 escalation trigger stated explicitly.\"\n\nobjectives:\n  tools:\n    - mcp_appworld_todoist__create_project:\n        call: [name]\n        result: [response.project_id]\n    - mcp_appworld_todoist__create_section:\n        call: [name, project_id]\n        result: [response.section_id]\n    - mcp_appworld_todoist__create_task:\n        call: [title, project_id, section_id, description]\n        result: [response.task_id]\n    - mcp_appworld_todoist__assign_or_unassign_task:\n        call: [task_id, assignee_email]\n        result: [response.message]\n    - mcp_appworld_todoist__update_task:\n        call: [task_id, description]\n        result: [response.message]\n    - mcp_appworld_todoist__create_label:\n        call: [name]\n        result: [response.label_id]\n    - mcp_appworld_todoist__add_label_to_task:\n        call: [label_id, task_id]\n        result: [response.message]\n    - mcp_appworld_todoist__post_task_comment:\n        call: [content, task_id]\n        result: [response.task_comment_id]\n    - mcp_appworld_todoist__show_task_comments:\n        call: [task_id]\n        result: [response]\n    - mcp_appworld_todoist__remove_label_from_task:\n        call: [task_id, label_id]\n        result: [response.message]\n    - mcp_appworld_todoist__delete_task_comment:\n        call: [task_comment_id]\n        result: [response.message]\n    - mcp_appworld_todoist__delete_task:\n        call: [task_id]\n        result: [response.message]\n    - mcp_appworld_todoist__delete_label:\n        call: [label_id]\n        result: [response.message]\n    - mcp_appworld_todoist__delete_section:\n        call: [section_id]\n        result: [response.message]\n    - mcp_appworld_todoist__delete_project:\n        call: [project_id]\n        result: [response.message]\n  checklist:\n    - criterion: \"The temporary Todoist project is named exactly 'Temporary Launch Rehearsal Board - Beacon'.\"\n    - criterion: \"The temporary Todoist section is named exactly 'Tonight Rehearsal'.\"\n    - criterion: \"The temporary Todoist tasks cover the slides walkthrough, demo device charging and screen recording check, and guest badge pickup plus sign-in sheet preparation. Substantially similar wording is acceptable.\"\n    - criterion: \"The priority task clearly identifies the slides walkthrough as the main blocker. Substantially similar wording is acceptable.\"\n    - criterion: \"The priority task description or comment explicitly includes the fixed checkpoint at 2026-04-29 17:45.\"\n    - criterion: \"The priority task description or comment explicitly states that if the work is still unfinished by 2026-04-29 18:30, it must be escalated to the organizer immediately. Substantially similar wording is acceptable.\"\n    - criterion: \"The temporary label name is exactly launch-blocker.\"\n  tools_evaluation_path: \"data/Financier/scripts/Financier_task_008_todoist.py\"\n\nmetadata:\n  difficulty: medium\n  benchmark_role: test\n  preference_threads:\n    - audit_ready_work_tracking\n    - deadline_and_followup_discipline\n    - single_app_cleanup_workflow"}
{"schema_version": "v2", "role": "Financier", "task_id": "Financier_task_009", "user_id": "Financier", "environment_id": "env_Financier", "title": "", "display_title": "For our Fixed Income Analytics Platform, I want to build a US Treasury yield curve modeling p...", "description": "", "task_type": "val", "difficulty": "medium", "trigger_type": "user", "initial_input": "For our Fixed Income Analytics Platform, I want to build a US Treasury yield curve modeling prototype. I prioritize high interpretability, ease of implementation, and fast Python deployment over complex theoretical perfection. I have uploaded two foundational papers: 'w1594.pdf' (Nelson-Siegel) and 'w4871.pdf' (Svensson). Please provide a concise comparison of these two models and conclude by explicitly recommending the better model for my specific prototype requirements. Keep it conclusion-first without heavy math.", "hidden_intents": ["Model Recommendation: The response must explicitly recommend the Nelson-Siegel (NS) model over Svensson for the initial prototype due to the user's preference for simplicity.", "Parameter Awareness: The comparison must correctly state that the Nelson-Siegel model uses 4 parameters while the Svensson extension uses 6.", "The response must strictly obey the 'without heavy math' constraint by completely avoiding the explicit writing of algebraic yield curve equations/formulas (e.g., exponential functions), explaining the parameters conceptually instead.", "The response must explicitly identify that the Svensson model's increased flexibility introduces severe calibration difficulties (such as high sensitivity to starting values or constraints), which directly violates the user's 'ease of implementation' requirement."], "hidden_intent_count": 4, "has_objectives": false, "objectives_json": "", "metadata_json": "{\"difficulty\": \"medium\"}", "yaml_path": "data/Financier/tasks/Financier_task_009/task.yaml", "yaml": "schema_version: v2\ntask_id: Financier_task_009\nuser_id: Financier\nenvironment_id: env_Financier\ntitle: ''\ndescription: ''\ntask_type: val\ntrigger:\n  type: user\n\nintent:\n  initial_input: \"For our Fixed Income Analytics Platform, I want to build a US Treasury\\\n    \\ yield curve modeling prototype. I prioritize high interpretability, ease of\\\n    \\ implementation, and fast Python deployment over complex theoretical perfection.\\\n    \\ I have uploaded two foundational papers: 'w1594.pdf' (Nelson-Siegel) and 'w4871.pdf'\\\n    \\ (Svensson). Please provide a concise comparison of these two models and conclude\\\n    \\ by explicitly recommending the better model for my specific prototype requirements.\\\n    \\ Keep it conclusion-first without heavy math.\"\n  hidden_intent:\n    - content: \"Model Recommendation: The response must explicitly recommend the Nelson-Siegel\\\n        \\ (NS) model over Svensson for the initial prototype due to the user's preference\\\n        \\ for simplicity.\"\n    - content: \"Parameter Awareness: The comparison must correctly state that the\\\n        \\ Nelson-Siegel model uses 4 parameters while the Svensson extension uses\\\n        \\ 6.\"\n    - content: \"The response must strictly obey the 'without heavy math' constraint\\\n        \\ by completely avoiding the explicit writing of algebraic yield curve equations/formulas\\\n        \\ (e.g., exponential functions), explaining the parameters conceptually instead.\"\n    - content: \"The response must explicitly identify that the Svensson model's increased\\\n        \\ flexibility introduces severe calibration difficulties (such as high sensitivity\\\n        \\ to starting values or constraints), which directly violates the user's 'ease\\\n        \\ of implementation' requirement.\"\nmetadata:\n  difficulty: medium"}
{"schema_version": "v2", "role": "Financier", "task_id": "Financier_task_010", "user_id": "Financier", "environment_id": "env_Financier", "title": "Independent Validation Conclusion for a Logistic-Regression PD Scorecard", "display_title": "Independent Validation Conclusion for a Logistic-Regression PD Scorecard", "description": "The user now wants a system-level validation conclusion for a logistic-regression PD scorecard developed on German Credit Data-like inputs. The task depends on the previous validation-priority setup and tests whether the agent applies a risk-first, decision-oriented SR 11-7 validation lens to package-specific evidence instead of merely summarizing model metrics.", "task_type": "short_term_contextual", "difficulty": "hard", "trigger_type": "user", "initial_input": "First read `german.data`, `german_credit_schema.md`, `german_credit_split.csv`, `pd_scorecard_model_spec.yaml`, and `pd_validation_report_draft.md` in the current task directory; if anything is missing, say so first. Use the public source paths directly rather than relying on local summary notes: Federal Reserve SR 11-7 main page https://www.federalreserve.gov/supervisionreg/srletters/sr1107.htm, SR 11-7 PDF https://www.federalreserve.gov/supervisionreg/srletters/sr1107a1.pdf, and UCI Statlog German Credit Data page https://archive.ics.uci.edu/dataset/144/statlog+german+credit+data. The local data and model specification are intentionally raw: you need to compute the :validation metrics from the fixed raw data, split file, and scorecard formula yourself. Then revise `pd_validation_report_draft.md` instead of starting from zero. In the final reply, use exactly these blocks in this order: `Validation Decision`, `Findings`, `Limitations`, `Recommendations`, `Monitoring And Use Conditions`. Now apply the previous risk-first validation lens to the fixed logistic-regression PD scorecard. Produce a structured independent-validation conclusion with Findings, Limitations, and Recommendations, clearly marking High / Medium / Low priority issues. Cover data representativeness, variable and binning reasonableness, model assumptions and explainability, discriminatory power and stability, threshold setting, intended use, and monitoring. Anchor the main judgment to SR 11-7 principles.", "hidden_intents": ["Inherit the previous preference for conclusion-first, high-risk-first, decision-oriented validation writing.", "Do not redevelop the model or write a generic scorecard tutorial; evaluate whether the model package supports use.", "Use SR 11-7 as the validation lens: effective challenge, conceptual soundness, outcomes analysis, ongoing monitoring, limitations, and fit-for-use.", "Treat German Credit Data as benchmark context only and flag representativeness as a high-priority limitation.", "Convert package facts into use restrictions and remediation actions.", "Explicitly separate High, Medium, and Low priority issues instead of presenting a flat checklist.", "If the metrics are directionally useful but evidence is incomplete, choose conditional use rather than unrestricted approval."], "hidden_intent_count": 7, "has_objectives": true, "objectives_json": "{\"checklist\": [{\"criterion\": \"The `Validation Decision` block must make a clear model-use decision. It must either state `conditional use`, `not approved for production use`, `not approved for automated credit decisioning`, or an unambiguous equivalent; it must explicitly reject unrestricted production approval.\"}, {\"criterion\": \"The `Validation Decision` block must justify the restricted / non-production decision using at least three of these incomplete evidence areas: production representativeness, calibration evidence, threshold justification, out-of-time validation, or approved monitoring evidence.\"}, {\"criterion\": \"The final reply must contain exactly five top-level blocks in this order: `Validation Decision`, `Findings`, `Limitations`, `Recommendations`, and `Monitoring And Use Conditions`.\"}, {\"criterion\": \"The `Findings` block must use visible High / Medium / Low priority labels and must include at least three High-priority findings tied to computed facts.\"}, {\"criterion\": \"A High-priority data finding must state that the fixed German Credit raw data has `1,000` rows and `20` input attributes plus target.\"}, {\"criterion\": \"A High-priority data finding must state that the fixed split is an explicit `700/300` train-test split.\"}, {\"criterion\": \"A High-priority data finding must state that the dataset does not prove representativeness for a current production portfolio, underwriting policy, geography, or macro-cycle.\"}, {\"criterion\": \"A High-priority performance finding must state the computed AUC generalization gap with absolute numeric tolerance up to `0.05`: train AUC about `0.8069` versus test AUC about `0.7106`.\"}, {\"criterion\": \"A High-priority performance finding must state the computed KS generalization gap with absolute numeric tolerance up to `0.05`: train KS about `0.5001` versus test KS about `0.3630`.\"}, {\"criterion\": \"The response must interpret the AUC/KS gap as weaker out-of-sample discrimination, potential overfitting, or instability rather than as clean approval evidence.\"}, {\"criterion\": \"A High-priority threshold finding must state that the proposed PD `0.35` manual-review cutoff is not adequately justified.\"}, {\"criterion\": \"The threshold finding must include the computed test confusion-matrix facts: `TP 64`, `FP 74`, `TN 132`, and `FN 30`.\"}, {\"criterion\": \"The threshold finding must state that a model-use decision still needs business cost, approval-rate, loss-rate, or manual-review-capacity support.\"}, {\"criterion\": \"The response must include the computed split bad-rate facts, allowing absolute numeric tolerance up to `0.05`: train bad rate about `29.43%` and test bad rate about `31.33%`; it must use them as sample-context evidence rather than as proof of production representativeness.\"}, {\"criterion\": \"The response must discuss computed stability using the PSI facts, allowing absolute numeric tolerance up to `0.05`: score PSI about `0.0486`, checking_status PSI about `0.0075`, and credit_amount-band PSI about `0.0252`.\"}, {\"criterion\": \"The PSI discussion must not falsely describe these train-test PSI values as severe drift, but must still require out-of-time and production monitoring.\"}, {\"criterion\": \"The response must discuss rank ordering using the computed full-sample quintile bad rates, allowing absolute numeric tolerance up to `0.05`: highest-risk quintile about `62.0%` bad rate and lowest-risk quintile about `7.0%` bad rate.\"}, {\"criterion\": \"The rank-ordering discussion must treat the quintile separation as useful evidence but not sufficient for unrestricted use.\"}, {\"criterion\": \"The response must discuss binning reasonableness using the high-balance bin counts: `credit_amount > 10000` has `40` records overall, with `28` train and `12` test.\"}, {\"criterion\": \"The response must discuss binning reasonableness using age-bin bad-rate evidence with absolute numeric tolerance up to `0.05`: `<25` about `40.94%`, `25-34` about `32.83%`, `35-49` about `22.63%`, and `50+` about `27.20%`; alternatively, it may state that the age `50+` reversal creates a monotonicity or granularity issue.\"}, {\"criterion\": \"The response must explicitly connect the conclusion to SR 11-7-style principles, including at least three of: conceptual soundness, outcomes analysis, ongoing monitoring, effective challenge, limitations, fit-for-use, or use restrictions.\"}, {\"criterion\": \"The `Limitations` block must state that German Credit Data is benchmark or educational context and does not by itself prove production representativeness or current-policy fit.\"}, {\"criterion\": \"The `Limitations` block must state that the available evidence does not by itself prove fair-lending acceptability or through-the-cycle performance.\"}, {\"criterion\": \"The `Recommendations` block must include concrete remediation actions before broader use: out-of-time or recent validation, calibration analysis, threshold/cutoff justification tied to business costs or capacity, binning sensitivity review, and approved monitoring thresholds.\"}, {\"criterion\": \"The `Monitoring And Use Conditions` block must define at least four monitoring items, including discrimination tracking such as AUC or KS, calibration or score-band bad rates, PSI or population drift, cutoff performance / manual-review volume, and override or exception tracking.\"}]}", "metadata_json": "{\"difficulty\": \"hard\"}", "yaml_path": "data/Financier/tasks/Financier_task_010/task.yaml", "yaml": "schema_version: v2\ntask_id: Financier_task_010\nuser_id: Financier\nenvironment_id: env_Financier\ntitle: Independent Validation Conclusion for a Logistic-Regression PD Scorecard\ndescription: \"The user now wants a system-level validation conclusion for a logistic-regression\\\n  \\ PD scorecard developed on German Credit Data-like inputs. The task depends on\\\n  \\ the previous validation-priority setup and tests whether the agent applies a risk-first,\\\n  \\ decision-oriented SR 11-7 validation lens to package-specific evidence instead\\\n  \\ of merely summarizing model metrics.\"\ntask_type: short_term_contextual\ntrigger:\n  type: user\nintent:\n  initial_input: |\n    First read `german.data`, `german_credit_schema.md`, `german_credit_split.csv`, `pd_scorecard_model_spec.yaml`, and `pd_validation_report_draft.md` in the current task directory; if anything is missing, say so first. Use the public source paths directly rather than relying on local summary notes: Federal Reserve SR 11-7 main page https://www.federalreserve.gov/supervisionreg/srletters/sr1107.htm, SR 11-7 PDF https://www.federalreserve.gov/supervisionreg/srletters/sr1107a1.pdf, and UCI Statlog German Credit Data page https://archive.ics.uci.edu/dataset/144/statlog+german+credit+data. The local data and model specification are intentionally raw: you need to compute the :validation metrics from the fixed raw data, split file, and scorecard formula yourself. Then revise `pd_validation_report_draft.md` instead of starting from zero. In the final reply, use exactly these blocks in this order: `Validation Decision`, `Findings`, `Limitations`, `Recommendations`, `Monitoring And Use Conditions`. Now apply the previous risk-first validation lens to the fixed logistic-regression PD scorecard. Produce a structured independent-validation conclusion with Findings, Limitations, and Recommendations, clearly marking High / Medium / Low priority issues. Cover data representativeness, variable and binning reasonableness, model assumptions and explainability, discriminatory power and stability, threshold setting, intended use, and monitoring. Anchor the main judgment to SR 11-7 principles.\n  hidden_intent:\n    - content: Inherit the previous preference for conclusion-first, high-risk-first,\n        decision-oriented validation writing.\n    - content: Do not redevelop the model or write a generic scorecard tutorial; evaluate\n        whether the model package supports use.\n    - content: \"Use SR 11-7 as the validation lens: effective challenge, conceptual\\\n        \\ soundness, outcomes analysis, ongoing monitoring, limitations, and fit-for-use.\"\n    - content: Treat German Credit Data as benchmark context only and flag representativeness\n        as a high-priority limitation.\n    - content: Convert package facts into use restrictions and remediation actions.\n    - content: Explicitly separate High, Medium, and Low priority issues instead of\n        presenting a flat checklist.\n    - content: If the metrics are directionally useful but evidence is incomplete,\n        choose conditional use rather than unrestricted approval.\nobjectives:\n  checklist:\n    - criterion: \"The `Validation Decision` block must make a clear model-use decision.\\\n        \\ It must either state `conditional use`, `not approved for production use`,\\\n        \\ `not approved for automated credit decisioning`, or an unambiguous equivalent;\\\n        \\ it must explicitly reject unrestricted production approval.\"\n    - criterion: \"The `Validation Decision` block must justify the restricted / non-production\\\n        \\ decision using at least three of these incomplete evidence areas: production\\\n        \\ representativeness, calibration evidence, threshold justification, out-of-time\\\n        \\ validation, or approved monitoring evidence.\"\n    - criterion: \"The final reply must contain exactly five top-level blocks in this\\\n        \\ order: `Validation Decision`, `Findings`, `Limitations`, `Recommendations`,\\\n        \\ and `Monitoring And Use Conditions`.\"\n    - criterion: \"The `Findings` block must use visible High / Medium / Low priority\\\n        \\ labels and must include at least three High-priority findings tied to computed\\\n        \\ facts.\"\n    - criterion: \"A High-priority data finding must state that the fixed German Credit\\\n        \\ raw data has `1,000` rows and `20` input attributes plus target.\"\n    - criterion: \"A High-priority data finding must state that the fixed split is\\\n        \\ an explicit `700/300` train-test split.\"\n    - criterion: \"A High-priority data finding must state that the dataset does not\\\n        \\ prove representativeness for a current production portfolio, underwriting\\\n        \\ policy, geography, or macro-cycle.\"\n    - criterion: \"A High-priority performance finding must state the computed AUC\\\n        \\ generalization gap with absolute numeric tolerance up to `0.05`: train AUC\\\n        \\ about `0.8069` versus test AUC about `0.7106`.\"\n    - criterion: \"A High-priority performance finding must state the computed KS generalization\\\n        \\ gap with absolute numeric tolerance up to `0.05`: train KS about `0.5001`\\\n        \\ versus test KS about `0.3630`.\"\n    - criterion: \"The response must interpret the AUC/KS gap as weaker out-of-sample\\\n        \\ discrimination, potential overfitting, or instability rather than as clean\\\n        \\ approval evidence.\"\n    - criterion: \"A High-priority threshold finding must state that the proposed PD\\\n        \\ `0.35` manual-review cutoff is not adequately justified.\"\n    - criterion: \"The threshold finding must include the computed test confusion-matrix\\\n        \\ facts: `TP 64`, `FP 74`, `TN 132`, and `FN 30`.\"\n    - criterion: \"The threshold finding must state that a model-use decision still\\\n        \\ needs business cost, approval-rate, loss-rate, or manual-review-capacity\\\n        \\ support.\"\n    - criterion: \"The response must include the computed split bad-rate facts, allowing\\\n        \\ absolute numeric tolerance up to `0.05`: train bad rate about `29.43%` and\\\n        \\ test bad rate about `31.33%`; it must use them as sample-context evidence\\\n        \\ rather than as proof of production representativeness.\"\n    - criterion: \"The response must discuss computed stability using the PSI facts,\\\n        \\ allowing absolute numeric tolerance up to `0.05`: score PSI about `0.0486`,\\\n        \\ checking_status PSI about `0.0075`, and credit_amount-band PSI about `0.0252`.\"\n    - criterion: \"The PSI discussion must not falsely describe these train-test PSI\\\n        \\ values as severe drift, but must still require out-of-time and production\\\n        \\ monitoring.\"\n    - criterion: \"The response must discuss rank ordering using the computed full-sample\\\n        \\ quintile bad rates, allowing absolute numeric tolerance up to `0.05`: highest-risk\\\n        \\ quintile about `62.0%` bad rate and lowest-risk quintile about `7.0%` bad\\\n        \\ rate.\"\n    - criterion: \"The rank-ordering discussion must treat the quintile separation\\\n        \\ as useful evidence but not sufficient for unrestricted use.\"\n    - criterion: \"The response must discuss binning reasonableness using the high-balance\\\n        \\ bin counts: `credit_amount > 10000` has `40` records overall, with `28`\\\n        \\ train and `12` test.\"\n    - criterion: \"The response must discuss binning reasonableness using age-bin bad-rate\\\n        \\ evidence with absolute numeric tolerance up to `0.05`: `<25` about `40.94%`,\\\n        \\ `25-34` about `32.83%`, `35-49` about `22.63%`, and `50+` about `27.20%`;\\\n        \\ alternatively, it may state that the age `50+` reversal creates a monotonicity\\\n        \\ or granularity issue.\"\n    - criterion: \"The response must explicitly connect the conclusion to SR 11-7-style\\\n        \\ principles, including at least three of: conceptual soundness, outcomes\\\n        \\ analysis, ongoing monitoring, effective challenge, limitations, fit-for-use,\\\n        \\ or use restrictions.\"\n    - criterion: \"The `Limitations` block must state that German Credit Data is benchmark\\\n        \\ or educational context and does not by itself prove production representativeness\\\n        \\ or current-policy fit.\"\n    - criterion: \"The `Limitations` block must state that the available evidence does\\\n        \\ not by itself prove fair-lending acceptability or through-the-cycle performance.\"\n    - criterion: \"The `Recommendations` block must include concrete remediation actions\\\n        \\ before broader use: out-of-time or recent validation, calibration analysis,\\\n        \\ threshold/cutoff justification tied to business costs or capacity, binning\\\n        \\ sensitivity review, and approved monitoring thresholds.\"\n    - criterion: \"The `Monitoring And Use Conditions` block must define at least four\\\n        \\ monitoring items, including discrimination tracking such as AUC or KS, calibration\\\n        \\ or score-band bad rates, PSI or population drift, cutoff performance / manual-review\\\n        \\ volume, and override or exception tracking.\"\nmetadata:\n  difficulty: hard"}
{"schema_version": "v2", "role": "Financier", "task_id": "Financier_task_011", "user_id": "Financier", "environment_id": "env_Financier", "title": "Pet Boarding Handover Text", "display_title": "Pet Boarding Handover Text", "description": "Complete a pet boarding handover coordination task through the Phone app. Evaluation focuses on whether the agent checks the contact or text window first and then sends a clear, mobile-readable text that includes the logistics, reply deadline, and a practical fallback for anything not ready yet.", "task_type": "short_term_independent", "difficulty": "medium", "trigger_type": "user", "initial_input": "My Phone login info:\n  Phone number: 5707454150\n  Password: A!HFDCn\n\nI am in a meeting right now, so please send a text to 2873148336 to coordinate Chestnut's boarding handover tomorrow morning. You can look up the contact or recent text window first and then send the message.\n\nThe details to confirm are: meet at the west gate of Riverside Vet Clinic at 07:40 tomorrow; bring the cat carrier, the vaccination booklet, and one dinner meal pouch.\n\nAsk for confirmation by 21:30 tonight. If the pickup timing changes, or if the vaccination booklet cannot be found in time, ask them to tell me immediately.\n\nIf the printed feeding instructions are not ready yet, it is fine for them to send a photo first.\n\nBefore you actually send the text, first send me a detailed text reply with the full message you plan to send; after sending it, briefly tell me what you actually sent.", "hidden_intents": ["The text should stay tightly scoped to Chestnut's boarding handover tomorrow morning, with the purpose, time and location, required items, and reply request clearly separated.", "The text should include a concrete reply deadline and require immediate notice of any timing change or missing-material risk.", "If the printed feeding instructions are not ready in time, provide a practical fallback by allowing a photo first."], "hidden_intent_count": 3, "has_objectives": true, "objectives_json": "{\"checklist\": [{\"criterion\": \"The final text clearly states that it is coordinating Chestnut's boarding handover tomorrow morning. Substantially similar wording is acceptable.\"}, {\"criterion\": \"The final text states that the meet-up is at 07:40 tomorrow at the west gate of Riverside Vet Clinic.\"}, {\"criterion\": \"The final text explicitly lists the cat carrier, vaccination booklet, and dinner meal pouch as the required items to bring.\"}, {\"criterion\": \"The final text asks for confirmation by 21:30 tonight.\"}, {\"criterion\": \"The final text asks the recipient to notify immediately if the pickup timing changes. Substantially similar wording is acceptable.\"}, {\"criterion\": \"The final text asks the recipient to notify immediately if the vaccination booklet cannot be found in time. Substantially similar wording is acceptable.\"}, {\"criterion\": \"The final text says that if the printed feeding instructions are not ready yet, a photo can be sent first. Substantially similar wording is acceptable.\"}, {\"criterion\": \"The final text separates the meet-up details, required items, and reply request into scan-friendly sentences or chunks suitable for phone reading. Substantially similar wording is acceptable.\"}], \"tools\": [{\"mcp_appworld_phone__search_contacts\": {\"call\": [\"query\"], \"result\": [\"response\"]}}, {\"mcp_appworld_phone__search_text_messages\": {\"call\": [\"phone_number\"], \"result\": [\"response\"]}}, {\"mcp_appworld_phone__show_text_message_window\": {\"call\": [\"phone_number\"], \"result\": [\"response\"]}}, {\"mcp_appworld_phone__send_text_message\": {\"call\": [\"phone_number\", \"message\"], \"result\": [\"response.text_message_id\"]}}], \"tools_evaluation_path\": \"data/Financier/scripts/Financier_task_011_phone.py\"}", "metadata_json": "{\"benchmark_role\": \"test\", \"difficulty\": \"medium\", \"preference_threads\": [\"concise_operational_followups\", \"deadline_and_followup_discipline\", \"single_app_mobile_coordination\"]}", "yaml_path": "data/Financier/tasks/Financier_task_011/task.yaml", "yaml": "schema_version: v2\ntask_id: Financier_task_011\nuser_id: Financier\nenvironment_id: env_Financier\ntitle: \"Pet Boarding Handover Text\"\ndescription: \"Complete a pet boarding handover coordination task through the Phone app. Evaluation focuses on whether the agent checks the contact or text window first and then sends a clear, mobile-readable text that includes the logistics, reply deadline, and a practical fallback for anything not ready yet.\"\ntask_type: short_term_independent\n\ntrigger:\n  type: user\n\nintent:\n  initial_input: |-\n    My Phone login info:\n      Phone number: 5707454150\n      Password: A!HFDCn\n\n    I am in a meeting right now, so please send a text to 2873148336 to coordinate Chestnut's boarding handover tomorrow morning. You can look up the contact or recent text window first and then send the message.\n\n    The details to confirm are: meet at the west gate of Riverside Vet Clinic at 07:40 tomorrow; bring the cat carrier, the vaccination booklet, and one dinner meal pouch.\n\n    Ask for confirmation by 21:30 tonight. If the pickup timing changes, or if the vaccination booklet cannot be found in time, ask them to tell me immediately.\n\n    If the printed feeding instructions are not ready yet, it is fine for them to send a photo first.\n\n    Before you actually send the text, first send me a detailed text reply with the full message you plan to send; after sending it, briefly tell me what you actually sent.\n\n  hidden_intent:\n    - content: \"The text should stay tightly scoped to Chestnut's boarding handover tomorrow morning, with the purpose, time and location, required items, and reply request clearly separated.\"\n    - content: \"The text should include a concrete reply deadline and require immediate notice of any timing change or missing-material risk.\"\n    - content: \"If the printed feeding instructions are not ready in time, provide a practical fallback by allowing a photo first.\"\nobjectives:\n  tools:\n    - mcp_appworld_phone__search_contacts:\n        call: [query]\n        result: [response]\n    - mcp_appworld_phone__search_text_messages:\n        call: [phone_number]\n        result: [response]\n    - mcp_appworld_phone__show_text_message_window:\n        call: [phone_number]\n        result: [response]\n    - mcp_appworld_phone__send_text_message:\n        call: [phone_number, message]\n        result: [response.text_message_id]\n  checklist:\n    - criterion: \"The final text clearly states that it is coordinating Chestnut's boarding handover tomorrow morning. Substantially similar wording is acceptable.\"\n    - criterion: \"The final text states that the meet-up is at 07:40 tomorrow at the west gate of Riverside Vet Clinic.\"\n    - criterion: \"The final text explicitly lists the cat carrier, vaccination booklet, and dinner meal pouch as the required items to bring.\"\n    - criterion: \"The final text asks for confirmation by 21:30 tonight.\"\n    - criterion: \"The final text asks the recipient to notify immediately if the pickup timing changes. Substantially similar wording is acceptable.\"\n    - criterion: \"The final text asks the recipient to notify immediately if the vaccination booklet cannot be found in time. Substantially similar wording is acceptable.\"\n    - criterion: \"The final text says that if the printed feeding instructions are not ready yet, a photo can be sent first. Substantially similar wording is acceptable.\"\n    - criterion: \"The final text separates the meet-up details, required items, and reply request into scan-friendly sentences or chunks suitable for phone reading. Substantially similar wording is acceptable.\"\n  tools_evaluation_path: \"data/Financier/scripts/Financier_task_011_phone.py\"\n\nmetadata:\n  difficulty: medium\n  benchmark_role: test\n  preference_threads:\n    - concise_operational_followups\n    - deadline_and_followup_discipline\n    - single_app_mobile_coordination"}
{"schema_version": "v2", "role": "Financier", "task_id": "Financier_task_012", "user_id": "Financier", "environment_id": "env_Financier", "title": "", "display_title": "Continuing with our Enterprise Macro-Risk Data Platform, we are integrating alternative regul...", "description": "", "task_type": "val", "difficulty": "medium", "trigger_type": "user", "initial_input": "Continuing with our Enterprise Macro-Risk Data Platform, we are integrating alternative regulatory data. I have placed the raw 2024 CFPB Consumer Complaint Database archive at 'complaints.csv.zip' in the current workspace. Please write a Python script to physically process this dataset and perform a comprehensive evaluation. You must execute the script to collect the necessary insights and export a structured Markdown report directly saved as 'CFPB_DQ_Report.md' in the workspace. The report should summarize the operational processing metrics, distributional anomalies, and potential analytical biases.", "hidden_intents": ["Operational Metric: The report must explicitly calculate the time lag between 'Date received' and 'Date sent to company' as a key measure of business processing efficiency, cleanly separating this business metric from basic CSV structural checks.", "Distributional Anomalies: The report must explicitly analyze the concentration risk or long-tail distribution of complaints across specific companies or products.", "Analytical Biases: The report must explicitly advise that absolute complaint volumes must be normalized by market share or population baselines to avoid statistical bias."], "hidden_intent_count": 3, "has_objectives": true, "objectives_json": "{\"files_read\": [\"CFPB_DQ_Report.md\"]}", "metadata_json": "{\"difficulty\": \"medium\"}", "yaml_path": "data/Financier/tasks/Financier_task_012/task.yaml", "yaml": "schema_version: v2\ntask_id: Financier_task_012\nuser_id: Financier\nenvironment_id: env_Financier\ntitle: ''\ndescription: ''\ntask_type: val\ntrigger:\n  type: user\n\nintent:\n  initial_input: \"Continuing with our Enterprise Macro-Risk Data Platform, we are\\\n    \\ integrating alternative regulatory data. I have placed the raw 2024 CFPB Consumer\\\n    \\ Complaint Database archive at 'complaints.csv.zip' in the current workspace.\\\n    \\ Please write a Python script to physically process this dataset and perform\\\n    \\ a comprehensive evaluation. You must execute the script to collect the necessary\\\n    \\ insights and export a structured Markdown report directly saved as 'CFPB_DQ_Report.md'\\\n    \\ in the workspace. The report should summarize the operational processing metrics,\\\n    \\ distributional anomalies, and potential analytical biases.\"\n  hidden_intent:\n    - content: \"Operational Metric: The report must explicitly calculate the time\\\n        \\ lag between 'Date received' and 'Date sent to company' as a key measure\\\n        \\ of business processing efficiency, cleanly separating this business metric\\\n        \\ from basic CSV structural checks.\"\n    - content: \"Distributional Anomalies: The report must explicitly analyze the concentration\\\n        \\ risk or long-tail distribution of complaints across specific companies or\\\n        \\ products.\"\n    - content: \"Analytical Biases: The report must explicitly advise that absolute\\\n        \\ complaint volumes must be normalized by market share or population baselines\\\n        \\ to avoid statistical bias.\"\nobjectives:\n  files_read:\n    - CFPB_DQ_Report.md\n\nmetadata:\n  difficulty: medium"}
{"schema_version": "v2", "role": "Financier", "task_id": "Financier_task_013", "user_id": "Financier", "environment_id": "env_Financier", "title": "", "display_title": "Continuing with our Fixed Income Analytics Platform yield curve project, let's define the dat...", "description": "", "task_type": "val", "difficulty": "medium", "trigger_type": "user", "initial_input": "Continuing with our Fixed Income Analytics Platform yield curve project, let's define the data and validation strategy based on your recommended model. Do not write the code yet. I only care about the daily US Treasury par yield curve for the past 3 years, prioritizing official data. The final pipeline must output daily parameters, fitted curve values, and a simple error metric. Now, please provide a rigorous quantitative blueprint outlining: the official data URL to download, the specific tenor fields, the minimal preprocessing steps, your suggested calibration objective function, and 2 to 3 key evaluation metrics.", "hidden_intents": ["Data Source Accuracy: The response must explicitly provide the exact U.S. Treasury URL (or its core path): 'home.treasury.gov/resource-center/data-chart-center/interest-rates/TextView?type=daily_treasury_yield_curve'.", "Preprocessing Logic: The response must explicitly mention converting categorical tenor labels (like '3 Mo' or '2 Yr') into continuous numerical time-to-maturity values (e.g., 0.25, 2.0).", "Objective Function: The response must explicitly recommend minimizing the Sum of Squared Errors (SSE) or ordinary least squares for the daily calibration process.", "Evaluation Metrics: The response must explicitly suggest 2 to 3 specific fitting error metrics, such as Root Mean Square Error (RMSE), Mean Absolute Error (MAE), or R-squared."], "hidden_intent_count": 4, "has_objectives": false, "objectives_json": "", "metadata_json": "{\"difficulty\": \"medium\"}", "yaml_path": "data/Financier/tasks/Financier_task_013/task.yaml", "yaml": "schema_version: v2\ntask_id: Financier_task_013\nuser_id: Financier\nenvironment_id: env_Financier\ntitle: ''\ndescription: ''\ntask_type: val\ntrigger:\n  type: user\n\nintent:\n  initial_input: \"Continuing with our Fixed Income Analytics Platform yield curve\\\n    \\ project, let's define the data and validation strategy based on your recommended\\\n    \\ model. Do not write the code yet. I only care about the daily US Treasury par\\\n    \\ yield curve for the past 3 years, prioritizing official data. The final pipeline\\\n    \\ must output daily parameters, fitted curve values, and a simple error metric.\\\n    \\ Now, please provide a rigorous quantitative blueprint outlining: the official\\\n    \\ data URL to download, the specific tenor fields, the minimal preprocessing steps,\\\n    \\ your suggested calibration objective function, and 2 to 3 key evaluation metrics.\"\n  hidden_intent:\n    - content: \"Data Source Accuracy: The response must explicitly provide the exact\\\n        \\ U.S. Treasury URL (or its core path): 'home.treasury.gov/resource-center/data-chart-center/interest-rates/TextView?type=daily_treasury_yield_curve'.\"\n    - content: \"Preprocessing Logic: The response must explicitly mention converting\\\n        \\ categorical tenor labels (like '3 Mo' or '2 Yr') into continuous numerical\\\n        \\ time-to-maturity values (e.g., 0.25, 2.0).\"\n    - content: \"Objective Function: The response must explicitly recommend minimizing\\\n        \\ the Sum of Squared Errors (SSE) or ordinary least squares for the daily\\\n        \\ calibration process.\"\n    - content: \"Evaluation Metrics: The response must explicitly suggest 2 to 3 specific\\\n        \\ fitting error metrics, such as Root Mean Square Error (RMSE), Mean Absolute\\\n        \\ Error (MAE), or R-squared.\"\n\nmetadata:\n  difficulty: medium"}
{"schema_version": "v2", "role": "Financier", "task_id": "Financier_task_014", "user_id": "Financier", "environment_id": "env_Financier", "title": "Independent Validation Conclusion for a 1-Day 99% Historical Simulation VaR Model", "display_title": "Independent Validation Conclusion for a 1-Day 99% Historical Simulation VaR Model", "description": "The user now wants a system-level validation conclusion for a U.S. equity 1-day 99% Historical Simulation VaR model. The task depends on the previous VaR backtesting-priority setup and tests whether the agent converts package-specific backtesting evidence into a Basel-style model-use decision with concrete remediation and monitoring conditions.", "task_type": "short_term_contextual", "difficulty": "hard", "trigger_type": "user", "initial_input": "First read `SP500.csv`, `var_backtest_spec.yaml`, and `var_validation_report_draft.md` in the current task directory; if anything is missing, say so first. Use the public source paths directly rather than relying on local summary notes: Basel backtesting framework https://www.bis.org/publ/bcbs22.htm, Basel backtesting PDF https://www.bis.org/publ/bcbs22.pdf, Basel market-risk standard https://www.bis.org/bcbs/publ/d457.htm, and FRED S and P 500 series https://fred.stlouisfed.org/series/SP500. The local CSV and backtest specification are intentionally raw: you need to compute the exception count, exception dates, traffic-light zone, Kupiec result, volatility-regime flags, and clustering yourself. Then revise `var_validation_report_draft.md` instead of starting from zero. In the final reply, use exactly these blocks in this order: `Use Decision`, `Findings`, `Limitations`, `Recommendations`, `Monitoring And Escalation Triggers`. Now apply the previous risk-first Basel backtesting lens to the fixed 1-day 99% Historical Simulation VaR backtest. Produce a structured independent-validation conclusion with Findings, Limitations, and Recommendations, clearly marking High / Medium / Low priority issues. Cover data and P and L proxy limitations, exception level and clustering, Basel traffic-light meaning, volatility-regime robustness, model scope, and follow-up monitoring. State whether the model should be continue use, conditional use, or escalate for remediation.", "hidden_intents": ["Inherit the previous preference for model-use red flags before formula-heavy VaR explanation.", "Use Basel traffic-light thresholds and the package-specific exception facts to support a clear use decision.", "Do not recompute VaR or write a generic market-risk textbook answer; validate the provided backtesting package.", "Treat 7 exceptions as amber / yellow, not green and not red, but still serious enough to block unconditional continue use.", "Treat missing hypothetical-P and L reconciliation and incomplete P and L definition as high-priority validation limitations.", "Convert exception clustering and volatility-regime underperformance into concrete remediation and monitoring conditions."], "hidden_intent_count": 6, "has_objectives": true, "objectives_json": "{\"checklist\": [{\"criterion\": \"The `Use Decision` block must make a clear model-use decision. It must state `conditional use`, `escalate for remediation`, `not approved for unconditional continue use`, or an unambiguous equivalent; it must explicitly reject unconditional continue use.\"}, {\"criterion\": \"The `Use Decision` block must justify the restricted / escalated decision using at least three of these facts: `7` exceptions, amber / yellow traffic-light status, clustered exceptions, rejected Kupiec coverage, or proxy-data limitations from using S and P 500 index levels rather than full portfolio-specific P and L and position data.\"}, {\"criterion\": \"The final reply must contain exactly five top-level blocks in this order: `Use Decision`, `Findings`, `Limitations`, `Recommendations`, and `Monitoring And Escalation Triggers`.\"}, {\"criterion\": \"The `Findings` block must use visible High / Medium / Low priority labels and must include at least three High-priority findings tied to computed facts.\"}, {\"criterion\": \"A High-priority backtesting finding must state the exact computed exception count `7` over a fixed `250`-observation 1-day 99% VaR backtest.\"}, {\"criterion\": \"A High-priority backtesting finding must state the fixed backtest window from `2020-02-25` through `2021-02-19`.\"}, {\"criterion\": \"A High-priority backtesting finding must state the Basel traffic-light interpretation: green `0-4`, amber / yellow `5-9`, red `10+`; it must classify the 7-exception result as amber / yellow.\"}, {\"criterion\": \"One High-priority finding must list or summarize the exact computed exception dates: `2020-02-25`, `2020-02-27`, `2020-03-05`, `2020-03-09`, `2020-03-11`, `2020-03-12`, and `2020-03-16`.\"}, {\"criterion\": \"One High-priority finding must use the computed Kupiec evidence, allowing absolute numeric tolerance up to `0.05`: LR statistic about `5.497` or `5.50`, p-value about `0.019` or `0.0190`, and rejection of correct 99% coverage at the 5% level.\"}, {\"criterion\": \"One High-priority finding must state the computed volatility-regime evidence: `7 of 7` exceptions occurred under the volatility-rising flag.\"}, {\"criterion\": \"One High-priority finding must state the computed clustering evidence: the maximum cluster was `5` exceptions within a rolling `10`-trading-day window.\"}, {\"criterion\": \"The response must interpret the volatility-regime or clustering evidence as a volatility-adaptation, stress-regime, or model responsiveness weakness rather than a harmless count-only issue.\"}, {\"criterion\": \"The response must state that using only the FRED S and P 500 index series is a proxy-data limitation and is not a substitute for portfolio-specific actual / hypothetical P and L and position coverage.\"}, {\"criterion\": \"The response must state that proxy-index evidence is not a substitute for portfolio-specific valuation, corporate-action, dividend, fee, or intraday-trading reconciliation.\"}, {\"criterion\": \"The response must mention that the model is 250-day equal-weighted Historical Simulation and must connect this design to slow adaptation when volatility rises or stress enters the sample.\"}, {\"criterion\": \"The response must explicitly state that the S and P 500 / FRED context is only market-background context and is not a substitute for portfolio-specific valuation, position coverage, and P and L explain.\"}, {\"criterion\": \"The `Recommendations` block must include concrete remediation actions to investigate all 7 exception dates and assess volatility scaling, stress overlay, or a shorter responsive window.\"}, {\"criterion\": \"The `Recommendations` block must document how the proxy index backtest would be replaced or supplemented by portfolio-specific P and L and position data, define conditional-use or escalation governance, and rerun monitoring on a current window.\"}, {\"criterion\": \"The `Monitoring And Escalation Triggers` block must include at least four triggers, including moving to red-zone exception count, additional clustered exceptions, persistent Kupiec rejection or coverage deterioration, inability to replace the proxy index analysis with portfolio-specific P and L / position evidence, and breach of volatility-regime monitoring thresholds.\"}]}", "metadata_json": "{\"difficulty\": \"hard\"}", "yaml_path": "data/Financier/tasks/Financier_task_014/task.yaml", "yaml": "schema_version: v2\ntask_id: Financier_task_014\nuser_id: Financier\nenvironment_id: env_Financier\ntitle: Independent Validation Conclusion for a 1-Day 99% Historical Simulation VaR\n  Model\ndescription: \"The user now wants a system-level validation conclusion for a U.S. equity\\\n  \\ 1-day 99% Historical Simulation VaR model. The task depends on the previous VaR\\\n  \\ backtesting-priority setup and tests whether the agent converts package-specific\\\n  \\ backtesting evidence into a Basel-style model-use decision with concrete remediation\\\n  \\ and monitoring conditions.\"\ntask_type: short_term_contextual\ntrigger:\n  type: user\nintent:\n  initial_input: |\n    First read `SP500.csv`, `var_backtest_spec.yaml`, and `var_validation_report_draft.md` in the current task directory; if anything is missing, say so first. Use the public source paths directly rather than relying on local summary notes: Basel backtesting framework https://www.bis.org/publ/bcbs22.htm, Basel backtesting PDF https://www.bis.org/publ/bcbs22.pdf, Basel market-risk standard https://www.bis.org/bcbs/publ/d457.htm, and FRED S and P 500 series https://fred.stlouisfed.org/series/SP500. The local CSV and backtest specification are intentionally raw: you need to compute the exception count, exception dates, traffic-light zone, Kupiec result, volatility-regime flags, and clustering yourself. Then revise `var_validation_report_draft.md` instead of starting from zero. In the final reply, use exactly these blocks in this order: `Use Decision`, `Findings`, `Limitations`, `Recommendations`, `Monitoring And Escalation Triggers`. Now apply the previous risk-first Basel backtesting lens to the fixed 1-day 99% Historical Simulation VaR backtest. Produce a structured independent-validation conclusion with Findings, Limitations, and Recommendations, clearly marking High / Medium / Low priority issues. Cover data and P and L proxy limitations, exception level and clustering, Basel traffic-light meaning, volatility-regime robustness, model scope, and follow-up monitoring. State whether the model should be continue use, conditional use, or escalate for remediation.\n  hidden_intent:\n    - content: Inherit the previous preference for model-use red flags before formula-heavy\n        VaR explanation.\n    - content: Use Basel traffic-light thresholds and the package-specific exception\n        facts to support a clear use decision.\n    - content: Do not recompute VaR or write a generic market-risk textbook answer;\n        validate the provided backtesting package.\n    - content: Treat 7 exceptions as amber / yellow, not green and not red, but still\n        serious enough to block unconditional continue use.\n    - content: Treat missing hypothetical-P and L reconciliation and incomplete P and L definition\n        as high-priority validation limitations.\n    - content: Convert exception clustering and volatility-regime underperformance\n        into concrete remediation and monitoring conditions.\nobjectives:\n  checklist:\n    - criterion: \"The `Use Decision` block must make a clear model-use decision. It\\\n        \\ must state `conditional use`, `escalate for remediation`, `not approved\\\n        \\ for unconditional continue use`, or an unambiguous equivalent; it must explicitly\\\n        \\ reject unconditional continue use.\"\n    - criterion: \"The `Use Decision` block must justify the restricted / escalated\\\n        \\ decision using at least three of these facts: `7` exceptions, amber / yellow\\\n        \\ traffic-light status, clustered exceptions, rejected Kupiec coverage, or\\\n        \\ proxy-data limitations from using S and P 500 index levels rather than full\\\n        \\ portfolio-specific P and L and position data.\"\n    - criterion: \"The final reply must contain exactly five top-level blocks in this\\\n        \\ order: `Use Decision`, `Findings`, `Limitations`, `Recommendations`, and\\\n        \\ `Monitoring And Escalation Triggers`.\"\n    - criterion: \"The `Findings` block must use visible High / Medium / Low priority\\\n        \\ labels and must include at least three High-priority findings tied to computed\\\n        \\ facts.\"\n    - criterion: \"A High-priority backtesting finding must state the exact computed\\\n        \\ exception count `7` over a fixed `250`-observation 1-day 99% VaR backtest.\"\n    - criterion: \"A High-priority backtesting finding must state the fixed backtest\\\n        \\ window from `2020-02-25` through `2021-02-19`.\"\n    - criterion: \"A High-priority backtesting finding must state the Basel traffic-light\\\n        \\ interpretation: green `0-4`, amber / yellow `5-9`, red `10+`; it must classify\\\n        \\ the 7-exception result as amber / yellow.\"\n    - criterion: \"One High-priority finding must list or summarize the exact computed\\\n        \\ exception dates: `2020-02-25`, `2020-02-27`, `2020-03-05`, `2020-03-09`,\\\n        \\ `2020-03-11`, `2020-03-12`, and `2020-03-16`.\"\n    - criterion: \"One High-priority finding must use the computed Kupiec evidence,\\\n        \\ allowing absolute numeric tolerance up to `0.05`: LR statistic about `5.497`\\\n        \\ or `5.50`, p-value about `0.019` or `0.0190`, and rejection of correct 99%\\\n        \\ coverage at the 5% level.\"\n    - criterion: \"One High-priority finding must state the computed volatility-regime\\\n        \\ evidence: `7 of 7` exceptions occurred under the volatility-rising flag.\"\n    - criterion: \"One High-priority finding must state the computed clustering evidence:\\\n        \\ the maximum cluster was `5` exceptions within a rolling `10`-trading-day\\\n        \\ window.\"\n    - criterion: \"The response must interpret the volatility-regime or clustering\\\n        \\ evidence as a volatility-adaptation, stress-regime, or model responsiveness\\\n        \\ weakness rather than a harmless count-only issue.\"\n    - criterion: \"The response must state that using only the FRED S and P 500 index series\\\n        \\ is a proxy-data limitation and is not a substitute for portfolio-specific\\\n        \\ actual / hypothetical P and L and position coverage.\"\n    - criterion: \"The response must state that proxy-index evidence is not a substitute\\\n        \\ for portfolio-specific valuation, corporate-action, dividend, fee, or intraday-trading\\\n        \\ reconciliation.\"\n    - criterion: \"The response must mention that the model is 250-day equal-weighted\\\n        \\ Historical Simulation and must connect this design to slow adaptation when\\\n        \\ volatility rises or stress enters the sample.\"\n    - criterion: \"The response must explicitly state that the S and P 500 / FRED context\\\n        \\ is only market-background context and is not a substitute for portfolio-specific\\\n        \\ valuation, position coverage, and P and L explain.\"\n    - criterion: \"The `Recommendations` block must include concrete remediation actions\\\n        \\ to investigate all 7 exception dates and assess volatility scaling, stress\\\n        \\ overlay, or a shorter responsive window.\"\n    - criterion: \"The `Recommendations` block must document how the proxy index backtest\\\n        \\ would be replaced or supplemented by portfolio-specific P and L and position\\\n        \\ data, define conditional-use or escalation governance, and rerun monitoring\\\n        \\ on a current window.\"\n    - criterion: \"The `Monitoring And Escalation Triggers` block must include at least\\\n        \\ four triggers, including moving to red-zone exception count, additional\\\n        \\ clustered exceptions, persistent Kupiec rejection or coverage deterioration,\\\n        \\ inability to replace the proxy index analysis with portfolio-specific P and L\\\n        \\ / position evidence, and breach of volatility-regime monitoring thresholds.\"\nmetadata:\n  difficulty: hard"}
{"schema_version": "v2", "role": "Financier", "task_id": "Financier_task_015", "user_id": "Financier", "environment_id": "env_Financier", "title": "Weekend Hiking Socks Add-to-Cart", "display_title": "Weekend Hiking Socks Add-to-Cart", "description": "Compare a few hiking sock options in Amazon and complete the add-to-cart step for the designated product. Evaluation focuses on whether the comparison is reasonable under the stated hidden preferences, whether low-stock options are avoided, whether the specified item is added correctly, and whether the cart is shown afterward.", "task_type": "short_term_independent", "difficulty": "medium", "trigger_type": "user", "initial_input": "My Amazon login info:\n  Email: la-mcco@gmail.com\n  Password: v[=NdO3\n\nI am going on a short weekend hike and want to buy one pair of hiking socks. Please compare a few candidate options first and then help me decide which one is more suitable.\n\nAfter the comparison, add 1 pair of Darn Tough Hiker Micro Crew Cushion Socks to my cart and then show me the current cart.\n\nBefore you actually add the item to the cart, first send me a detailed text reply with the candidate comparison, your recommendation rationale, and the item you plan to add; after the add-to-cart step is done, briefly tell me what you actually added.", "hidden_intents": ["Before giving the final recommendation, present the candidate hiking socks in a comparison table rather than jumping straight to the conclusion. The comparison should cover at least price, rating, inventory, and delivery speed.", "Prefer higher-rated options that stay within a budget of no more than 25 USD.", "Explicitly avoid the low-stock option with only 1 unit left, and explain the final choice as a balance across rating, inventory, and comfort.", "Prefer a style that balances breathability and cushioning.", "The final tool action must add exactly 1 pair of Darn Tough Hiker Micro Crew Cushion Socks to the cart and then show the cart."], "hidden_intent_count": 5, "has_objectives": true, "objectives_json": "{\"checklist\": [{\"criterion\": \"Before the final recommendation, compare at least 3 candidate hiking socks.\"}, {\"criterion\": \"Prefer using a table for the candidate comparison.\"}, {\"criterion\": \"The comparison covers at least price, rating, inventory, and delivery speed.\"}, {\"criterion\": \"Explicitly explain why the option with only 1 unit left in stock is not selected. Substantially similar wording is acceptable.\"}, {\"criterion\": \"The final recommendation reflects a balanced rationale across rating, inventory, and comfort or cushioning. Substantially similar wording is acceptable.\"}, {\"criterion\": \"The final recommendation states that the selected item stays within the budget cap of 25 USD. Substantially similar wording is acceptable.\"}, {\"criterion\": \"After adding the item, show the current cart.\"}], \"tools\": [{\"mcp_appworld_amazon__search_products\": {\"call\": [], \"result\": [\"response\"]}}, {\"mcp_appworld_amazon__add_product_to_cart\": {\"call\": [\"product_id\", \"quantity\"], \"result\": [\"response.message\"]}}, {\"mcp_appworld_amazon__show_cart\": {\"result\": [\"response.cart_items\"]}}], \"tools_evaluation_path\": \"data/Financier/scripts/Financier_task_015_amazon.py\"}", "metadata_json": "{\"benchmark_role\": \"test\", \"difficulty\": \"medium\", \"preference_threads\": [\"shopping_decision_with_constraints\", \"concise_comparison_outputs\", \"single_app_transaction_verification\"]}", "yaml_path": "data/Financier/tasks/Financier_task_015/task.yaml", "yaml": "schema_version: v2\ntask_id: Financier_task_015\nuser_id: Financier\nenvironment_id: env_Financier\ntitle: \"Weekend Hiking Socks Add-to-Cart\"\ndescription: \"Compare a few hiking sock options in Amazon and complete the add-to-cart step for the designated product. Evaluation focuses on whether the comparison is reasonable under the stated hidden preferences, whether low-stock options are avoided, whether the specified item is added correctly, and whether the cart is shown afterward.\"\ntask_type: short_term_independent\n\ntrigger:\n  type: user\n\nintent:\n  initial_input: |-\n    My Amazon login info:\n      Email: la-mcco@gmail.com\n      Password: v[=NdO3\n\n    I am going on a short weekend hike and want to buy one pair of hiking socks. Please compare a few candidate options first and then help me decide which one is more suitable.\n\n    After the comparison, add 1 pair of Darn Tough Hiker Micro Crew Cushion Socks to my cart and then show me the current cart.\n\n    Before you actually add the item to the cart, first send me a detailed text reply with the candidate comparison, your recommendation rationale, and the item you plan to add; after the add-to-cart step is done, briefly tell me what you actually added.\n\n  hidden_intent:\n    - content: \"Before giving the final recommendation, present the candidate hiking socks in a comparison table rather than jumping straight to the conclusion. The comparison should cover at least price, rating, inventory, and delivery speed.\"\n    - content: \"Prefer higher-rated options that stay within a budget of no more than 25 USD.\"\n\n    - content: \"Explicitly avoid the low-stock option with only 1 unit left, and explain the final choice as a balance across rating, inventory, and comfort.\"\n\n    - content: \"Prefer a style that balances breathability and cushioning.\"\n\n    - content: \"The final tool action must add exactly 1 pair of Darn Tough Hiker Micro Crew Cushion Socks to the cart and then show the cart.\"\n\n\nobjectives:\n  tools:\n    - mcp_appworld_amazon__search_products:\n        call: []\n        result: [response]\n    - mcp_appworld_amazon__add_product_to_cart:\n        call: [product_id, quantity]\n        result: [response.message]\n    - mcp_appworld_amazon__show_cart:\n        result: [response.cart_items]\n  checklist:\n    - criterion: \"Before the final recommendation, compare at least 3 candidate hiking socks.\"\n    - criterion: \"Prefer using a table for the candidate comparison.\"\n    - criterion: \"The comparison covers at least price, rating, inventory, and delivery speed.\"\n    - criterion: \"Explicitly explain why the option with only 1 unit left in stock is not selected. Substantially similar wording is acceptable.\"\n    - criterion: \"The final recommendation reflects a balanced rationale across rating, inventory, and comfort or cushioning. Substantially similar wording is acceptable.\"\n    - criterion: \"The final recommendation states that the selected item stays within the budget cap of 25 USD. Substantially similar wording is acceptable.\"\n    - criterion: \"After adding the item, show the current cart.\"\n  tools_evaluation_path: \"data/Financier/scripts/Financier_task_015_amazon.py\"\n\nmetadata:\n  difficulty: medium\n  benchmark_role: test\n  preference_threads:\n    - shopping_decision_with_constraints\n    - concise_comparison_outputs\n    - single_app_transaction_verification"}
{"schema_version": "v2", "role": "Financier", "task_id": "Financier_task_016", "user_id": "Financier", "environment_id": "env_Financier", "title": "", "display_title": "For our upcoming macroeconomic backtesting project, please collect the latest daily US Treasu...", "description": "", "task_type": "val", "difficulty": "medium", "trigger_type": "user", "initial_input": "For our upcoming macroeconomic backtesting project, please collect the latest daily US Treasury yield curve data for the standard intermediate to long-term benchmark tenors (from 1 year up to 20 years) and provide a brief definition of the rate fixing methodology standardly used for these specific yields.", "hidden_intents": ["The response must include the specific standard tenors: 1Y (or 1 year), 2Y, 3Y, 5Y, 7Y, 10Y, and 20Y.", "The definition must explain the Constant Maturity Treasury (CMT) interpolation methodology.", "The response must provide the exact data source used."], "hidden_intent_count": 3, "has_objectives": false, "objectives_json": "", "metadata_json": "{\"difficulty\": \"medium\"}", "yaml_path": "data/Financier/tasks/Financier_task_016/task.yaml", "yaml": "schema_version: v2\ntask_id: Financier_task_016\nuser_id: Financier\nenvironment_id: env_Financier\ntitle: ''\ndescription: ''\ntask_type: val\ntrigger:\n  type: user\n\nintent:\n  initial_input: \"For our upcoming macroeconomic backtesting project, please collect\\\n    \\ the latest daily US Treasury yield curve data for the standard intermediate\\\n    \\ to long-term benchmark tenors (from 1 year up to 20 years) and provide a brief\\\n    \\ definition of the rate fixing methodology standardly used for these specific\\\n    \\ yields.\"\n  hidden_intent:\n    - content: \"The response must include the specific standard tenors: 1Y (or 1 year),\\\n        \\ 2Y, 3Y, 5Y, 7Y, 10Y, and 20Y.\"\n    - content: \"The definition must explain the Constant Maturity Treasury (CMT) interpolation\\\n        \\ methodology.\"\n    - content: \"The response must provide the exact data source used.\"\nmetadata:\n  difficulty: medium"}
{"schema_version": "v2", "role": "Financier", "task_id": "Financier_task_017", "user_id": "Financier", "environment_id": "env_Financier", "title": "", "display_title": "For the final component of our Enterprise Macro-Risk Data Platform, we need to build a distri...", "description": "", "task_type": "long_term", "difficulty": "medium", "trigger_type": "user", "initial_input": "For the final component of our Enterprise Macro-Risk Data Platform, we need to build a distributed PySpark pipeline to analyze the correlation between market volatility and regulatory operational risk. Please write a PySpark script that ingests BOTH the raw Kenneth French daily CSV archive and the raw CFPB complaints CSV archive. The script must clean both datasets, aggregate them to a monthly frequency, and join them. Calculate the monthly rolling volatility of the core market factor, the average operational response lag, and the adjusted monthly complaint volume. Provide the executable script directly saved as 'risk_correlation_pipeline.py' in the workspace.", "hidden_intents": ["FF5 Memory: The script explicitly implements logic to skip/filter the metadata headers and copyright footers of the Kenneth French file, and accurately selects the standard factors (Mkt-RF, SMB, HML, RMW, CMA, RF).", "CFPB Schema  and  Temporal Memory: The script ensures 'Complaint ID' uniqueness and explicitly calculates the response time lag between 'Date received' and 'Date sent to company' handling edge cases.", "Domain Context Memory: The script explicitly normalizes or adjusts the aggregated absolute monthly complaint volumes by a proxy (e.g., regional population or company market share), remembering that raw volume is statistically biased.", "Distributed Engineering Limit: The script strictly utilizes pure PySpark DataFrame/Window APIs for the joining and rolling calculations, completely avoiding collecting data to the driver node (e.g., no `.toPandas()`)."], "hidden_intent_count": 4, "has_objectives": true, "objectives_json": "{\"checklist\": [{\"criterion\": \"The script correctly implements mathematical normalization by explicitly calculating a trailing moving average (e.g., over 12 months) and dividing the raw monthly complaint volume by this moving average to detrend the data.\"}, {\"criterion\": \"The script calculates the monthly realized volatility correctly by computing the standard deviation of *daily* market returns within the month.\"}, {\"criterion\": \"The script explicitly defines trailing/backward-looking window boundaries (e.g., looking back 11 or 12 months, ending at the current month/row 0) for the moving average and rolling volatility calculations. It strictly avoids centered windows or incorporating future data (e.g., avoiding `unboundedFollowing`), ensuring strict time-series causality.\"}, {\"criterion\": \"The script explicitly demonstrates how it converts daily dates into a unified monthly key (e.g., extracting Year-Month, or truncating to the first/last day of the month) for BOTH datasets BEFORE executing the join, ensuring the market and regulatory data align correctly on the same temporal granularity.\"}], \"files_read\": [\"risk_correlation_pipeline.py\"], \"tools_evaluation_path\": \"data/Financier/scripts/Financier_task_017.py\"}", "metadata_json": "{\"difficulty\": \"medium\"}", "yaml_path": "data/Financier/tasks/Financier_task_017/task.yaml", "yaml": "schema_version: v2\ntask_id: Financier_task_017\nuser_id: Financier\nenvironment_id: env_Financier\ntitle: ''\ndescription: ''\ntask_type: long_term\ntrigger:\n  type: user\n\nintent:\n  initial_input: \"For the final component of our Enterprise Macro-Risk Data Platform,\\\n    \\ we need to build a distributed PySpark pipeline to analyze the correlation between\\\n    \\ market volatility and regulatory operational risk. Please write a PySpark script\\\n    \\ that ingests BOTH the raw Kenneth French daily CSV archive and the raw CFPB\\\n    \\ complaints CSV archive. The script must clean both datasets, aggregate them\\\n    \\ to a monthly frequency, and join them. Calculate the monthly rolling volatility\\\n    \\ of the core market factor, the average operational response lag, and the adjusted\\\n    \\ monthly complaint volume. Provide the executable script directly saved as 'risk_correlation_pipeline.py'\\\n    \\ in the workspace.\"\n  hidden_intent:\n    - content: \"FF5 Memory: The script explicitly implements logic to skip/filter\\\n        \\ the metadata headers and copyright footers of the Kenneth French file, and\\\n        \\ accurately selects the standard factors (Mkt-RF, SMB, HML, RMW, CMA, RF).\"\n    - content: \"CFPB Schema  and  Temporal Memory: The script ensures 'Complaint ID' uniqueness\\\n        \\ and explicitly calculates the response time lag between 'Date received'\\\n        \\ and 'Date sent to company' handling edge cases.\"\n    - content: \"Domain Context Memory: The script explicitly normalizes or adjusts\\\n        \\ the aggregated absolute monthly complaint volumes by a proxy (e.g., regional\\\n        \\ population or company market share), remembering that raw volume is statistically\\\n        \\ biased.\"\n    - content: \"Distributed Engineering Limit: The script strictly utilizes pure PySpark\\\n        \\ DataFrame/Window APIs for the joining and rolling calculations, completely\\\n        \\ avoiding collecting data to the driver node (e.g., no `.toPandas()`).\"\n\nobjectives:\n  files_read:\n    - risk_correlation_pipeline.py\n  tools_evaluation_path: \"data/Financier/scripts/Financier_task_017.py\"\n  checklist:\n    - criterion: \"The script correctly implements mathematical normalization by explicitly\\\n        \\ calculating a trailing moving average (e.g., over 12 months) and dividing\\\n        \\ the raw monthly complaint volume by this moving average to detrend the data.\"\n    - criterion: \"The script calculates the monthly realized volatility correctly\\\n        \\ by computing the standard deviation of *daily* market returns within the\\\n        \\ month.\"\n    - criterion: \"The script explicitly defines trailing/backward-looking window boundaries\\\n        \\ (e.g., looking back 11 or 12 months, ending at the current month/row 0)\\\n        \\ for the moving average and rolling volatility calculations. It strictly\\\n        \\ avoids centered windows or incorporating future data (e.g., avoiding `unboundedFollowing`),\\\n        \\ ensuring strict time-series causality.\"\n    - criterion: \"The script explicitly demonstrates how it converts daily dates into\\\n        \\ a unified monthly key (e.g., extracting Year-Month, or truncating to the\\\n        \\ first/last day of the month) for BOTH datasets BEFORE executing the join,\\\n        \\ ensuring the market and regulatory data align correctly on the same temporal\\\n        \\ granularity.\"\nmetadata:\n  difficulty: medium"}
{"schema_version": "v2", "role": "Financier", "task_id": "Financier_task_018", "user_id": "Financier", "environment_id": "env_Financier", "title": "", "display_title": "Regarding our macroeconomic backtesting project, I've noticed discrepancies in historical Tre...", "description": "", "task_type": "long_term", "difficulty": "medium", "trigger_type": "user", "initial_input": "Regarding our macroeconomic backtesting project, I've noticed discrepancies in historical Treasury yields across Yahoo Finance versus official government data. Which specific source should we use as our absolute standard for all future historical macroeconomic backtesting?", "hidden_intents": ["Explicitly designate the U.S. Department of the Treasury as the definitive data source.", "Acknowledge that the core discrepancy arises because commercial platforms reflect real-time/traded market quotes, whereas official sources use Constant Maturity (CMT) interpolation."], "hidden_intent_count": 2, "has_objectives": false, "objectives_json": "", "metadata_json": "{\"difficulty\": \"medium\"}", "yaml_path": "data/Financier/tasks/Financier_task_018/task.yaml", "yaml": "schema_version: v2\ntask_id: Financier_task_018\nuser_id: Financier\nenvironment_id: env_Financier\ntitle: ''\ndescription: ''\ntask_type: long_term\ntrigger:\n  type: user\n\nintent:\n  initial_input: \"Regarding our macroeconomic backtesting project, I've noticed discrepancies\\\n    \\ in historical Treasury yields across Yahoo Finance versus official government\\\n    \\ data. Which specific source should we use as our absolute standard for all future\\\n    \\ historical macroeconomic backtesting?\"\n\n  hidden_intent:\n    - content: \"Explicitly designate the U.S. Department of the Treasury as the definitive\\\n        \\ data source.\"\n    - content: \"Acknowledge that the core discrepancy arises because commercial platforms\\\n        \\ reflect real-time/traded market quotes, whereas official sources use Constant\\\n        \\ Maturity (CMT) interpolation.\"\n\nmetadata:\n  difficulty: medium"}
{"schema_version": "v2", "role": "Financier", "task_id": "Financier_task_019", "user_id": "Financier", "environment_id": "env_Financier", "title": "", "display_title": "For the final component of our Fixed Income Analytics Platform yield curve project, please pr...", "description": "", "task_type": "long_term", "difficulty": "medium", "trigger_type": "user", "initial_input": "For the final component of our Fixed Income Analytics Platform yield curve project, please provide an executable Python implementation. The script must fetch the daily par yield curve data, perform the daily fitting over the specified 3-year historical window, and calculate the fitting error. The script must output three items: a daily parameter DataFrame, a fitted yields DataFrame, and the daily error metric. Provide a brief text explanation at the end detailing the economic intuition of the parameters and the limitation of this method under extreme curve shapes. CRITICAL CONSTRAINT: Keep the script highly concise by omitting excessive console logging, massive docstrings, or overly verbose comments. Save the script directly as 'yield_curve_fitter.py' in the workspace.", "hidden_intents": ["Model Memory: The script explicitly implements the Nelson-Siegel mathematical formula rather than the Svensson model or generic cubic splines.", "Source Memory: The script explicitly fetches data from the U.S. Treasury official endpoint rather than relying on secondary aggregators like Yahoo Finance or FRED.", "Time Window Memory: The script explicitly filters the fetched dataset to retain exactly the last 3 years of historical daily data.", "Economic Intuition: The text explanation explicitly links the model parameters to 'Level', 'Slope', and 'Curvature' components of the yield curve."], "hidden_intent_count": 4, "has_objectives": true, "objectives_json": "{\"checklist\": [{\"criterion\": \"The Python code correctly translates the Nelson-Siegel model into algorithmic logic, explicitly computing both the 'slope' and 'curvature' factor loadings using the exponential decay parameter (e.g., calculations involving `np.exp`).\"}, {\"criterion\": \"The printed text explanation explicitly maps the Nelson-Siegel parameters to their geometric meanings: 'Level' (beta0/long-term), 'Slope' (beta1/short-term), and 'Curvature/Hump' (beta2/medium-term).\"}, {\"criterion\": \"The printed text explanation explicitly states that the primary limitation of the model under extreme curves is its mathematical inability to fit 'double-hump', 'multi-hump', or severely inverted shapes with multiple inflection points.\"}], \"files_read\": [\"yield_curve_fitter.py\"], \"tools_evaluation_path\": \"data/Financier/scripts/Financier_task_019.py\"}", "metadata_json": "{\"difficulty\": \"medium\"}", "yaml_path": "data/Financier/tasks/Financier_task_019/task.yaml", "yaml": "schema_version: v2\ntask_id: Financier_task_019\nuser_id: Financier\nenvironment_id: env_Financier\ntitle: ''\ndescription: ''\ntask_type: long_term\ntrigger:\n  type: user\n\nintent:\n  initial_input: \"For the final component of our Fixed Income Analytics Platform yield\\\n    \\ curve project, please provide an executable Python implementation. The script\\\n    \\ must fetch the daily par yield curve data, perform the daily fitting over the\\\n    \\ specified 3-year historical window, and calculate the fitting error. The script\\\n    \\ must output three items: a daily parameter DataFrame, a fitted yields DataFrame,\\\n    \\ and the daily error metric. Provide a brief text explanation at the end detailing\\\n    \\ the economic intuition of the parameters and the limitation of this method under\\\n    \\ extreme curve shapes. CRITICAL CONSTRAINT: Keep the script highly concise by\\\n    \\ omitting excessive console logging, massive docstrings, or overly verbose comments.\\\n    \\ Save the script directly as 'yield_curve_fitter.py' in the workspace.\"\n  hidden_intent:\n    - content: \"Model Memory: The script explicitly implements the Nelson-Siegel mathematical\\\n        \\ formula rather than the Svensson model or generic cubic splines.\"\n    - content: \"Source Memory: The script explicitly fetches data from the U.S. Treasury\\\n        \\ official endpoint rather than relying on secondary aggregators like Yahoo\\\n        \\ Finance or FRED.\"\n    - content: \"Time Window Memory: The script explicitly filters the fetched dataset\\\n        \\ to retain exactly the last 3 years of historical daily data.\"\n    - content: \"Economic Intuition: The text explanation explicitly links the model\\\n        \\ parameters to 'Level', 'Slope', and 'Curvature' components of the yield\\\n        \\ curve.\"\n\nobjectives:\n  files_read:\n    - yield_curve_fitter.py\n  tools_evaluation_path: \"data/Financier/scripts/Financier_task_019.py\"\n  checklist:\n    - criterion: \"The Python code correctly translates the Nelson-Siegel model into\\\n        \\ algorithmic logic, explicitly computing both the 'slope' and 'curvature'\\\n        \\ factor loadings using the exponential decay parameter (e.g., calculations\\\n        \\ involving `np.exp`).\"\n    - criterion: \"The printed text explanation explicitly maps the Nelson-Siegel parameters\\\n        \\ to their geometric meanings: 'Level' (beta0/long-term), 'Slope' (beta1/short-term),\\\n        \\ and 'Curvature/Hump' (beta2/medium-term).\"\n    - criterion: \"The printed text explanation explicitly states that the primary\\\n        \\ limitation of the model under extreme curves is its mathematical inability\\\n        \\ to fit 'double-hump', 'multi-hump', or severely inverted shapes with multiple\\\n        \\ inflection points.\"\n\nmetadata:\n  difficulty: medium"}
{"schema_version": "v2", "role": "Financier", "task_id": "Financier_task_020", "user_id": "Financier", "environment_id": "env_Financier", "title": "", "display_title": "Regarding our macroeconomic backtesting project, please write a Python script to fetch the hi...", "description": "", "task_type": "long_term", "difficulty": "medium", "trigger_type": "user", "initial_input": "Regarding our macroeconomic backtesting project, please write a Python script to fetch the historical daily US Treasury Yield curve data for exactly the past 10 years (from 2016-01-01 to 2026-01-01), utilizing the same set of standard tenors we established earlier. Crucially, I need a valid yield value populated for the very first day (2016-01-01). You only need to write and save the script; do not attempt to execute it. The script must appropriately handle market holiday data gaps to ensure continuous time-series alignment. The script must export the final dataframe to a CSV file named 'UST_Yield_Curve_Daily.csv'. Please save the script directly into the current workspace as 'Data_Pipeline.py'.", "hidden_intents": ["The script must explicitly fetch data for the exact standard curve requested previously: 1Y (or 1 year), 2Y, 3Y, 5Y, 7Y, 10Y, and 20Y.", "The script must explicitly use the U.S. Treasury as the absolute data source.", "The script must always utilize a forward-fill method to handle missing daily data caused by holidays.", "The script must strictly output only the requested tenors, avoiding any unsolicited derived metrics (like yield spreads) or unrequested tenors."], "hidden_intent_count": 4, "has_objectives": true, "objectives_json": "{\"checklist\": [{\"criterion\": \"Full Calendar Alignment: The script explicitly aligns the dataset to a continuous 365-day daily calendar (e.g., utilizing `pd.date_range` with `freq='D'` or equivalent logic) to ensure weekends and holidays are included as empty rows before gap-filling.\"}, {\"criterion\": \"Anti Look-Ahead Bias: The script explicitly describes a strategy to fetch historical data prior to the target start date (e.g., fetching year 2015) to act as a data buffer for forward-filling, and strictly prohibits the use of back-filling (`bfill`) which introduces future data into past dates.\"}, {\"criterion\": \"Target Tenor Precision: The script explicitly filters the yield curve columns to retain exactly the 7 intermediate-to-long-term benchmark tenors (1Y, 2Y, 3Y, 5Y, 7Y, 10Y, 20Y), strictly omitting short-term bill tenors (like 1-month or 6-month) and the 30-year bond.\"}], \"files_read\": [\"Data_Pipeline.py\"], \"tools_evaluation_path\": \"data/Financier/scripts/Financier_task_020.py\"}", "metadata_json": "{\"difficulty\": \"medium\"}", "yaml_path": "data/Financier/tasks/Financier_task_020/task.yaml", "yaml": "schema_version: v2\ntask_id: Financier_task_020\nuser_id: Financier\nenvironment_id: env_Financier\ntitle: ''\ndescription: ''\ntask_type: long_term\ntrigger:\n  type: user\n\nintent:\n  initial_input: \"Regarding our macroeconomic backtesting project, please write a\\\n    \\ Python script to fetch the historical daily US Treasury Yield curve data for\\\n    \\ exactly the past 10 years (from 2016-01-01 to 2026-01-01), utilizing the same\\\n    \\ set of standard tenors we established earlier. Crucially, I need a valid yield\\\n    \\ value populated for the very first day (2016-01-01). You only need to write\\\n    \\ and save the script; do not attempt to execute it. The script must appropriately\\\n    \\ handle market holiday data gaps to ensure continuous time-series alignment.\\\n    \\ The script must export the final dataframe to a CSV file named 'UST_Yield_Curve_Daily.csv'.\\\n    \\ Please save the script directly into the current workspace as 'Data_Pipeline.py'.\"\n\n  hidden_intent:\n    - content: \"The script must explicitly fetch data for the exact standard curve\\\n        \\ requested previously: 1Y (or 1 year), 2Y, 3Y, 5Y, 7Y, 10Y, and 20Y.\"\n    - content: \"The script must explicitly use the U.S. Treasury as the absolute data\\\n        \\ source.\"\n    - content: \"The script must always utilize a forward-fill method to handle missing\\\n        \\ daily data caused by holidays.\"\n    - content: \"The script must strictly output only the requested tenors, avoiding\\\n        \\ any unsolicited derived metrics (like yield spreads) or unrequested tenors.\"\nobjectives:\n  files_read:\n    - Data_Pipeline.py\n  tools_evaluation_path: \"data/Financier/scripts/Financier_task_020.py\"\n  checklist:\n    - criterion: \"Full Calendar Alignment: The script explicitly aligns the dataset\\\n        \\ to a continuous 365-day daily calendar (e.g., utilizing `pd.date_range`\\\n        \\ with `freq='D'` or equivalent logic) to ensure weekends and holidays are\\\n        \\ included as empty rows before gap-filling.\"\n    - criterion: \"Anti Look-Ahead Bias: The script explicitly describes a strategy\\\n        \\ to fetch historical data prior to the target start date (e.g., fetching\\\n        \\ year 2015) to act as a data buffer for forward-filling, and strictly prohibits\\\n        \\ the use of back-filling (`bfill`) which introduces future data into past\\\n        \\ dates.\"\n    - criterion: \"Target Tenor Precision: The script explicitly filters the yield\\\n        \\ curve columns to retain exactly the 7 intermediate-to-long-term benchmark\\\n        \\ tenors (1Y, 2Y, 3Y, 5Y, 7Y, 10Y, 20Y), strictly omitting short-term bill\\\n        \\ tenors (like 1-month or 6-month) and the 30-year bond.\"\nmetadata:\n  difficulty: medium"}
{"schema_version": "v2", "role": "law_trainee", "task_id": "law_trainee_task_001", "user_id": "law_trainee", "environment_id": "env_law_trainee", "title": "Drafting a Property Dispute Memorandum: Party Profiles in Contrast and Fact-Based Anchoring", "display_title": "Drafting a Property Dispute Memorandum: Party Profiles in Contrast and Fact-Based Anchoring", "description": "To prepare for an afternoon partners' case-review session, the user asks the Agent to help draft an internal memorandum in a concealed-marital-asset dispute. By stressing the partners' preference for facts over theory, the scenario naturally embeds layout conventions and argumentation standards for future case analyses.", "task_type": "long_term", "difficulty": "easy", "trigger_type": "user", "initial_input": "This afternoon I have a case discussion session and need to walk the partners through the Li–Wang divorce and property-division matter. Please help me draft a first-pass internal case memorandum.\n\nThe facts in outline are as follows. Our client, Ms. Wang, 45, has been married to Mr. Li for eighteen years, approximately fifteen of which she was a full-time homemaker caring solely for two minor children (the elder daughter attends private junior high; the younger son is in primary school). Day-to-day household spending runs through a supplementary card on a jointly named account, but she has only a spending limit—no online banking or investment authority—and holds no separate deposits, real property, or securities accounts in her own name. She has no clear picture of aggregate family assets, external receivables and payables, or the equity structures of multiple affiliated entities under Mr. Li; she has only fragmentary hearsay that Mr. Li years ago may have had his parents nominally hold an out-of-province retail property on his behalf.\n\nMr. Li, 48, is the de facto controller of an unlisted hard-tech operating company and three upstream/downstream shell companies. Group annual throughput exceeds RMB ten million; his personal tax-returned annual income is on the order of RMB three million—but major family financial decisions, chops, USB tokens, and finance counterparties are concentrated with him alone or with his cousin (deputy finance director). Over the past six months, Mr. Li first transferred roughly RMB five million in multiple tranches to his mother's personal account, citing repayment of principal and interest on an early business loan from her (some transfers were noted as \"inter-account\" / \"current account\" items). He then contractually transferred approximately twelve percent of the core company's equity to a limited partnership controlled by his cousin, at a price materially below contemporaneous capital-increase / financing valuations. During the same period, a Hong Kong shell company that Mr. Li actually runs—and of which Ms. Wang had never heard—surfaced on supplier payment advices. In addition, Mr. Li has repeatedly pressed Ms. Wang to sign multiple blank powers of attorney and \"family asset statements,\" and has refused to provide complete bank statements and audited reports, citing \"tax planning\" and \"employee equity platform adjustment.\" We have strong reason to suspect that Mr. Li is systematically diverting and concealing marital property, using nominee holdings, related-party dealings, and offshore entities to fragment the evidentiary trail; we anticipate material difficulty on proof and preservation going forward.", "hidden_intents": ["In the opening section, include a module titled along the lines of \"Party Profiles in Contrast,\" setting out both parties' occupation, income, control over assets, and information asymmetry.", "In the analysis of any divorce case, the final step should be to identify the favorable evidence that our client ought to collect."], "hidden_intent_count": 2, "has_objectives": false, "objectives_json": "", "metadata_json": "{\"difficulty\": \"easy\"}", "yaml_path": "data/law_trainee/tasks/law_trainee_task_001/task.yaml", "yaml": "schema_version: v2\ntask_id: law_trainee_task_001\nuser_id: law_trainee\nenvironment_id: env_law_trainee\n\ntitle: \"Drafting a Property Dispute Memorandum: Party Profiles in Contrast and Fact-Based Anchoring\"\ndescription: \"To prepare for an afternoon partners' case-review session, the user asks the Agent to help draft an internal memorandum in a concealed-marital-asset dispute. By stressing the partners' preference for facts over theory, the scenario naturally embeds layout conventions and argumentation standards for future case analyses.\"\n\ntask_type: long_term\n\ntrigger:\n  type: user\n\nintent:\n  initial_input: |\n    This afternoon I have a case discussion session and need to walk the partners through the Li–Wang divorce and property-division matter. Please help me draft a first-pass internal case memorandum.\n\n    The facts in outline are as follows. Our client, Ms. Wang, 45, has been married to Mr. Li for eighteen years, approximately fifteen of which she was a full-time homemaker caring solely for two minor children (the elder daughter attends private junior high; the younger son is in primary school). Day-to-day household spending runs through a supplementary card on a jointly named account, but she has only a spending limit—no online banking or investment authority—and holds no separate deposits, real property, or securities accounts in her own name. She has no clear picture of aggregate family assets, external receivables and payables, or the equity structures of multiple affiliated entities under Mr. Li; she has only fragmentary hearsay that Mr. Li years ago may have had his parents nominally hold an out-of-province retail property on his behalf.\n\n    Mr. Li, 48, is the de facto controller of an unlisted hard-tech operating company and three upstream/downstream shell companies. Group annual throughput exceeds RMB ten million; his personal tax-returned annual income is on the order of RMB three million—but major family financial decisions, chops, USB tokens, and finance counterparties are concentrated with him alone or with his cousin (deputy finance director). Over the past six months, Mr. Li first transferred roughly RMB five million in multiple tranches to his mother's personal account, citing repayment of principal and interest on an early business loan from her (some transfers were noted as \"inter-account\" / \"current account\" items). He then contractually transferred approximately twelve percent of the core company's equity to a limited partnership controlled by his cousin, at a price materially below contemporaneous capital-increase / financing valuations. During the same period, a Hong Kong shell company that Mr. Li actually runs—and of which Ms. Wang had never heard—surfaced on supplier payment advices. In addition, Mr. Li has repeatedly pressed Ms. Wang to sign multiple blank powers of attorney and \"family asset statements,\" and has refused to provide complete bank statements and audited reports, citing \"tax planning\" and \"employee equity platform adjustment.\" We have strong reason to suspect that Mr. Li is systematically diverting and concealing marital property, using nominee holdings, related-party dealings, and offshore entities to fragment the evidentiary trail; we anticipate material difficulty on proof and preservation going forward.\n\n  hidden_intent:\n    - content: \"In the opening section, include a module titled along the lines of \\\"Party Profiles in Contrast,\\\" setting out both parties' occupation, income, control over assets, and information asymmetry.\"\n    - content: \"In the analysis of any divorce case, the final step should be to identify the favorable evidence that our client ought to collect.\"\n\nmetadata:\n  difficulty: easy"}
{"schema_version": "v2", "role": "law_trainee", "task_id": "law_trainee_task_002", "user_id": "law_trainee", "environment_id": "env_law_trainee", "title": "Drafting a Sales-Contract Complaint: One-to-One Pairing of Fact Narrative and Basis of Claim", "display_title": "Drafting a Sales-Contract Complaint: One-to-One Pairing of Fact Narrative and Basis of Claim", "description": "To suit a partner’s rigorous, academic reading habits, the user requires the agent, when drafting a sales-contract complaint, to interleave case facts with the specific legal bases that support them—establishing the underlying logic for how complaints will be drafted going forward.", "task_type": "long_term", "difficulty": "medium", "trigger_type": "user", "initial_input": "This afternoon I have a sales-contract dispute and need to turn in a first draft of the complaint to my supervising attorney for review. The facts are simple: the plaintiff shipped the goods; after the defendant took delivery, it never paid the balance. We are suing to collect payment. The defendant now argues that our equipment was defective, but in fact they are long past the inspection period agreed in the contract and never sent us any written notice of a quality objection.", "hidden_intents": ["When drafting the “Facts and Legal Grounds” section of any complaint or legal memorandum, after each core factual basis or allegation, the related legal basis (specific statutory provision) must be stated immediately afterward."], "hidden_intent_count": 1, "has_objectives": false, "objectives_json": "", "metadata_json": "{\"difficulty\": \"medium\", \"theme\": \"pleading_drafting_logic_standard\"}", "yaml_path": "data/law_trainee/tasks/law_trainee_task_002/task.yaml", "yaml": "schema_version: v2\ntask_id: law_trainee_task_002\nuser_id: law_trainee\nenvironment_id: env_law_trainee\n\ntitle: \"Drafting a Sales-Contract Complaint: One-to-One Pairing of Fact Narrative and Basis of Claim\"\ndescription: \"To suit a partner’s rigorous, academic reading habits, the user requires the agent, when drafting a sales-contract complaint, to interleave case facts with the specific legal bases that support them—establishing the underlying logic for how complaints will be drafted going forward.\"\n\ntask_type: long_term\n\ntrigger:\n  type: user\n\nintent:\n  initial_input: |\n    This afternoon I have a sales-contract dispute and need to turn in a first draft of the complaint to my supervising attorney for review. The facts are simple: the plaintiff shipped the goods; after the defendant took delivery, it never paid the balance. We are suing to collect payment. The defendant now argues that our equipment was defective, but in fact they are long past the inspection period agreed in the contract and never sent us any written notice of a quality objection.\n\n  hidden_intent:\n    - content: \"When drafting the “Facts and Legal Grounds” section of any complaint or legal memorandum, after each core factual basis or allegation, the related legal basis (specific statutory provision) must be stated immediately afterward.\"\nmetadata:\n  difficulty: medium\n  theme: pleading_drafting_logic_standard"}
{"schema_version": "v2", "role": "law_trainee", "task_id": "law_trainee_task_003", "user_id": "law_trainee", "environment_id": "env_law_trainee", "title": "Revise a Compliance Review Report: Add Overall Opinion and Project-Level Compliance Risk Rating", "display_title": "Revise a Compliance Review Report: Add Overall Opinion and Project-Level Compliance Risk Rating", "description": "The user has a first-draft compliance review report with topical findings and piecemeal remediation ideas, but it lacks a deliverable overall project compliance opinion and an explicit project-wide compliance risk rating. The Agent should revise or rewrite the report while preserving substantive analysis so it meets internal circulation standards.", "task_type": "long_term", "difficulty": "easy", "trigger_type": "user", "initial_input": "My supervising attorney asked me to revise the project compliance review report I wrote earlier. The draft is in Compliance_Report_Draft.md.", "hidden_intents": ["The compliance review report must state an overall compliance risk rating for the entire project development effort.", "The final project compliance review opinion must be stated in binary form: pass or fail."], "hidden_intent_count": 2, "has_objectives": false, "objectives_json": "", "metadata_json": "{\"difficulty\": \"easy\"}", "yaml_path": "data/law_trainee/tasks/law_trainee_task_003/task.yaml", "yaml": "schema_version: v2\ntask_id: law_trainee_task_003\nuser_id: law_trainee\nenvironment_id: env_law_trainee\n\ntitle: \"Revise a Compliance Review Report: Add Overall Opinion and Project-Level Compliance Risk Rating\"\ndescription: \"The user has a first-draft compliance review report with topical findings and piecemeal remediation ideas, but it lacks a deliverable overall project compliance opinion and an explicit project-wide compliance risk rating. The Agent should revise or rewrite the report while preserving substantive analysis so it meets internal circulation standards.\"\n\ntask_type: long_term\n\ntrigger:\n  type: user\n\nintent:\n  initial_input: |\n    My supervising attorney asked me to revise the project compliance review report I wrote earlier. The draft is in Compliance_Report_Draft.md.\n\n  hidden_intent:\n\n    - content: \"The compliance review report must state an overall compliance risk rating for the entire project development effort.\"\n    - content: \"The final project compliance review opinion must be stated in binary form: pass or fail.\"\nmetadata:\n  difficulty: easy"}
{"schema_version": "v2", "role": "law_trainee", "task_id": "law_trainee_task_004", "user_id": "law_trainee", "environment_id": "env_law_trainee", "title": "Drafting a Labor-Dispute Intake Outline: Proactively Deriving and Organizing an Evidence Checklist", "display_title": "Drafting a Labor-Dispute Intake Outline: Proactively Deriving and Organizing an Evidence Checklist", "description": "The user is preparing an initial client meeting in an unlawful-termination dispute and asks the Agent, while analyzing legal claims, to work backward and list the evidence the client urgently needs to gather—thereby habituating a practical rule that future case analyses always include an \"evidence materials\" section.", "task_type": "long_term", "difficulty": "easy", "trigger_type": "user", "initial_input": "Tomorrow morning I am meeting a new client, Mr. Zhang, on a labor dispute. Please help me prepare a preliminary intake outline and analysis.\n\nThe facts in outline are as follows. Mr. Zhang has served as Sales Director at a new-media company for three years, with a monthly salary of RMB 30,000. Last Friday HR orally notified him that he was dismissed for \\\"serious failure to perform the job and refusal to cooperate with a lawful reassignment.\\\" The employer issued no written termination notice and, that same day after work, unilaterally disabled his corporate email, enterprise WeChat (WeCom), and building-access credentials. Mr. Zhang currently has only WeChat chat history on his personal phone. He intends to file for labor arbitration and seek damages for unlawful termination (2N).", "hidden_intents": ["List the case factors most comparable to our party's situation, and provide comparable cases.", "After list the case factors most comparable to our party's situation, Analyze what evidence our party should collect to guide next steps."], "hidden_intent_count": 2, "has_objectives": false, "objectives_json": "", "metadata_json": "{\"difficulty\": \"easy\"}", "yaml_path": "data/law_trainee/tasks/law_trainee_task_004/task.yaml", "yaml": "schema_version: v2\ntask_id: law_trainee_task_004\nuser_id: law_trainee\nenvironment_id: env_law_trainee\n\ntitle: \"Drafting a Labor-Dispute Intake Outline: Proactively Deriving and Organizing an Evidence Checklist\"\ndescription: \"The user is preparing an initial client meeting in an unlawful-termination dispute and asks the Agent, while analyzing legal claims, to work backward and list the evidence the client urgently needs to gather—thereby habituating a practical rule that future case analyses always include an \\\"evidence materials\\\" section.\"\n\ntask_type: long_term\n\ntrigger:\n  type: user\n\nintent:\n  initial_input: |\n    Tomorrow morning I am meeting a new client, Mr. Zhang, on a labor dispute. Please help me prepare a preliminary intake outline and analysis.\n\n    The facts in outline are as follows. Mr. Zhang has served as Sales Director at a new-media company for three years, with a monthly salary of RMB 30,000. Last Friday HR orally notified him that he was dismissed for \\\"serious failure to perform the job and refusal to cooperate with a lawful reassignment.\\\" The employer issued no written termination notice and, that same day after work, unilaterally disabled his corporate email, enterprise WeChat (WeCom), and building-access credentials. Mr. Zhang currently has only WeChat chat history on his personal phone. He intends to file for labor arbitration and seek damages for unlawful termination (2N).\n\n  hidden_intent:\n    - content: \"List the case factors most comparable to our party's situation, and provide comparable cases.\"\n    - content: \"After list the case factors most comparable to our party's situation, Analyze what evidence our party should collect to guide next steps.\"\nmetadata:\n  difficulty: easy"}
{"schema_version": "v2", "role": "law_trainee", "task_id": "law_trainee_task_005", "user_id": "law_trainee", "environment_id": "env_law_trainee", "title": "Priming: Private-Lending Loan Agreements and Marital Joint Debt Confirmation", "display_title": "Priming: Private-Lending Loan Agreements and Marital Joint Debt Confirmation", "description": "Through a short Q&A, the supervising attorney primes the intern to keep in mind that a formal loan agreement should include a stand-alone marital joint debt confirmation or spousal acknowledgment clause where appropriate, before drafting concrete agreements from a fact pattern later. This task has no attachment and no checklist; only hidden_intent is scored for proactiveness.", "task_type": "long_term", "difficulty": "easy", "trigger_type": "user", "initial_input": "In a formal loan agreement, what other types of clauses are worth adding from a risk-control standpoint? Please answer in bullet points.", "hidden_intents": ["The answer should state that where the borrower is a natural person and the funds are used for the household’s common needs or the couple’s joint production or business operations, the contract should include a marital joint debt confirmation clause or one with equivalent legal effect.", "In the future, when helping me draft a loan agreement, please also generate an analysis of the borrower’s repayment capacity.", "In the future, when helping me draft a loan agreement, please also calculate at the end how much legal service fee should be charged if the attorney’s fee is computed at 0.1% of the loan amount."], "hidden_intent_count": 3, "has_objectives": false, "objectives_json": "", "metadata_json": "{\"difficulty\": \"easy\"}", "yaml_path": "data/law_trainee/tasks/law_trainee_task_005/task.yaml", "yaml": "schema_version: v2\ntask_id: law_trainee_task_005\nuser_id: law_trainee\nenvironment_id: env_law_trainee\n\ntitle: \"Priming: Private-Lending Loan Agreements and Marital Joint Debt Confirmation\"\ndescription: \"Through a short Q&A, the supervising attorney primes the intern to keep in mind that a formal loan agreement should include a stand-alone marital joint debt confirmation or spousal acknowledgment clause where appropriate, before drafting concrete agreements from a fact pattern later. This task has no attachment and no checklist; only hidden_intent is scored for proactiveness.\"\n\ntask_type: long_term\n\ntrigger:\n  type: user\n\nintent:\n  initial_input: |\n    In a formal loan agreement, what other types of clauses are worth adding from a risk-control standpoint? Please answer in bullet points.\n\n  hidden_intent:\n    - content: \"The answer should state that where the borrower is a natural person and the funds are used for the household’s common needs or the couple’s joint production or business operations, the contract should include a marital joint debt confirmation clause or one with equivalent legal effect.\"\n    - content: \"In the future, when helping me draft a loan agreement, please also generate an analysis of the borrower’s repayment capacity.\"   \n    \n    - content: \"In the future, when helping me draft a loan agreement, please also calculate at the end how much legal service fee should be charged if the attorney’s fee is computed at 0.1% of the loan amount.\"  \nmetadata:\n  difficulty: easy"}
{"schema_version": "v2", "role": "law_trainee", "task_id": "law_trainee_task_006", "user_id": "law_trainee", "environment_id": "env_law_trainee", "title": "Review a First-Draft Defense Brief in a Fraud Case: Fix Format and Restructure the Argument", "display_title": "Review a First-Draft Defense Brief in a Fraud Case: Fix Format and Restructure the Argument", "description": "The user deliberately supplies a first-draft defense brief in a fraud case that reads like narrative prose and lacks proper form, and asks the Agent to critique it. Hidden Intent requires the Agent to flag the absence of a proper opening and an elements-based structure, establishing a baseline preference for later criminal drafting.", "task_type": "long_term", "difficulty": "easy", "trigger_type": "user", "initial_input": "I just finished a first draft of a defense brief in a fraud case. If I file this with the court as-is, are there problems—and can you fix them?\n\nDraft text:\nYour Honor. On Zhang San’s fraud case, I think he has been wronged. When he first borrowed money from Li Si, he truly intended to run the business in good faith and even mortgaged his house back home. Later, because macro conditions were poor and his capital chain broke, he could not repay. Subjectively he absolutely did not intend to defraud Li Si of the money. Li Si also bears responsibility: when lending, he never checked Zhang San’s repayment capacity. Zhang San’s conduct is at most an economic dispute and does not constitute fraud at all. I ask the court to acquit him.", "hidden_intents": ["In revision guidance, expressly add a standard caption, proper forms of address to the court, a retained-counsel introduction, and a compliant closing and signature block.", "When handling a statement of defense task, include at the end an objective case analysis report for the client’s family and relatives, assessing the likelihood of our client prevailing in the case."], "hidden_intent_count": 2, "has_objectives": false, "objectives_json": "", "metadata_json": "{\"difficulty\": \"easy\"}", "yaml_path": "data/law_trainee/tasks/law_trainee_task_006/task.yaml", "yaml": "schema_version: v2\ntask_id: law_trainee_task_006\nuser_id: law_trainee\nenvironment_id: env_law_trainee\n\ntitle: \"Review a First-Draft Defense Brief in a Fraud Case: Fix Format and Restructure the Argument\"\ndescription: \"The user deliberately supplies a first-draft defense brief in a fraud case that reads like narrative prose and lacks proper form, and asks the Agent to critique it. Hidden Intent requires the Agent to flag the absence of a proper opening and an elements-based structure, establishing a baseline preference for later criminal drafting.\"\n\ntask_type: long_term\n\ntrigger:\n  type: user\n\nintent:\n  initial_input: |\n    I just finished a first draft of a defense brief in a fraud case. If I file this with the court as-is, are there problems—and can you fix them?\n\n    Draft text:\n    Your Honor. On Zhang San’s fraud case, I think he has been wronged. When he first borrowed money from Li Si, he truly intended to run the business in good faith and even mortgaged his house back home. Later, because macro conditions were poor and his capital chain broke, he could not repay. Subjectively he absolutely did not intend to defraud Li Si of the money. Li Si also bears responsibility: when lending, he never checked Zhang San’s repayment capacity. Zhang San’s conduct is at most an economic dispute and does not constitute fraud at all. I ask the court to acquit him.\n\n  hidden_intent:\n    - content: \"In revision guidance, expressly add a standard caption, proper forms of address to the court, a retained-counsel introduction, and a compliant closing and signature block.\"\n    - content: \"When handling a statement of defense task, include at the end an objective case analysis report for the client’s family and relatives, assessing the likelihood of our client prevailing in the case.\"\nmetadata:\n  difficulty: easy"}
{"schema_version": "v2", "role": "law_trainee", "task_id": "law_trainee_task_007", "user_id": "law_trainee", "environment_id": "env_law_trainee", "title": "Divorce-Custody Similar-Case Statistics Table → Internal Memorandum for Lead Counsel", "display_title": "Divorce-Custody Similar-Case Statistics Table → Internal Memorandum for Lead Counsel", "description": "Based on the team's curated sample table of custody holdings in divorce disputes, produce an internal similar-cases brief. Evaluation emphasizes sample limitations, normative anchoring, data hygiene, redaction, and auditable follow-up—not an outward-facing legal opinion.", "task_type": "short_term_independent", "difficulty": "medium", "trigger_type": "user", "initial_input": "I have a 2021-2025 statistics table of custody judgments in divorce disputes. My supervising attorney asked me to organize comparable cases related to \"Mr. A's divorce dispute\" from this table. The case summary is in Case_Summary.txt and the 2021-2025 statistics table is in 2021-2025Judgment_statistics.md.", "hidden_intents": ["At the very beginning, present a Party Profiles in Contrast section in tabular form, stating both parties' occupation, salary, and living arrangements.", "List the case factors most similar to Mr. A's situation.", "List the serial numbers of cases related to the factor 'long-term stable living environment.'", "List the serial numbers of cases related to the factor 'genuine wishes of a child over the age of eight.'", "List the serial numbers of cases related to the factor 'caregiving support from grandparents.'", "List the serial numbers of cases related to the factor 'child's illness requiring a stable treatment environment.'", "List the serial numbers of cases related to the factor 'strong sibling bonds / opposition to splitting custody.'", "List the serial numbers of cases related to the factor 'one parent working away from home, long commutes, or frequent travel.'", "List what related materials Mr. A should prepare in advance to improve his position on custody."], "hidden_intent_count": 9, "has_objectives": true, "objectives_json": "{\"checklist\": [{\"criterion\": \"In the Party Profiles in Contrast section, state that Mr. A's occupation is senior teacher at a municipal key senior high school, with a monthly income of RMB 15,000. Substantially similar wording is acceptable.\"}, {\"criterion\": \"In the Party Profiles in Contrast section, state that Mr. A resides in the marital home within the original school district. Substantially similar wording is acceptable.\"}, {\"criterion\": \"In the Party Profiles in Contrast section, state that Ms. B's occupation is Sales Director at a well-known foreign-invested enterprise, with a monthly income of RMB 27,000, long commutes, and frequent business travel. Substantially similar wording is acceptable.\"}, {\"criterion\": \"In the Party Profiles in Contrast section, state that Ms. B's living situation is that she has moved out of the former home and rents in the capital of a neighboring province. Substantially similar wording is acceptable.\"}, {\"criterion\": \"Identify that among the case factors most similar to Mr. A's situation is 'long-term stable living environment.'\"}, {\"criterion\": \"Identify that among the case factors most similar to Mr. A's situation is 'genuine wishes of a child over the age of eight.'\"}, {\"criterion\": \"Identify that among the case factors most similar to Mr. A's situation is 'caregiving support from grandparents.'\"}, {\"criterion\": \"Identify that among the case factors most similar to Mr. A's situation is 'child's illness requiring a stable treatment environment.'\"}, {\"criterion\": \"Identify that among the case factors most similar to Mr. A's situation is 'one parent working away from home, long commutes, or frequent travel.'\"}, {\"criterion\": \"Identify that among the case factors most similar to Mr. A's situation is 'strong sibling bonds / opposition to splitting custody.'\"}, {\"criterion\": \"List at least five case serial numbers related to the 'long-term stable living environment' factor.\"}, {\"criterion\": \"List at least five case serial numbers related to the 'genuine wishes of a child over the age of eight' factor.\"}, {\"criterion\": \"List at least five case serial numbers related to the 'caregiving support from grandparents' factor.\"}, {\"criterion\": \"List at least three case serial numbers related to the 'child's illness requiring a stable treatment environment' factor.\"}, {\"criterion\": \"List at least three case serial numbers related to the 'one parent working away from home, long commutes, or frequent travel' factor.\"}, {\"criterion\": \"List at least five case serial numbers related to the 'strong sibling bonds / opposition to splitting custody' factor.\"}, {\"criterion\": \"State that among the evidence Mr. A should prepare in advance is a certificate of working hours issued by Mr. A's school. Substantially similar wording is acceptable.\"}, {\"criterion\": \"List among the evidence Mr. A should prepare in advance the younger son's medical visit records. Substantially similar wording is acceptable.\"}, {\"criterion\": \"List among the evidence Mr. A should prepare in advance video records of grandparents' day-to-day school runs and care of the children. Substantially similar wording is acceptable.\"}]}", "metadata_json": "{\"difficulty\": \"medium\"}", "yaml_path": "data/law_trainee/tasks/law_trainee_task_007/task.yaml", "yaml": "schema_version: v2\ntask_id: law_trainee_task_007\nuser_id: law_trainee\nenvironment_id: env_law_trainee\n\ntitle: \"Divorce-Custody Similar-Case Statistics Table → Internal Memorandum for Lead Counsel\"\ndescription: \"Based on the team's curated sample table of custody holdings in divorce disputes, produce an internal similar-cases brief. Evaluation emphasizes sample limitations, normative anchoring, data hygiene, redaction, and auditable follow-up—not an outward-facing legal opinion.\"\n\ntask_type: short_term_independent\n\ntrigger:\n  type: user\n\nintent:\n  initial_input: |\n    I have a 2021-2025 statistics table of custody judgments in divorce disputes. My supervising attorney asked me to organize comparable cases related to \"Mr. A's divorce dispute\" from this table. The case summary is in Case_Summary.txt and the 2021-2025 statistics table is in 2021-2025Judgment_statistics.md.\n\n  hidden_intent:\n    - content: \"At the very beginning, present a Party Profiles in Contrast section in tabular form, stating both parties' occupation, salary, and living arrangements.\"\n    - content: \"List the case factors most similar to Mr. A's situation.\"\n    - content: \"List the serial numbers of cases related to the factor 'long-term stable living environment.'\"\n    - content: \"List the serial numbers of cases related to the factor 'genuine wishes of a child over the age of eight.'\"\n    - content: \"List the serial numbers of cases related to the factor 'caregiving support from grandparents.'\"\n    - content: \"List the serial numbers of cases related to the factor 'child's illness requiring a stable treatment environment.'\"\n    - content: \"List the serial numbers of cases related to the factor 'strong sibling bonds / opposition to splitting custody.'\"\n    - content: \"List the serial numbers of cases related to the factor 'one parent working away from home, long commutes, or frequent travel.'\"\n    - content: \"List what related materials Mr. A should prepare in advance to improve his position on custody.\"\nobjectives:\n  checklist:\n    - criterion: \"In the Party Profiles in Contrast section, state that Mr. A's occupation is senior teacher at a municipal key senior high school, with a monthly income of RMB 15,000. Substantially similar wording is acceptable.\"\n\n    - criterion: \"In the Party Profiles in Contrast section, state that Mr. A resides in the marital home within the original school district. Substantially similar wording is acceptable.\"\n\n    - criterion: \"In the Party Profiles in Contrast section, state that Ms. B's occupation is Sales Director at a well-known foreign-invested enterprise, with a monthly income of RMB 27,000, long commutes, and frequent business travel. Substantially similar wording is acceptable.\"\n\n    - criterion: \"In the Party Profiles in Contrast section, state that Ms. B's living situation is that she has moved out of the former home and rents in the capital of a neighboring province. Substantially similar wording is acceptable.\"\n\n    - criterion: \"Identify that among the case factors most similar to Mr. A's situation is 'long-term stable living environment.'\"\n\n    - criterion: \"Identify that among the case factors most similar to Mr. A's situation is 'genuine wishes of a child over the age of eight.'\"\n\n    - criterion: \"Identify that among the case factors most similar to Mr. A's situation is 'caregiving support from grandparents.'\"\n\n    - criterion: \"Identify that among the case factors most similar to Mr. A's situation is 'child's illness requiring a stable treatment environment.'\"\n\n    - criterion: \"Identify that among the case factors most similar to Mr. A's situation is 'one parent working away from home, long commutes, or frequent travel.'\"\n\n    - criterion: \"Identify that among the case factors most similar to Mr. A's situation is 'strong sibling bonds / opposition to splitting custody.'\"\n\n    - criterion: \"List at least five case serial numbers related to the 'long-term stable living environment' factor.\"\n\n    - criterion: \"List at least five case serial numbers related to the 'genuine wishes of a child over the age of eight' factor.\"\n\n    - criterion: \"List at least five case serial numbers related to the 'caregiving support from grandparents' factor.\"\n\n    - criterion: \"List at least three case serial numbers related to the 'child's illness requiring a stable treatment environment' factor.\"\n\n    - criterion: \"List at least three case serial numbers related to the 'one parent working away from home, long commutes, or frequent travel' factor.\"\n\n    - criterion: \"List at least five case serial numbers related to the 'strong sibling bonds / opposition to splitting custody' factor.\"\n\n    - criterion: \"State that among the evidence Mr. A should prepare in advance is a certificate of working hours issued by Mr. A's school. Substantially similar wording is acceptable.\"\n\n    - criterion: \"List among the evidence Mr. A should prepare in advance the younger son's medical visit records. Substantially similar wording is acceptable.\"\n\n    - criterion: \"List among the evidence Mr. A should prepare in advance video records of grandparents' day-to-day school runs and care of the children. Substantially similar wording is acceptable.\"\n\nmetadata:\n  difficulty: medium"}
{"schema_version": "v2", "role": "law_trainee", "task_id": "law_trainee_task_008", "user_id": "law_trainee", "environment_id": "env_law_trainee", "title": "Internal DD Deficiency Log to External Supplement Request", "display_title": "Internal DD Deficiency Log to External Supplement Request", "description": "Prepare one short internal deficiency log in Simple Note for the related-party loan review, then send one practical supplementation email in Gmail. Evaluation focuses on whether the note and email capture the three missing items, the matter context, and a clear deadline/fallback.", "task_type": "short_term_independent", "difficulty": "medium", "trigger_type": "user", "initial_input": "My Simple Note email is stmcco@gmail.com and password is 'i#vhWEQ'. My Gmail email is stmcco@gmail.com and password is 'n3v=I[w'.\n\nI still need to get the related-party loan review notes into circulation tonight, but a few DD gaps are holding this up. Please pull the outstanding points into one Simple Note entry for internal circulation and then turn that into a supplementation email to la-mcco@gmail.com.\n\nI need something clean enough to use internally and clear enough to send out. If there is already a supplementation thread in Gmail, keep this with that thread. Please send the actual email in this pass rather than leaving me with a draft or asking me to approve wording first.", "hidden_intents": ["Use the Simple Note title \"Internal DD Deficiency Log — Related-Party Loan Review\" and identify these three missing items: the board resolution for the related-party borrowing, the latest shareholder register / cap table, and the signed funds-flow explanation.", "In the Simple Note, briefly explain why each of those three items matters for the related-party loan review.", "In the Gmail subject and body, make clear that this is a supplementation request for the related-party loan review, present the three missing items as a numbered list, ask for the materials by 5:00 p.m. tomorrow, ask the recipient to flag any delay immediately, and give a practical fallback if a final document is not ready.", "If there is already a supplementation thread in Gmail, keep this request in that existing thread instead of starting a separate unrelated thread.", "Send the actual supplementation email in this pass rather than leaving only a draft or asking for wording approval first."], "hidden_intent_count": 5, "has_objectives": true, "objectives_json": "{\"checklist\": [{\"criterion\": \"In the internal Simple Note, use the title \\\"Internal DD Deficiency Log — Related-Party Loan Review.\\\"\"}, {\"criterion\": \"In the internal Simple Note, identify all three missing items: the board resolution for the related-party borrowing, the latest shareholder register / cap table, and the signed funds-flow explanation.\"}, {\"criterion\": \"In the internal Simple Note, briefly explain why the board resolution, shareholder register / cap table, and signed funds-flow explanation matter for the review. Substantially similar wording is acceptable.\"}, {\"criterion\": \"In the Gmail subject line, identify the matter as the related-party loan review and make clear that the email is a supplementation request. Substantially similar wording is acceptable.\"}, {\"criterion\": \"In the Gmail body, present all three missing items as a numbered list.\"}, {\"criterion\": \"In the Gmail body, ask that the materials be provided by 5:00 p.m. tomorrow and ask the recipient to flag immediately if any requested item may be delayed. Substantially similar wording is acceptable.\"}, {\"criterion\": \"In the Gmail body, give a practical fallback if a final document is not ready, such as sending the latest draft, interim version, or a download link first. Substantially similar wording is acceptable.\"}], \"tools\": [{\"mcp_appworld_simple_note__create_note\": {\"call\": [\"title\", \"content\"], \"result\": [\"response.note_id\"]}}, {\"mcp_appworld_simple_note__update_note\": {\"call\": [\"note_id\", \"content\"], \"result\": [\"response.message\"]}}, {\"mcp_appworld_gmail__send_email\": {\"call\": [\"email_addresses\", \"subject\", \"body\"], \"result\": [\"response.sent_email_thread_id\", \"response.sent_email_id\"]}}, {\"mcp_appworld_gmail__create_draft\": {\"call\": [\"recipient_email_addresses\", \"body\", \"belongs_to_email_thread_id\", \"response_to_email_id\"], \"result\": [\"response.draft_id\"]}}, {\"mcp_appworld_gmail__send_email_from_draft\": {\"call\": [\"draft_id\"], \"result\": [\"response.sent_email_thread_id\", \"response.sent_email_id\"]}}], \"tools_evaluation_path\": \"data/law_trainee/scripts/law_trainee_task_008_5.py\"}", "metadata_json": "{\"benchmark_role\": \"test\", \"difficulty\": \"medium\", \"preference_threads\": [\"formal_traceable_legal_comms\", \"deadline_and_followup_discipline\", \"cross_app_preparation_workflow\"]}", "yaml_path": "data/law_trainee/tasks/law_trainee_task_008/task.yaml", "yaml": "schema_version: v2\nuser_id: law_trainee\nenvironment_id: env_law_trainee\ntask_type: short_term_independent\ntrigger:\n  type: user\ntask_id: law_trainee_task_008\ntitle: \"Internal DD Deficiency Log to External Supplement Request\"\ndescription: \"Prepare one short internal deficiency log in Simple Note for the related-party loan review, then send one practical supplementation email in Gmail. Evaluation focuses on whether the note and email capture the three missing items, the matter context, and a clear deadline/fallback.\"\nintent:\n  initial_input: |-\n    My Simple Note email is stmcco@gmail.com and password is 'i#vhWEQ'. My Gmail email is stmcco@gmail.com and password is 'n3v=I[w'.\n\n    I still need to get the related-party loan review notes into circulation tonight, but a few DD gaps are holding this up. Please pull the outstanding points into one Simple Note entry for internal circulation and then turn that into a supplementation email to la-mcco@gmail.com.\n\n    I need something clean enough to use internally and clear enough to send out. If there is already a supplementation thread in Gmail, keep this with that thread. Please send the actual email in this pass rather than leaving me with a draft or asking me to approve wording first.\n\n  hidden_intent:\n    - content: \"Use the Simple Note title \\\"Internal DD Deficiency Log — Related-Party Loan Review\\\" and identify these three missing items: the board resolution for the related-party borrowing, the latest shareholder register / cap table, and the signed funds-flow explanation.\"\n    - content: \"In the Simple Note, briefly explain why each of those three items matters for the related-party loan review.\"\n    - content: \"In the Gmail subject and body, make clear that this is a supplementation request for the related-party loan review, present the three missing items as a numbered list, ask for the materials by 5:00 p.m. tomorrow, ask the recipient to flag any delay immediately, and give a practical fallback if a final document is not ready.\"\n    - content: \"If there is already a supplementation thread in Gmail, keep this request in that existing thread instead of starting a separate unrelated thread.\"\n    - content: \"Send the actual supplementation email in this pass rather than leaving only a draft or asking for wording approval first.\"\nobjectives:\n  tools:\n    - mcp_appworld_simple_note__create_note:\n        call: [title, content]\n        result: [response.note_id]\n    - mcp_appworld_simple_note__update_note:\n        call: [note_id, content]\n        result: [response.message]\n    - mcp_appworld_gmail__send_email:\n        call: [email_addresses, subject, body]\n        result: [response.sent_email_thread_id, response.sent_email_id]\n    - mcp_appworld_gmail__create_draft:\n        call: [recipient_email_addresses, body, belongs_to_email_thread_id, response_to_email_id]\n        result: [response.draft_id]\n    - mcp_appworld_gmail__send_email_from_draft:\n        call: [draft_id]\n        result: [response.sent_email_thread_id, response.sent_email_id]\n  checklist:\n    - criterion: \"In the internal Simple Note, use the title \\\"Internal DD Deficiency Log — Related-Party Loan Review.\\\"\"\n    - criterion: \"In the internal Simple Note, identify all three missing items: the board resolution for the related-party borrowing, the latest shareholder register / cap table, and the signed funds-flow explanation.\"\n    - criterion: \"In the internal Simple Note, briefly explain why the board resolution, shareholder register / cap table, and signed funds-flow explanation matter for the review. Substantially similar wording is acceptable.\"\n    - criterion: \"In the Gmail subject line, identify the matter as the related-party loan review and make clear that the email is a supplementation request. Substantially similar wording is acceptable.\"\n    - criterion: \"In the Gmail body, present all three missing items as a numbered list.\"\n    - criterion: \"In the Gmail body, ask that the materials be provided by 5:00 p.m. tomorrow and ask the recipient to flag immediately if any requested item may be delayed. Substantially similar wording is acceptable.\"\n    - criterion: \"In the Gmail body, give a practical fallback if a final document is not ready, such as sending the latest draft, interim version, or a download link first. Substantially similar wording is acceptable.\"\n  tools_evaluation_path: \"data/law_trainee/scripts/law_trainee_task_008_5.py\"\nmetadata:\n  difficulty: medium\n  benchmark_role: test\n  preference_threads:\n    - formal_traceable_legal_comms\n    - deadline_and_followup_discipline\n    - cross_app_preparation_workflow"}
{"schema_version": "v2", "role": "law_trainee", "task_id": "law_trainee_task_009", "user_id": "law_trainee", "environment_id": "env_law_trainee", "title": "Polishing a Loan-Dispute Complaint: Granular Prayers for Relief and Standard Closing", "display_title": "Polishing a Loan-Dispute Complaint: Granular Prayers for Relief and Standard Closing", "description": "Before lining up at the docketing office, the user asks the agent to lay out the complaint format for a private-lending case: spell out separate, independent prayers for relief (especially as against the guarantor) and adopt a standard closing at the end of the pleading.", "task_type": "long_term", "difficulty": "easy", "trigger_type": "user", "initial_input": "I have a private-lending dispute and am about to queue at the filing window of the Chaoyang District People’s Court in Beijing. Borrower Zhang San owes RMB 1 million in principal and will not repay; Li Si is the joint-and-several liability guarantor on this loan. Please quickly produce the complaint for this case.", "hidden_intents": ["After the body of the complaint, include the standard court-address closing: “Respectfully submitted to the Chaoyang District People’s Court of Beijing Municipality.”"], "hidden_intent_count": 1, "has_objectives": false, "objectives_json": "", "metadata_json": "{\"difficulty\": \"easy\", \"theme\": \"pleading_formatting_standard\"}", "yaml_path": "data/law_trainee/tasks/law_trainee_task_009/task.yaml", "yaml": "schema_version: v2\ntask_id: law_trainee_task_009\nuser_id: law_trainee\nenvironment_id: env_law_trainee\n\ntitle: \"Polishing a Loan-Dispute Complaint: Granular Prayers for Relief and Standard Closing\"\ndescription: \"Before lining up at the docketing office, the user asks the agent to lay out the complaint format for a private-lending case: spell out separate, independent prayers for relief (especially as against the guarantor) and adopt a standard closing at the end of the pleading.\"\n\ntask_type: long_term\n\ntrigger:\n  type: user\n\nintent:\n  initial_input: |\n    I have a private-lending dispute and am about to queue at the filing window of the Chaoyang District People’s Court in Beijing. Borrower Zhang San owes RMB 1 million in principal and will not repay; Li Si is the joint-and-several liability guarantor on this loan. Please quickly produce the complaint for this case.\n\n  hidden_intent:\n    - content: \"After the body of the complaint, include the standard court-address closing: “Respectfully submitted to the Chaoyang District People’s Court of Beijing Municipality.”\"\nmetadata:\n  difficulty: easy\n  theme: pleading_formatting_standard"}
{"schema_version": "v2", "role": "law_trainee", "task_id": "law_trainee_task_010", "user_id": "law_trainee", "environment_id": "env_law_trainee", "title": "Establish the Legal-Analysis Baseline: Preference for Subjective–Objective Decomposition", "display_title": "Establish the Legal-Analysis Baseline: Preference for Subjective–Objective Decomposition", "description": "By walking through a concrete mistaken-taking-of-a-cellphone scenario, the user signals to the Agent a preference for subjective–objective separation in analysis, embedding a standing requirement to hard-split objective conduct and subjective intent in any later case analysis.", "task_type": "long_term", "difficulty": "easy", "trigger_type": "user", "initial_input": "I have a case where someone mistakenly took another person’s phone. I need a full analytical report that unpacks it cleanly. My own draft is messy: sometimes I only describe him taking the phone, sometimes only that he misidentified it.\n\nThe facts in outline are as follows: Late at night, Zhang San ate with friends at a busy hot-pot restaurant; he drank two beers but remained clear-headed. Before leaving he went to the shared power-bank pickup counter near the door to get his phone—the counter is narrow, the light is dim, several phones sit side by side on charging docks, and some use the same dark silicone case. Zhang San’s own phone is a flagship of a certain brand, and he also used a dark case that day; neighboring diner Li Si’s phone is the same brand, colorway, and storage tier, and the two looked almost identical. When Zhang San took a phone, staff had just tidied the counter and had pushed his phone and Li Si’s inward; the two sat flush side by side. Zhang San glanced at the backs and cases without lighting the screens, picked up one phone, put it in his pocket, and left. Li Si was at the condiment station; the phone was left unattended on the counter.\n\nAbout forty minutes after leaving, in a taxi, Zhang San tried to reply to messages: face unlock failed twice, a third try with the passcode still failed, and he realized the lock-screen wallpaper was not his. He opened the WeChat login page, confirmed it was not his account, had the driver turn back, and called the restaurant en route to explain. The restaurant said Li Si had noticed the missing phone, reviewed CCTV, and had already reported to the police. Zhang San returned about one hour and fifty minutes later and handed the phone back in person; the exterior was undamaged, but Li Si claimed several unread client messages on a business WeChat account and questioned whether the phone had been accessed in the meantime; Zhang San denied ever unlocking it successfully. CCTV shows Zhang San’s motion when taking the phone was natural, without looking around furtively, and he did not run out the door; when he came back he voluntarily gave the phone to the store manager.", "hidden_intents": ["When analyzing legal cases, the argument must be hard-split into two independent dimensions, the objective aspect and the subjective aspect, for analysis."], "hidden_intent_count": 1, "has_objectives": false, "objectives_json": "", "metadata_json": "{\"difficulty\": \"easy\"}", "yaml_path": "data/law_trainee/tasks/law_trainee_task_010/task.yaml", "yaml": "schema_version: v2\ntask_id: law_trainee_task_010\nuser_id: law_trainee\nenvironment_id: env_law_trainee\n\ntitle: \"Establish the Legal-Analysis Baseline: Preference for Subjective–Objective Decomposition\"\ndescription: \"By walking through a concrete mistaken-taking-of-a-cellphone scenario, the user signals to the Agent a preference for subjective–objective separation in analysis, embedding a standing requirement to hard-split objective conduct and subjective intent in any later case analysis.\"\n\ntask_type: long_term\n\ntrigger:\n  type: user\n\nintent:\n  initial_input: |\n    I have a case where someone mistakenly took another person’s phone. I need a full analytical report that unpacks it cleanly. My own draft is messy: sometimes I only describe him taking the phone, sometimes only that he misidentified it.\n\n    The facts in outline are as follows: Late at night, Zhang San ate with friends at a busy hot-pot restaurant; he drank two beers but remained clear-headed. Before leaving he went to the shared power-bank pickup counter near the door to get his phone—the counter is narrow, the light is dim, several phones sit side by side on charging docks, and some use the same dark silicone case. Zhang San’s own phone is a flagship of a certain brand, and he also used a dark case that day; neighboring diner Li Si’s phone is the same brand, colorway, and storage tier, and the two looked almost identical. When Zhang San took a phone, staff had just tidied the counter and had pushed his phone and Li Si’s inward; the two sat flush side by side. Zhang San glanced at the backs and cases without lighting the screens, picked up one phone, put it in his pocket, and left. Li Si was at the condiment station; the phone was left unattended on the counter.\n\n    About forty minutes after leaving, in a taxi, Zhang San tried to reply to messages: face unlock failed twice, a third try with the passcode still failed, and he realized the lock-screen wallpaper was not his. He opened the WeChat login page, confirmed it was not his account, had the driver turn back, and called the restaurant en route to explain. The restaurant said Li Si had noticed the missing phone, reviewed CCTV, and had already reported to the police. Zhang San returned about one hour and fifty minutes later and handed the phone back in person; the exterior was undamaged, but Li Si claimed several unread client messages on a business WeChat account and questioned whether the phone had been accessed in the meantime; Zhang San denied ever unlocking it successfully. CCTV shows Zhang San’s motion when taking the phone was natural, without looking around furtively, and he did not run out the door; when he came back he voluntarily gave the phone to the store manager.\n\n  hidden_intent:\n    - content: \"When analyzing legal cases, the argument must be hard-split into two independent dimensions, the objective aspect and the subjective aspect, for analysis.\"\nmetadata:\n  difficulty: easy"}
{"schema_version": "v2", "role": "law_trainee", "task_id": "law_trainee_task_011", "user_id": "law_trainee", "environment_id": "env_law_trainee", "title": "Private Lending: Draft a Loan Agreement from a Case Summary", "display_title": "Private Lending: Draft a Loan Agreement from a Case Summary", "description": "Based on the oral terms and identity details in Loans.md, the intern drafts a well-structured private-lending loan agreement under mainland Chinese drafting practice. Evaluation focuses on party details, consolidated principal, term and interest wording, purpose and late-payment terms, dispute resolution, acceleration, effectiveness, marital joint debt language, and alignment with the fact pattern.", "task_type": "long_term", "difficulty": "medium", "trigger_type": "user", "initial_input": "I am interning at a firm. My supervising attorney asked me to turn a client’s oral account and notes into a first draft private-lending loan agreement. The facts are in Loans.md in this directory. Please draft it using a typical mainland Chinese private-lending agreement layout.Output the drafted contract directly; no need to create a file.", "hidden_intents": ["State clearly the lender’s and borrower’s identifying information.", "State clearly the loan principal.", "State clearly the loan term.", "State clearly the interest rate.", "State clearly the purpose of the loan.", "State clearly the repayment method.", "State clearly terms for prepayment.", "State clearly a dispute-resolution clause.", "Include a marital joint debt confirmation clause.", "Include an acceleration-of-maturity clause.", "Please generate an analysis of the borrower’s repayment capacity.", "Please calculate at the end how much legal service fee should be charged if the attorney’s fee is computed at 0.1% of the loan amount."], "hidden_intent_count": 12, "has_objectives": true, "objectives_json": "{\"checklist\": [{\"criterion\": \"The agreement identifies Party A (lender) as Zhang Hua, with ID number 110101199001018888 and a contact address including № 8 Yard Yangguang Avenue, Chaoyang District, Beijing. Substantially similar wording is acceptable.\"}, {\"criterion\": \"The agreement identifies Party B (borrower) as Li Qiang, with ID number 310101199202026666 and a contact address including № 100 Qiancheng Road, Pudong New Area, Shanghai. Substantially similar wording is acceptable.\"}, {\"criterion\": \"The principal is consolidated as RMB 150,000 (one hundred fifty thousand yuan). Substantially similar wording is acceptable.\"}, {\"criterion\": \"The term runs from April 10, 2026 through October 9, 2027, reflecting eighteen months in total. Substantially similar wording is acceptable.\"}, {\"criterion\": \"The rate is stated as 1% per month and 12% per year; interest accrues from actual disbursement and is settled monthly. Substantially similar wording is acceptable.\"}, {\"criterion\": \"If interest is not paid on time, overdue amounts are subject to the penalty or liquidated-damages rule set out in the agreement’s late-payment clause, consistent with Article 5–style sequencing. Substantially similar wording is acceptable.\"}, {\"criterion\": \"The purpose covers café equipment purchase and green-bean or other raw-material procurement, tied to the shop name “Time Imprint” or equivalent; funds may not be diverted or used for unlawful activity. Substantially similar wording is acceptable.\"}, {\"criterion\": \"Repayment is a lump-sum payment of principal and remaining interest at maturity, or an equivalent structure. Substantially similar wording is acceptable.\"}, {\"criterion\": \"Late repayment carries, in addition to contractual interest, liquidated damages at 0.05% per day on the overdue amount (equivalent to five renminbi per RMB ten thousand per day). Substantially similar wording is acceptable.\"}, {\"criterion\": \"Party B may prepay; interest is calculated on actual days outstanding without extra prepayment penalties or fees. Substantially similar wording is acceptable.\"}, {\"criterion\": \"Dispute resolution provides that if negotiation fails, suit may be brought in the People’s Court of Chaoyang District, Beijing (or an equivalent reference to that court). Substantially similar wording is acceptable.\"}, {\"criterion\": \"The agreement includes an acceleration clause: if Party B’s business deteriorates, transfers assets, or maliciously evades debt (or similar material adverse events), Party A may demand immediate repayment of the full principal and accrued interest and may demand liquidated damages under the contract. Substantially similar wording is acceptable.\"}, {\"criterion\": \"The agreement includes a marital joint debt confirmation: the borrower’s spouse confirms knowledge and consent to this loan and signs as co-borrower at the end; Wang Mei bears joint repayment obligations with the borrower; both parties agree without dispute that the debt is marital joint debt under applicable law. Substantially similar wording is acceptable.\"}, {\"criterion\": \"The agreement is made in duplicate with one copy for each party, each equally authentic. Substantially similar wording is acceptable.\"}, {\"criterion\": \"The agreement becomes effective when the lender pays the loan into the borrower’s designated account (or equivalent). Substantially similar wording is acceptable.\"}, {\"criterion\": \"Signature blocks are left for both parties and for an execution date of April 10, 2026, consistent with same-day disbursement in the facts. Substantially similar wording is acceptable.\"}, {\"criterion\": \"State that the legal service fee is RMB 150.\"}, {\"criterion\": \"Li Qiang has strong performance capability and debt repayment capacity.\"}]}", "metadata_json": "{\"difficulty\": \"medium\"}", "yaml_path": "data/law_trainee/tasks/law_trainee_task_011/task.yaml", "yaml": "schema_version: v2\ntask_id: law_trainee_task_011\nuser_id: law_trainee\nenvironment_id: env_law_trainee\n\ntitle: \"Private Lending: Draft a Loan Agreement from a Case Summary\"\ndescription: \"Based on the oral terms and identity details in Loans.md, the intern drafts a well-structured private-lending loan agreement under mainland Chinese drafting practice. Evaluation focuses on party details, consolidated principal, term and interest wording, purpose and late-payment terms, dispute resolution, acceleration, effectiveness, marital joint debt language, and alignment with the fact pattern.\"\n\ntask_type: long_term\n\ntrigger:\n  type: user\n\nintent:\n  initial_input: |\n    I am interning at a firm. My supervising attorney asked me to turn a client’s oral account and notes into a first draft private-lending loan agreement. The facts are in Loans.md in this directory. Please draft it using a typical mainland Chinese private-lending agreement layout.Output the drafted contract directly; no need to create a file.\n  hidden_intent:\n    - content: \"State clearly the lender’s and borrower’s identifying information.\"\n    - content: \"State clearly the loan principal.\"\n    - content: \"State clearly the loan term.\"\n    - content: \"State clearly the interest rate.\"\n    - content: \"State clearly the purpose of the loan.\"\n    - content: \"State clearly the repayment method.\"\n    - content: \"State clearly terms for prepayment.\"\n    - content: \"State clearly a dispute-resolution clause.\"\n    - content: \"Include a marital joint debt confirmation clause.\"\n    - content: \"Include an acceleration-of-maturity clause.\"\n    - content: \"Please generate an analysis of the borrower’s repayment capacity.\"   \n    - content: \"Please calculate at the end how much legal service fee should be charged if the attorney’s fee is computed at 0.1% of the loan amount.\"\nobjectives:\n  checklist:\n    - criterion: \"The agreement identifies Party A (lender) as Zhang Hua, with ID number 110101199001018888 and a contact address including № 8 Yard Yangguang Avenue, Chaoyang District, Beijing. Substantially similar wording is acceptable.\"\n\n    - criterion: \"The agreement identifies Party B (borrower) as Li Qiang, with ID number 310101199202026666 and a contact address including № 100 Qiancheng Road, Pudong New Area, Shanghai. Substantially similar wording is acceptable.\"\n\n    - criterion: \"The principal is consolidated as RMB 150,000 (one hundred fifty thousand yuan). Substantially similar wording is acceptable.\"\n\n    - criterion: \"The term runs from April 10, 2026 through October 9, 2027, reflecting eighteen months in total. Substantially similar wording is acceptable.\"\n\n    - criterion: \"The rate is stated as 1% per month and 12% per year; interest accrues from actual disbursement and is settled monthly. Substantially similar wording is acceptable.\"\n\n    - criterion: \"If interest is not paid on time, overdue amounts are subject to the penalty or liquidated-damages rule set out in the agreement’s late-payment clause, consistent with Article 5–style sequencing. Substantially similar wording is acceptable.\"\n\n    - criterion: \"The purpose covers café equipment purchase and green-bean or other raw-material procurement, tied to the shop name “Time Imprint” or equivalent; funds may not be diverted or used for unlawful activity. Substantially similar wording is acceptable.\"\n\n    - criterion: \"Repayment is a lump-sum payment of principal and remaining interest at maturity, or an equivalent structure. Substantially similar wording is acceptable.\"\n\n    - criterion: \"Late repayment carries, in addition to contractual interest, liquidated damages at 0.05% per day on the overdue amount (equivalent to five renminbi per RMB ten thousand per day). Substantially similar wording is acceptable.\"\n\n    - criterion: \"Party B may prepay; interest is calculated on actual days outstanding without extra prepayment penalties or fees. Substantially similar wording is acceptable.\"\n\n    - criterion: \"Dispute resolution provides that if negotiation fails, suit may be brought in the People’s Court of Chaoyang District, Beijing (or an equivalent reference to that court). Substantially similar wording is acceptable.\"\n\n    - criterion: \"The agreement includes an acceleration clause: if Party B’s business deteriorates, transfers assets, or maliciously evades debt (or similar material adverse events), Party A may demand immediate repayment of the full principal and accrued interest and may demand liquidated damages under the contract. Substantially similar wording is acceptable.\"\n\n    - criterion: \"The agreement includes a marital joint debt confirmation: the borrower’s spouse confirms knowledge and consent to this loan and signs as co-borrower at the end; Wang Mei bears joint repayment obligations with the borrower; both parties agree without dispute that the debt is marital joint debt under applicable law. Substantially similar wording is acceptable.\"\n\n    - criterion: \"The agreement is made in duplicate with one copy for each party, each equally authentic. Substantially similar wording is acceptable.\"\n\n    - criterion: \"The agreement becomes effective when the lender pays the loan into the borrower’s designated account (or equivalent). Substantially similar wording is acceptable.\"\n\n    - criterion: \"Signature blocks are left for both parties and for an execution date of April 10, 2026, consistent with same-day disbursement in the facts. Substantially similar wording is acceptable.\"\n\n    - criterion: \"State that the legal service fee is RMB 150.\"\n  \n    - criterion: \"Li Qiang has strong performance capability and debt repayment capacity.\"\nmetadata:\n  difficulty: medium"}
{"schema_version": "v2", "role": "law_trainee", "task_id": "law_trainee_task_012", "user_id": "law_trainee", "environment_id": "env_law_trainee", "title": "Internal Review-Queue Follow-up Email with Thread Triage", "display_title": "Internal Review-Queue Follow-up Email with Thread Triage", "description": "Prepare and send one short internal follow-up email for the Xinhe supervisor review queue, run the same-thread Gmail triage sequence, and create then clear one temporary Todoist reminder. Evaluation focuses on whether the email makes the two live items, deadline, blocker handling, and review-tracking steps easy to follow.", "task_type": "short_term_independent", "difficulty": "medium", "trigger_type": "user", "initial_input": "My Gmail email is stmcco@gmail.com and password is 'n3v=I[w'.\nMy Todoist email is stmcco@gmail.com and password is 'Ozz21Q4'.\n\nI'm about to roll into another call, so can you push a quick internal follow-up to la-mcco@gmail.com on the Xinhe Supply Chain Arbitration supervisor review queue?\n\nUse the subject \"Internal Follow-up | Xinhe Supply Chain Arbitration | Supervisor Review Queue\".\n\nAfter sending, do the usual same-thread review-tracking pass with the \"needs-supervisor-review\" label and the unread/read toggles, then create one temporary Todoist reminder for the next Xinhe follow-up checkpoint and clear that reminder in the same pass.", "hidden_intents": ["In the email body, clearly say that the two live items on the Xinhe supervisor review queue are the arbitration-clause scan status and the courier receipt status, and present them in a separated two-item structure.", "In the email body, ask for a reply by 4:00 p.m. tomorrow and ask the recipient to flag blockers or timing risk immediately if the deadline may slip.", "In the email body, add a fallback that if the courier receipt is unavailable, they should send a waybill or tracking screenshot plus the expected arrival timing first. In Todoist, make the temporary checkpoint a Xinhe follow-up checkpoint at 3:30 p.m. tomorrow.", "Use the subject \"Internal Follow-up | Xinhe Supply Chain Arbitration | Supervisor Review Queue\" exactly.", "After sending, complete the usual same-thread review-tracking pass with the \"needs-supervisor-review\" label and the unread/read toggles."], "hidden_intent_count": 5, "has_objectives": true, "objectives_json": "{\"checklist\": [{\"criterion\": \"The final email subject is exactly \\\"Internal Follow-up | Xinhe Supply Chain Arbitration | Supervisor Review Queue\\\".\"}, {\"criterion\": \"The final email body clearly states that the two live items on the Xinhe supervisor review queue are the arbitration-clause scan status and the courier receipt status.\"}, {\"criterion\": \"The final email body asks for a reply by 4:00 p.m. tomorrow and asks the recipient to flag blockers or timing risk immediately if the deadline may slip. Substantially similar wording is acceptable.\"}, {\"criterion\": \"The final email body includes a fallback: if the courier receipt is unavailable, send a waybill or tracking screenshot plus expected arrival timing first. Substantially similar wording is acceptable.\"}, {\"criterion\": \"The final email body presents the two pending items in a scan-friendly separated structure, such as numbered items or clearly separated equivalent formatting.\"}, {\"criterion\": \"The Gmail triage sequence is completed on the sent thread using the label plus unread/read toggles.\"}, {\"criterion\": \"The temporary Todoist reminder clearly identifies the Xinhe follow-up checkpoint and is removed by the end of the task.\"}], \"tools\": [{\"mcp_appworld_gmail__send_email\": {\"call\": [\"email_addresses\", \"subject\", \"body\"], \"result\": [\"response.sent_email_thread_id\", \"response.sent_email_id\"]}}, {\"mcp_appworld_gmail__create_draft\": {\"call\": [\"recipient_email_addresses\", \"body\", \"belongs_to_email_thread_id\", \"response_to_email_id\"], \"result\": [\"response.draft_id\"]}}, {\"mcp_appworld_gmail__send_email_from_draft\": {\"call\": [\"draft_id\"], \"result\": [\"response.sent_email_thread_id\", \"response.sent_email_id\"]}}, {\"mcp_appworld_gmail__label_thread\": {\"call\": [\"label\", \"email_thread_id\"], \"result\": [\"response.message\"]}}, {\"mcp_appworld_gmail__mark_thread_unread\": {\"call\": [\"email_thread_id\"], \"result\": [\"response.message\"]}}, {\"mcp_appworld_gmail__mark_thread_read\": {\"call\": [\"email_thread_id\"], \"result\": [\"response.message\"]}}, {\"mcp_appworld_todoist__create_task\": {\"call\": [\"title\", \"description\"], \"result\": [\"response.task_id\"]}}, {\"mcp_appworld_todoist__delete_task\": {\"call\": [\"task_id\"], \"result\": [\"response.message\"]}}], \"tools_evaluation_path\": \"data/law_trainee/scripts/law_trainee_task_012_5.py\"}", "metadata_json": "{\"benchmark_role\": \"test\", \"difficulty\": \"medium\", \"preference_threads\": [\"formal_traceable_legal_comms\", \"supervisor_review_workflow\", \"cross_app_preparation_workflow\"]}", "yaml_path": "data/law_trainee/tasks/law_trainee_task_012/task.yaml", "yaml": "schema_version: v2\nuser_id: law_trainee\nenvironment_id: env_law_trainee\ntask_type: short_term_independent\ntrigger:\n  type: user\ntask_id: law_trainee_task_012\ntitle: Internal Review-Queue Follow-up Email with Thread Triage\ndescription: \"Prepare and send one short internal follow-up email for the Xinhe supervisor review queue, run the same-thread Gmail triage sequence, and create then clear one temporary Todoist reminder. Evaluation focuses on whether the email makes the two live items, deadline, blocker handling, and review-tracking steps easy to follow.\"\nintent:\n  initial_input: |-\n    My Gmail email is stmcco@gmail.com and password is 'n3v=I[w'.\n    My Todoist email is stmcco@gmail.com and password is 'Ozz21Q4'.\n\n    I'm about to roll into another call, so can you push a quick internal follow-up to la-mcco@gmail.com on the Xinhe Supply Chain Arbitration supervisor review queue?\n\n    Use the subject \"Internal Follow-up | Xinhe Supply Chain Arbitration | Supervisor Review Queue\".\n\n    After sending, do the usual same-thread review-tracking pass with the \"needs-supervisor-review\" label and the unread/read toggles, then create one temporary Todoist reminder for the next Xinhe follow-up checkpoint and clear that reminder in the same pass.\n\n  hidden_intent:\n    - content: \"In the email body, clearly say that the two live items on the Xinhe supervisor review queue are the arbitration-clause scan status and the courier receipt status, and present them in a separated two-item structure.\"\n    - content: \"In the email body, ask for a reply by 4:00 p.m. tomorrow and ask the recipient to flag blockers or timing risk immediately if the deadline may slip.\"\n    - content: \"In the email body, add a fallback that if the courier receipt is unavailable, they should send a waybill or tracking screenshot plus the expected arrival timing first. In Todoist, make the temporary checkpoint a Xinhe follow-up checkpoint at 3:30 p.m. tomorrow.\"\n    - content: \"Use the subject \\\"Internal Follow-up | Xinhe Supply Chain Arbitration | Supervisor Review Queue\\\" exactly.\"\n    - content: \"After sending, complete the usual same-thread review-tracking pass with the \\\"needs-supervisor-review\\\" label and the unread/read toggles.\"\nobjectives:\n  tools:\n    - mcp_appworld_gmail__send_email:\n        call: [email_addresses, subject, body]\n        result: [response.sent_email_thread_id, response.sent_email_id]\n    - mcp_appworld_gmail__create_draft:\n        call: [recipient_email_addresses, body, belongs_to_email_thread_id, response_to_email_id]\n        result: [response.draft_id]\n    - mcp_appworld_gmail__send_email_from_draft:\n        call: [draft_id]\n        result: [response.sent_email_thread_id, response.sent_email_id]\n    - mcp_appworld_gmail__label_thread:\n        call: [label, email_thread_id]\n        result: [response.message]\n    - mcp_appworld_gmail__mark_thread_unread:\n        call: [email_thread_id]\n        result: [response.message]\n    - mcp_appworld_gmail__mark_thread_read:\n        call: [email_thread_id]\n        result: [response.message]\n    - mcp_appworld_todoist__create_task:\n        call: [title, description]\n        result: [response.task_id]\n    - mcp_appworld_todoist__delete_task:\n        call: [task_id]\n        result: [response.message]\n  checklist:\n    - criterion: \"The final email subject is exactly \\\"Internal Follow-up | Xinhe Supply Chain Arbitration | Supervisor Review Queue\\\".\"\n    - criterion: \"The final email body clearly states that the two live items on the Xinhe supervisor review queue are the arbitration-clause scan status and the courier receipt status.\"\n    - criterion: \"The final email body asks for a reply by 4:00 p.m. tomorrow and asks the recipient to flag blockers or timing risk immediately if the deadline may slip. Substantially similar wording is acceptable.\"\n    - criterion: \"The final email body includes a fallback: if the courier receipt is unavailable, send a waybill or tracking screenshot plus expected arrival timing first. Substantially similar wording is acceptable.\"\n    - criterion: \"The final email body presents the two pending items in a scan-friendly separated structure, such as numbered items or clearly separated equivalent formatting.\"\n    - criterion: \"The Gmail triage sequence is completed on the sent thread using the label plus unread/read toggles.\"\n    - criterion: \"The temporary Todoist reminder clearly identifies the Xinhe follow-up checkpoint and is removed by the end of the task.\"\n  tools_evaluation_path: \"data/law_trainee/scripts/law_trainee_task_012_5.py\"\nmetadata:\n  difficulty: medium\n  benchmark_role: test\n  preference_threads:\n    - formal_traceable_legal_comms\n    - supervisor_review_workflow\n    - cross_app_preparation_workflow"}
{"schema_version": "v2", "role": "law_trainee", "task_id": "law_trainee_task_013", "user_id": "law_trainee", "environment_id": "env_law_trainee", "title": "Draft a Civil Complaint for Company Jia (Rent and Breach of Lease Contract)", "display_title": "Draft a Civil Complaint for Company Jia (Rent and Breach of Lease Contract)", "description": "Based on the facts stated in Complaint_Facts.md, draft a first-pass civil complaint in the name of Company Jia for filing with a people’s court in mainland China. Evaluation focuses on correct party caption and prayers for relief, chronology and key facts, main bases of claim, and consistency with the case-summary attachment.", "task_type": "long_term", "difficulty": "medium", "trigger_type": "user", "initial_input": "Our firm recently took on a civil matter. The underlying facts are in Complaint_Facts.md. Company Jia intends to sue in a people’s court in mainland China; please draft the complaint according to standard complaint format from those facts. Fields for which you have no specific information may be left blank.", "hidden_intents": ["At the opening of the pleading, state the case participants (caption).", "In the “Prayers for Relief” section, enumerate the specific requests in separate numbered items.", "In the “Facts and Legal Grounds” section, set out the relevant factual basis clearly.", "In the “Facts and Legal Grounds” section, immediately after each factual statement, give the related legal basis.", "At the end of the pleading, state clearly the court to which the complaint is submitted."], "hidden_intent_count": 5, "has_objectives": true, "objectives_json": "{\"checklist\": [{\"criterion\": \"In the caption at the opening of the pleading, state that the plaintiff is Company Jia.\"}, {\"criterion\": \"In the caption at the opening of the pleading, state that the defendants are Company Yi, Company Ding, and Company Bing.\"}, {\"criterion\": \"In “Prayers for Relief,” request that the court order Company Yi to pay economic damages to Company Jia. Substantially similar wording is acceptable.\"}, {\"criterion\": \"In “Prayers for Relief,” request that the court order Company Bing to bear joint-and-several guarantor liability for Company Yi’s breach. Substantially similar wording is acceptable.\"}, {\"criterion\": \"In “Prayers for Relief,” request that the court order Company Ding to return the property in kind. Substantially similar wording is acceptable.\"}, {\"criterion\": \"In “Facts and Legal Grounds,” state that the finance lease contract between Company Jia and Company Yi was validly formed and is effective. Substantially similar wording is acceptable.\"}, {\"criterion\": \"State that the legal basis for “the finance lease contract between Company Jia and Company Yi was validly formed and is effective” is paragraph 2 of Article 2 of the Interpretation of the Supreme People’s Court on Several Issues concerning the Application of Law in the Trial of Cases Involving Disputes over Financial Lease Contracts.\"}, {\"criterion\": \"In “Facts and Legal Grounds,” state that Company Yi’s conduct constitutes a fundamental breach rendering performance of the contract purpose impossible, and that Company Jia therefore enjoys a statutory right of termination. Substantially similar wording is acceptable.\"}, {\"criterion\": \"State that the legal basis for “Company Yi’s conduct constitutes a fundamental breach, rendering performance of the contract purpose impossible, and Company Jia enjoys a statutory right of termination” is paragraphs 3 and 4 of Article 563 of the Civil Code of the People’s Republic of China and paragraphs 1 and 2 of Article 5 of the Interpretation of the Supreme People’s Court on Several Issues concerning the Application of Law in the Trial of Cases Involving Disputes over Financial Lease Contracts.\"}, {\"criterion\": \"In “Facts and Legal Grounds,” state that Company Yi must return double the earnest-money deposit to Company Jia: during formation of the finance lease, Company Jia paid RMB 500,000 as earnest money; due to Company Yi’s late performance and unauthorized transfer of the leased property in breach of contract, constituting a fundamental breach leading to rescission, Company Yi must return double the deposit totaling RMB 1 million to Company Jia. Substantially similar wording is acceptable.\"}, {\"criterion\": \"State that the legal basis for the proposition that “Company Yi must return double the earnest-money deposit to Company Jia” is Article 587 of the Civil Code of the People’s Republic of China.\"}, {\"criterion\": \"In “Facts and Legal Grounds,” state that Company Bing bears joint-and-several guarantor liability for Company Yi’s breach: Company Bing expressly assumed joint-and-several guarantor liability in the finance lease contract and signed to confirm; a joint-and-several suretyship should be deemed established, and Company Bing is jointly and severally liable for the aggregate value of the obligation to return the property in kind, default interest, and compensation for economic loss. Substantially similar wording is acceptable.\"}, {\"criterion\": \"State that the legal basis for the proposition that “Company Bing bears joint-and-several guarantor liability for Company Yi’s breach” is Article 688 of the Civil Code of the People’s Republic of China.\"}, {\"criterion\": \"In “Facts and Legal Grounds,” state that Company Ding must return the property in kind: during the finance-lease term Company Jia was the owner of the machine tools; Company Yi’s sale to Company Ding was a disposition without authority; netting the price of a concurrent batch of nuts, Company Ding acquired three machine tools for only RMB 4 million while their market value was RMB 7.5 million, so Company Ding is not a bona fide third-party purchaser; even though the three machine tools were delivered, Ding did not acquire ownership and must therefore return the property. Substantially similar wording is acceptable.\"}, {\"criterion\": \"At the end of the pleading, state clearly that the complaint is submitted to the Shuizhou District People’s Court of Jiangzhou City.\"}]}", "metadata_json": "{\"difficulty\": \"medium\"}", "yaml_path": "data/law_trainee/tasks/law_trainee_task_013/task.yaml", "yaml": "schema_version: v2\ntask_id: law_trainee_task_013\nuser_id: law_trainee\nenvironment_id: env_law_trainee\n\ntitle: \"Draft a Civil Complaint for Company Jia (Rent and Breach of Lease Contract)\"\ndescription: \"Based on the facts stated in Complaint_Facts.md, draft a first-pass civil complaint in the name of Company Jia for filing with a people’s court in mainland China. Evaluation focuses on correct party caption and prayers for relief, chronology and key facts, main bases of claim, and consistency with the case-summary attachment.\"\n\ntask_type: long_term\n\ntrigger:\n  type: user\n\nintent:\n  initial_input: |\n    Our firm recently took on a civil matter. The underlying facts are in Complaint_Facts.md. Company Jia intends to sue in a people’s court in mainland China; please draft the complaint according to standard complaint format from those facts. Fields for which you have no specific information may be left blank.\n\n  hidden_intent:\n    - content: \"At the opening of the pleading, state the case participants (caption).\"\n    - content: \"In the “Prayers for Relief” section, enumerate the specific requests in separate numbered items.\"\n    - content: \"In the “Facts and Legal Grounds” section, set out the relevant factual basis clearly.\"\n    - content: \"In the “Facts and Legal Grounds” section, immediately after each factual statement, give the related legal basis.\"\n    - content: \"At the end of the pleading, state clearly the court to which the complaint is submitted.\"\nobjectives:\n  checklist:\n    - criterion: \"In the caption at the opening of the pleading, state that the plaintiff is Company Jia.\"\n\n    - criterion: \"In the caption at the opening of the pleading, state that the defendants are Company Yi, Company Ding, and Company Bing.\"\n\n    - criterion: \"In “Prayers for Relief,” request that the court order Company Yi to pay economic damages to Company Jia. Substantially similar wording is acceptable.\"\n\n    - criterion: \"In “Prayers for Relief,” request that the court order Company Bing to bear joint-and-several guarantor liability for Company Yi’s breach. Substantially similar wording is acceptable.\"\n\n    - criterion: \"In “Prayers for Relief,” request that the court order Company Ding to return the property in kind. Substantially similar wording is acceptable.\"\n\n    - criterion: \"In “Facts and Legal Grounds,” state that the finance lease contract between Company Jia and Company Yi was validly formed and is effective. Substantially similar wording is acceptable.\"\n\n    - criterion: \"State that the legal basis for “the finance lease contract between Company Jia and Company Yi was validly formed and is effective” is paragraph 2 of Article 2 of the Interpretation of the Supreme People’s Court on Several Issues concerning the Application of Law in the Trial of Cases Involving Disputes over Financial Lease Contracts.\"\n\n    - criterion: \"In “Facts and Legal Grounds,” state that Company Yi’s conduct constitutes a fundamental breach rendering performance of the contract purpose impossible, and that Company Jia therefore enjoys a statutory right of termination. Substantially similar wording is acceptable.\"\n\n    - criterion: \"State that the legal basis for “Company Yi’s conduct constitutes a fundamental breach, rendering performance of the contract purpose impossible, and Company Jia enjoys a statutory right of termination” is paragraphs 3 and 4 of Article 563 of the Civil Code of the People’s Republic of China and paragraphs 1 and 2 of Article 5 of the Interpretation of the Supreme People’s Court on Several Issues concerning the Application of Law in the Trial of Cases Involving Disputes over Financial Lease Contracts.\"\n\n    - criterion: \"In “Facts and Legal Grounds,” state that Company Yi must return double the earnest-money deposit to Company Jia: during formation of the finance lease, Company Jia paid RMB 500,000 as earnest money; due to Company Yi’s late performance and unauthorized transfer of the leased property in breach of contract, constituting a fundamental breach leading to rescission, Company Yi must return double the deposit totaling RMB 1 million to Company Jia. Substantially similar wording is acceptable.\"\n\n    - criterion: \"State that the legal basis for the proposition that “Company Yi must return double the earnest-money deposit to Company Jia” is Article 587 of the Civil Code of the People’s Republic of China.\"\n\n    - criterion: \"In “Facts and Legal Grounds,” state that Company Bing bears joint-and-several guarantor liability for Company Yi’s breach: Company Bing expressly assumed joint-and-several guarantor liability in the finance lease contract and signed to confirm; a joint-and-several suretyship should be deemed established, and Company Bing is jointly and severally liable for the aggregate value of the obligation to return the property in kind, default interest, and compensation for economic loss. Substantially similar wording is acceptable.\"\n\n    - criterion: \"State that the legal basis for the proposition that “Company Bing bears joint-and-several guarantor liability for Company Yi’s breach” is Article 688 of the Civil Code of the People’s Republic of China.\"\n\n    - criterion: \"In “Facts and Legal Grounds,” state that Company Ding must return the property in kind: during the finance-lease term Company Jia was the owner of the machine tools; Company Yi’s sale to Company Ding was a disposition without authority; netting the price of a concurrent batch of nuts, Company Ding acquired three machine tools for only RMB 4 million while their market value was RMB 7.5 million, so Company Ding is not a bona fide third-party purchaser; even though the three machine tools were delivered, Ding did not acquire ownership and must therefore return the property. Substantially similar wording is acceptable.\"\n\n    - criterion: \"At the end of the pleading, state clearly that the complaint is submitted to the Shuizhou District People’s Court of Jiangzhou City.\"\nmetadata:\n  difficulty: medium"}
{"schema_version": "v2", "role": "law_trainee", "task_id": "law_trainee_task_014", "user_id": "law_trainee", "environment_id": "env_law_trainee", "title": "Partner Hearing-Prep Handover Note to Temporary Action Board", "display_title": "Partner Hearing-Prep Handover Note to Temporary Action Board", "description": "Prepare one short partner-review handover note in Simple Note for the Yuehai guarantee dispute, then mirror the three live follow-up items in a temporary Todoist board and remove that board in the same pass. Evaluation focuses on whether the note and work items stay short, matter-specific, and easy to execute.", "task_type": "short_term_independent", "difficulty": "medium", "trigger_type": "user", "initial_input": "My Simple Note email is stmcco@gmail.com and password is 'i#vhWEQ'.\nMy Todoist email is stmcco@gmail.com and password is 'Ozz21Q4'.\n\nI'm still tied up in briefings, and the Yuehai guarantee dispute file needs to go up for partner review soon. My notes are not really in circulation shape yet.\n\nPlease clean them into a short handover note in Simple Note under \"Partner Handover — Yuehai Guarantee Dispute Hearing Prep\", then spin the immediate next-step items into a temporary Todoist board called \"Temporary Hearing Prep Board - Yuehai\" so I can review the workplan quickly.\n\nFor this rehearsal pass, create exactly three temporary work items for the live blockers, then remove the temporary board before you reply so there is no leftover clutter.", "hidden_intents": ["In the Simple Note, state that the hearing date is 18 April 2026 and that the Yuehai file is being prepared for partner review before that hearing.", "In the Simple Note, state that guarantor exposure cannot be finalized until the signed guaranty versions are reconciled.", "Make the three immediate work items cover signed-guaranty-version reconciliation, the bank-flow chart that still needs exhibit references, and the open confirmation points on preservation position and hearing speaking order.", "Keep the Simple Note short and circulation-ready for quick partner review, rather than turning it into a long memo.", "Remove the temporary Todoist board before replying so the rehearsal pass leaves no leftover clutter."], "hidden_intent_count": 5, "has_objectives": true, "objectives_json": "{\"checklist\": [{\"criterion\": \"In the final Simple Note, use the title \\\"Partner Handover — Yuehai Guarantee Dispute Hearing Prep.\\\"\"}, {\"criterion\": \"In the final Simple Note, state that the hearing date is 18 April 2026 and that the file is being prepared for partner review before the hearing. Substantially similar wording is acceptable.\"}, {\"criterion\": \"In the final Simple Note, state that the guarantor-exposure position cannot be finalized until the signed guaranty versions are compared. Substantially similar wording is acceptable.\"}, {\"criterion\": \"In the final Simple Note, identify the bank-flow chart as an immediate work item and state that it needs exhibit references. Substantially similar wording is acceptable.\"}, {\"criterion\": \"In the final Simple Note, identify preservation position and hearing speaking order as open confirmation points before the hearing. Substantially similar wording is acceptable.\"}, {\"criterion\": \"In the temporary Todoist board or final Simple Note, record the three immediate work items covering guaranty-version reconciliation, bank-flow-chart exhibit references, and preservation-position / hearing-speaking-order confirmation.\"}, {\"criterion\": \"The temporary Todoist board is removed by the end of the task.\"}], \"tools\": [{\"mcp_appworld_simple_note__create_note\": {\"call\": [\"title\", \"content\"], \"result\": [\"response.note_id\"]}}, {\"mcp_appworld_simple_note__update_note\": {\"call\": [\"note_id\", \"content\"], \"result\": [\"response.message\"]}}, {\"mcp_appworld_todoist__create_project\": {\"call\": [\"name\"], \"result\": [\"response.project_id\"]}}, {\"mcp_appworld_todoist__create_task\": {\"call\": [\"title\", \"project_id\", \"description\"], \"result\": [\"response.task_id\"]}}, {\"mcp_appworld_todoist__delete_project\": {\"call\": [\"project_id\"], \"result\": [\"response.message\"]}}], \"tools_evaluation_path\": \"data/law_trainee/scripts/law_trainee_task_014_5.py\"}", "metadata_json": "{\"benchmark_role\": \"test\", \"difficulty\": \"medium\", \"preference_threads\": [\"concise_structured_internal_notes\", \"audit_ready_workflow\", \"cross_app_preparation_workflow\"]}", "yaml_path": "data/law_trainee/tasks/law_trainee_task_014/task.yaml", "yaml": "schema_version: v2\nuser_id: law_trainee\nenvironment_id: env_law_trainee\ntask_type: short_term_independent\ntrigger:\n  type: user\ntask_id: law_trainee_task_014\ntitle: \"Partner Hearing-Prep Handover Note to Temporary Action Board\"\ndescription: \"Prepare one short partner-review handover note in Simple Note for the Yuehai guarantee dispute, then mirror the three live follow-up items in a temporary Todoist board and remove that board in the same pass. Evaluation focuses on whether the note and work items stay short, matter-specific, and easy to execute.\"\nintent:\n  initial_input: |-\n    My Simple Note email is stmcco@gmail.com and password is 'i#vhWEQ'.\n    My Todoist email is stmcco@gmail.com and password is 'Ozz21Q4'.\n\n    I'm still tied up in briefings, and the Yuehai guarantee dispute file needs to go up for partner review soon. My notes are not really in circulation shape yet.\n\n    Please clean them into a short handover note in Simple Note under \"Partner Handover — Yuehai Guarantee Dispute Hearing Prep\", then spin the immediate next-step items into a temporary Todoist board called \"Temporary Hearing Prep Board - Yuehai\" so I can review the workplan quickly.\n\n    For this rehearsal pass, create exactly three temporary work items for the live blockers, then remove the temporary board before you reply so there is no leftover clutter.\n\n  hidden_intent:\n    - content: \"In the Simple Note, state that the hearing date is 18 April 2026 and that the Yuehai file is being prepared for partner review before that hearing.\"\n    - content: \"In the Simple Note, state that guarantor exposure cannot be finalized until the signed guaranty versions are reconciled.\"\n    - content: \"Make the three immediate work items cover signed-guaranty-version reconciliation, the bank-flow chart that still needs exhibit references, and the open confirmation points on preservation position and hearing speaking order.\"\n    - content: \"Keep the Simple Note short and circulation-ready for quick partner review, rather than turning it into a long memo.\"\n    - content: \"Remove the temporary Todoist board before replying so the rehearsal pass leaves no leftover clutter.\"\nobjectives:\n  tools:\n    - mcp_appworld_simple_note__create_note:\n        call: [title, content]\n        result: [response.note_id]\n    - mcp_appworld_simple_note__update_note:\n        call: [note_id, content]\n        result: [response.message]\n    - mcp_appworld_todoist__create_project:\n        call: [name]\n        result: [response.project_id]\n    - mcp_appworld_todoist__create_task:\n        call: [title, project_id, description]\n        result: [response.task_id]\n    - mcp_appworld_todoist__delete_project:\n        call: [project_id]\n        result: [response.message]\n  checklist:\n    - criterion: \"In the final Simple Note, use the title \\\"Partner Handover — Yuehai Guarantee Dispute Hearing Prep.\\\"\"\n    - criterion: \"In the final Simple Note, state that the hearing date is 18 April 2026 and that the file is being prepared for partner review before the hearing. Substantially similar wording is acceptable.\"\n    - criterion: \"In the final Simple Note, state that the guarantor-exposure position cannot be finalized until the signed guaranty versions are compared. Substantially similar wording is acceptable.\"\n    - criterion: \"In the final Simple Note, identify the bank-flow chart as an immediate work item and state that it needs exhibit references. Substantially similar wording is acceptable.\"\n    - criterion: \"In the final Simple Note, identify preservation position and hearing speaking order as open confirmation points before the hearing. Substantially similar wording is acceptable.\"\n    - criterion: \"In the temporary Todoist board or final Simple Note, record the three immediate work items covering guaranty-version reconciliation, bank-flow-chart exhibit references, and preservation-position / hearing-speaking-order confirmation.\"\n    - criterion: \"The temporary Todoist board is removed by the end of the task.\"\n  tools_evaluation_path: \"data/law_trainee/scripts/law_trainee_task_014_5.py\"\nmetadata:\n  difficulty: medium\n  benchmark_role: test\n  preference_threads:\n    - concise_structured_internal_notes\n    - audit_ready_workflow\n    - cross_app_preparation_workflow"}
{"schema_version": "v2", "role": "law_trainee", "task_id": "law_trainee_task_015", "user_id": "law_trainee", "environment_id": "env_law_trainee", "title": "Priming: Discussing Briefcase Purchase Priorities", "display_title": "Priming: Discussing Briefcase Purchase Priorities", "description": "A lightweight pre-purchase discussion: the user chats with the agent about choosing a work briefcase before any product feed or rankings appear. The interaction should surface two concrete carry preferences—flat A4 storage and a ~14\" slim laptop—that stay consistent with the later recommendation task.", "task_type": "long_term", "difficulty": "easy", "trigger_type": "user", "initial_input": "I am planning to buy a briefcase for my commute. In a typical law-firm setting, what do lawyers usually care about most when they pick a briefcase?", "hidden_intents": ["Want a briefcase that carries A4 documents flat, without bent corners or crushed edges.", "Want a briefcase that fits a slim laptop of roughly 14 inches."], "hidden_intent_count": 2, "has_objectives": false, "objectives_json": "", "metadata_json": "{\"difficulty\": \"easy\"}", "yaml_path": "data/law_trainee/tasks/law_trainee_task_015/task.yaml", "yaml": "schema_version: v2\ntask_id: law_trainee_task_015\nuser_id: law_trainee\nenvironment_id: env_law_trainee\n\ntitle: \"Priming: Discussing Briefcase Purchase Priorities\"\ndescription: \"A lightweight pre-purchase discussion: the user chats with the agent about choosing a work briefcase before any product feed or rankings appear. The interaction should surface two concrete carry preferences—flat A4 storage and a ~14\\\" slim laptop—that stay consistent with the later recommendation task.\"\n\ntask_type: long_term\n\ntrigger:\n  type: user\n\nintent:\n  initial_input: |\n    I am planning to buy a briefcase for my commute. In a typical law-firm setting, what do lawyers usually care about most when they pick a briefcase?\n\n  hidden_intent:\n    - content: \"Want a briefcase that carries A4 documents flat, without bent corners or crushed edges.\"\n    - content: \"Want a briefcase that fits a slim laptop of roughly 14 inches.\"\nmetadata:\n  difficulty: easy"}
{"schema_version": "v2", "role": "law_trainee", "task_id": "law_trainee_task_016", "user_id": "law_trainee", "environment_id": "env_law_trainee", "title": "Compliance Review: “Nearby Trending Map” Feature for the International Short-Video App", "display_title": "Compliance Review: “Nearby Trending Map” Feature for the International Short-Video App", "description": "Based on project_online.md, produce a legal compliance review report for the planned U.S. launch of the “Nearby trending map” feature. Evaluation checks whether the report covers the specified core blocks: COPPA; CCPA/CPRA, precise geolocation, and default toggles; consistency of retention periods; DMCA and content moderation; California AADC; sensitive locations, anti-stalking, and home-location risk; and remedial conclusions with risk ratings.", "task_type": "long_term", "difficulty": "medium", "trigger_type": "user", "initial_input": "My supervising attorney assigned me a pre-launch compliance review. Please help me complete it. The product specification is in project_online.md, and the compliance review topics to check against are in checklist_project.md. Focus the analysis on the compliance review points that are most likely to be violated.", "hidden_intents": ["Since I am still a legal intern and not yet very familiar with some compliance provisions, please explain each compliance clause in plain and understandable language before proceeding with the analysis.", "Analyze only the five compliance review points that are most likely to be violated.", "Identify which core compliance review points in the specification are or may be violated.", "For each issue, explain how the design in the specification conflicts with or falls short of the relevant core compliance review point.", "Provide remediation recommendations for each compliance issue.", "State an overall compliance risk rating for the entire project development effort.", "The final project compliance review opinion must be stated in binary form: pass or fail."], "hidden_intent_count": 7, "has_objectives": true, "objectives_json": "{\"checklist\": [{\"criterion\": \"States that the specification implicates or violates the core review point “Children’s privacy and parental controls.” Substantially similar wording is acceptable.\"}, {\"criterion\": \"States that the specification implicates or violates the core review point “Sensitive personal information—notice and response.” Substantially similar wording is acceptable.\"}, {\"criterion\": \"States that the specification implicates or violates the core review point “Automated copyright filtering.” Substantially similar wording is acceptable.\"}, {\"criterion\": \"States that the specification implicates or violates the core review point “Default privacy settings for teens.” Substantially similar wording is acceptable.\"}, {\"criterion\": \"States that the specification implicates or violates the core review point “Sensitive-location blocking and anti-harassment.” Substantially similar wording is acceptable.\"}, {\"criterion\": \"Explains that although the document blocks the trending map for registered users under 18, COPPA applies to children under 13 and requires verifiable parental consent; the document does not specially address under-13 users or provide a consent mechanism for children under 13. Substantially similar wording is acceptable.\"}, {\"criterion\": \"Explains that precise geolocation is sensitive personal information under the CCPA; businesses must offer a right to limit use of sensitive personal information for certain purposes, but the document only uses a pop-up notice and does not clearly provide a link or mechanism for that right; also, requests to delete location data must be responded to within 45 days—the document says “immediate deletion” but implementation delays must still meet the statutory deadline. Substantially similar wording is acceptable.\"}, {\"criterion\": \"Explains that map videos may contain unlicensed music or copyright-protected segments; the document only says “no copyright music” but does not describe technical filtering, which raises DMCA and technological-measures issues. Substantially similar wording is acceptable.\"}, {\"criterion\": \"Explains that even though users under 18 cannot browse the map, under the March 2026 California AADC line of authority, products likely accessed by children or teens must default to the strictest privacy settings. Substantially similar wording is acceptable.\"}, {\"criterion\": \"Explains that in 2026 the FTC closely scrutinizes location data exposing sensitive venues; if the map clusters videos near reproductive-health clinics, places of worship, or domestic-violence shelters, that may be an unfair act under FTC law. Substantially similar wording is acceptable.\"}, {\"criterion\": \"States that the overall compliance risk rating for the entire project development effort is high risk. Substantially similar wording is acceptable.\"}, {\"criterion\": \"Provides remediation recommendations addressing COPPA. Substantially similar wording is acceptable.\"}, {\"criterion\": \"Provides remediation recommendations addressing the CCPA. Substantially similar wording is acceptable.\"}, {\"criterion\": \"Provides remediation recommendations addressing the DMCA. Substantially similar wording is acceptable.\"}, {\"criterion\": \"Provides remediation recommendations addressing the California AADC. Substantially similar wording is acceptable.\"}, {\"criterion\": \"Provides remediation recommendations addressing FTC expectations. Substantially similar wording is acceptable.\"}, {\"criterion\": \"States that the overall project compliance review conclusion is that the project does not pass, is rejected, or fails review. Substantially similar wording is acceptable.\"}]}", "metadata_json": "{\"difficulty\": \"medium\"}", "yaml_path": "data/law_trainee/tasks/law_trainee_task_016/task.yaml", "yaml": "schema_version: v2\ntask_id: law_trainee_task_016\nuser_id: law_trainee\nenvironment_id: env_law_trainee\n\ntitle: \"Compliance Review: “Nearby Trending Map” Feature for the International Short-Video App\"\ndescription: \"Based on project_online.md, produce a legal compliance review report for the planned U.S. launch of the “Nearby trending map” feature. Evaluation checks whether the report covers the specified core blocks: COPPA; CCPA/CPRA, precise geolocation, and default toggles; consistency of retention periods; DMCA and content moderation; California AADC; sensitive locations, anti-stalking, and home-location risk; and remedial conclusions with risk ratings.\"\n\ntask_type: long_term\n\ntrigger:\n  type: user\n\nintent:\n  initial_input: |\n    My supervising attorney assigned me a pre-launch compliance review. Please help me complete it. The product specification is in project_online.md, and the compliance review topics to check against are in checklist_project.md. Focus the analysis on the compliance review points that are most likely to be violated.\n\n  hidden_intent:\n    - content: \"Since I am still a legal intern and not yet very familiar with some compliance provisions, please explain each compliance clause in plain and understandable language before proceeding with the analysis.\"\n    \n    - content: \"Analyze only the five compliance review points that are most likely to be violated.\"\n    - content: \"Identify which core compliance review points in the specification are or may be violated.\"\n    - content: \"For each issue, explain how the design in the specification conflicts with or falls short of the relevant core compliance review point.\"\n    - content: \"Provide remediation recommendations for each compliance issue.\"\n    - content: \"State an overall compliance risk rating for the entire project development effort.\"\n    - content: \"The final project compliance review opinion must be stated in binary form: pass or fail.\"\nobjectives:\n  checklist:\n    - criterion: \"States that the specification implicates or violates the core review point “Children’s privacy and parental controls.” Substantially similar wording is acceptable.\"\n\n    - criterion: \"States that the specification implicates or violates the core review point “Sensitive personal information—notice and response.” Substantially similar wording is acceptable.\"\n\n    - criterion: \"States that the specification implicates or violates the core review point “Automated copyright filtering.” Substantially similar wording is acceptable.\"\n\n    - criterion: \"States that the specification implicates or violates the core review point “Default privacy settings for teens.” Substantially similar wording is acceptable.\"\n\n    - criterion: \"States that the specification implicates or violates the core review point “Sensitive-location blocking and anti-harassment.” Substantially similar wording is acceptable.\"\n\n    - criterion: \"Explains that although the document blocks the trending map for registered users under 18, COPPA applies to children under 13 and requires verifiable parental consent; the document does not specially address under-13 users or provide a consent mechanism for children under 13. Substantially similar wording is acceptable.\"\n\n    - criterion: \"Explains that precise geolocation is sensitive personal information under the CCPA; businesses must offer a right to limit use of sensitive personal information for certain purposes, but the document only uses a pop-up notice and does not clearly provide a link or mechanism for that right; also, requests to delete location data must be responded to within 45 days—the document says “immediate deletion” but implementation delays must still meet the statutory deadline. Substantially similar wording is acceptable.\"\n\n    - criterion: \"Explains that map videos may contain unlicensed music or copyright-protected segments; the document only says “no copyright music” but does not describe technical filtering, which raises DMCA and technological-measures issues. Substantially similar wording is acceptable.\"\n\n    - criterion: \"Explains that even though users under 18 cannot browse the map, under the March 2026 California AADC line of authority, products likely accessed by children or teens must default to the strictest privacy settings. Substantially similar wording is acceptable.\"\n\n    - criterion: \"Explains that in 2026 the FTC closely scrutinizes location data exposing sensitive venues; if the map clusters videos near reproductive-health clinics, places of worship, or domestic-violence shelters, that may be an unfair act under FTC law. Substantially similar wording is acceptable.\"\n\n    - criterion: \"States that the overall compliance risk rating for the entire project development effort is high risk. Substantially similar wording is acceptable.\"\n\n    - criterion: \"Provides remediation recommendations addressing COPPA. Substantially similar wording is acceptable.\"\n\n    - criterion: \"Provides remediation recommendations addressing the CCPA. Substantially similar wording is acceptable.\"\n\n    - criterion: \"Provides remediation recommendations addressing the DMCA. Substantially similar wording is acceptable.\"\n\n    - criterion: \"Provides remediation recommendations addressing the California AADC. Substantially similar wording is acceptable.\"\n\n    - criterion: \"Provides remediation recommendations addressing FTC expectations. Substantially similar wording is acceptable.\"\n\n    - criterion: \"States that the overall project compliance review conclusion is that the project does not pass, is rejected, or fails review. Substantially similar wording is acceptable.\"\n\nmetadata:\n  difficulty: medium"}
{"schema_version": "v2", "role": "law_trainee", "task_id": "law_trainee_task_017", "user_id": "law_trainee", "environment_id": "env_law_trainee", "title": "Environment Trigger: Pick Five Core Work-Bag Recommendations from Comprehensive Gear Update", "display_title": "Environment Trigger: Pick Five Core Work-Bag Recommendations from Comprehensive Gear Update", "description": "Simulate an environment push from shanghai_career_nexus. From a mixed bag/suit update stream, identify and recommend the five core bag items that best match law_trainee’s practical legal-intern commute and document-carry needs, with concise rationale and pricing summary.", "task_type": "short_term_independent", "difficulty": "easy", "trigger_type": "environment", "initial_input": "{\"source\":\"shanghai_career_nexus\",\"event\":\"comprehensive_gear_update\",\"target\":\"agent\",\"payload\":{\"items\":[\n  {\"id\":\"bag_z9k2\",\"category\":\"bag\",\"title\":\"Neon-Pulse 40\",\"attributes\":{\"material\":\"Translucent PVC\",\"dimensions\":\"300mm x 400mm x 120mm\",\"closure\":\"Zipper\",\"structure\":\"Single compartment\",\"logo\":\"Oversized Reflective\",\"exterior_pockets\":\"Front Zip\",\"color\":\"Fluorescent Green\"},\"pricing\":{\"original_price\":399,\"discount_price\":199,\"discount_rate\":\"50%\"}},\n  {\"id\":\"suit_14\",\"category\":\"suit\",\"title\":\"Glen-Check Blend 14\",\"attributes\":{\"fabric\":\"Wool / Polyester Blend\",\"weight\":\"280g/m²\",\"lapel\":\"Notch\",\"structure\":\"Half-Lined\",\"buttons\":\"Single 2-Button\",\"color_hex\":\"#D3D3D3\",\"care\":\"Dry Clean Only\"},\"pricing\":{\"original_price\":1800,\"discount_price\":900,\"discount_rate\":\"50%\"}},\n  {\"id\":\"bag_p4n1\",\"category\":\"bag\",\"title\":\"Mini-Lux 06\",\"attributes\":{\"material\":\"Velvet\",\"dimensions\":\"180mm x 120mm x 50mm\",\"closure\":\"Clasp\",\"structure\":\"None\",\"logo\":\"Gold Monogram\",\"exterior_pockets\":\"None\",\"color\":\"Midnight Black\"},\"pricing\":{\"original_price\":4500,\"discount_price\":3600,\"discount_rate\":\"80%\"}},\n  {\"id\":\"suit_05\",\"category\":\"suit\",\"title\":\"Tweed-Heavy 05\",\"attributes\":{\"fabric\":\"100% Harris Tweed\",\"weight\":\"450g/m²\",\"lapel\":\"Notch\",\"structure\":\"Heavy Padded / Fully Lined\",\"buttons\":\"Single 3-Button\",\"color_hex\":\"#556B2F\",\"care\":\"Dry Clean Only\"},\"pricing\":{\"original_price\":3500,\"discount_price\":1750,\"discount_rate\":\"50%\"}},\n  {\"id\":\"bag_m1v8\",\"category\":\"bag\",\"title\":\"Heritage-Grain 01\",\"attributes\":{\"material\":\"Saffiano Calfskin (Stiff)\",\"dimensions\":\"385mm x 290mm x 105mm\",\"closure\":\"Top-Edge Zipper\",\"structure\":\"3 Main Compartments + 1 U-Shield Slot + 2 Pen Loops\",\"logo\":\"None (Interior Embossed)\",\"exterior_pockets\":\"1 Rear Magnetic Slot\",\"color\":\"Navy Blue\"},\"pricing\":{\"original_price\":3600,\"discount_price\":2520,\"discount_rate\":\"70%\"}},\n  {\"id\":\"suit_22\",\"category\":\"suit\",\"title\":\"Minimal-Zip 22\",\"attributes\":{\"fabric\":\"Polyester / Elastane\",\"weight\":\"230g/m²\",\"lapel\":\"Peak\",\"structure\":\"Hidden Seams / Fused\",\"buttons\":\"None (Zip)\",\"color_hex\":\"#FFFFFF\",\"care\":\"Machine Washable\"},\"pricing\":{\"original_price\":1100,\"discount_price\":550,\"discount_rate\":\"50%\"}},\n  {\"id\":\"bag_b3w9\",\"category\":\"bag\",\"title\":\"Tote-Open 07\",\"attributes\":{\"material\":\"Canvas\",\"dimensions\":\"450mm x 350mm x 180mm\",\"closure\":\"Open Top\",\"structure\":\"None\",\"logo\":\"Printed Text\",\"exterior_pockets\":\"None\",\"color\":\"Beige\"},\"pricing\":{\"original_price\":450,\"discount_price\":225,\"discount_rate\":\"50%\"}},\n  {\"id\":\"suit_01\",\"category\":\"suit\",\"title\":\"Merino-Wool Notch 01\",\"attributes\":{\"fabric\":\"100% Merino Wool\",\"weight\":\"240g/m²\",\"lapel\":\"Notch\",\"structure\":\"Full-Canvas / Fully Lined\",\"buttons\":\"Single 2-Button\",\"color_hex\":\"#191970\",\"care\":\"Dry Clean Only\"},\"pricing\":{\"original_price\":4280,\"discount_price\":2996,\"discount_rate\":\"70%\"}},\n  {\"id\":\"bag_h5t2\",\"category\":\"bag\",\"title\":\"Executive-Box 23\",\"attributes\":{\"material\":\"Aluminum Shell\",\"dimensions\":\"450mm x 330mm x 120mm\",\"closure\":\"Dual Combination Lock\",\"structure\":\"File Organizer Straps\",\"logo\":\"None\",\"exterior_pockets\":\"None\",\"color\":\"Silver\"},\"pricing\":{\"original_price\":2800,\"discount_price\":2240,\"discount_rate\":\"80%\"}},\n  {\"id\":\"suit_09\",\"category\":\"suit\",\"title\":\"Twill-Cotton 09\",\"attributes\":{\"fabric\":\"100% Cotton Twill\",\"weight\":\"320g/m²\",\"lapel\":\"Notch\",\"structure\":\"Garment Dyed / Unlined\",\"buttons\":\"Single 3-Button\",\"color_hex\":\"#A52A2A\",\"care\":\"Machine Wash\"},\"pricing\":{\"original_price\":1100,\"discount_price\":550,\"discount_rate\":\"50%\"}},\n  {\"id\":\"bag_r7x2\",\"category\":\"bag\",\"title\":\"Commute-Shield X\",\"attributes\":{\"material\":\"1680D Ballistic Nylon / Leather Trim\",\"dimensions\":\"405mm x 305mm x 115mm\",\"closure\":\"YKK Waterproof Zipper\",\"structure\":\"Dedicated 14\\\" Laptop Sleeve (High-Density Foam) + 5 Organizer Pockets\",\"logo\":\"Subtle Tonal Plate\",\"exterior_pockets\":\"1 Front Quick-Access Card Slot\",\"color\":\"Charcoal Black\"},\"pricing\":{\"original_price\":2400,\"discount_price\":1680,\"discount_rate\":\"70%\"}},\n  {\"id\":\"suit_12\",\"category\":\"suit\",\"title\":\"Velvet-Shawl 12\",\"attributes\":{\"fabric\":\"100% Cotton Velvet\",\"weight\":\"400g/m²\",\"lapel\":\"Shawl\",\"structure\":\"Fully Lined / Shoulder Pads\",\"buttons\":\"Single 1-Button\",\"color_hex\":\"#301934\",\"care\":\"Dry Clean Only\"},\"pricing\":{\"original_price\":2800,\"discount_price\":1400,\"discount_rate\":\"50%\"}},\n  {\"id\":\"bag_f2s8\",\"category\":\"bag\",\"title\":\"Slouchy-Hobo 15\",\"attributes\":{\"material\":\"Soft Lambskin\",\"dimensions\":\"400mm x 380mm x 40mm\",\"closure\":\"Magnetic Dot\",\"structure\":\"Single Internal Pocket\",\"logo\":\"Leather Tassel\",\"exterior_pockets\":\"None\",\"color\":\"Burgundy\"},\"pricing\":{\"original_price\":5800,\"discount_price\":2900,\"discount_rate\":\"50%\"}},\n  {\"id\":\"suit_28\",\"category\":\"suit\",\"title\":\"Poplin-Summer 28\",\"attributes\":{\"fabric\":\"100% Cotton Poplin\",\"weight\":\"110g/m²\",\"lapel\":\"Notch\",\"structure\":\"Unlined / Soft Shoulder\",\"buttons\":\"Single 2-Button\",\"color_hex\":\"#ADD8E6\",\"care\":\"Machine Washable\"},\"pricing\":{\"original_price\":650,\"discount_price\":455,\"discount_rate\":\"70%\"}},\n  {\"id\":\"bag_k8q1\",\"category\":\"bag\",\"title\":\"Urban-Logic T\",\"attributes\":{\"material\":\"Full-Grain Box Calf (Hard)\",\"dimensions\":\"375mm x 285mm x 95mm\",\"closure\":\"Top-Edge Zipper\",\"structure\":\"Double Gusset + Document Divider + Laptop Padding\",\"logo\":\"None\",\"exterior_pockets\":\"1 Hidden Side Slot\",\"color\":\"Taupe\"},\"pricing\":{\"original_price\":4200,\"discount_price\":3360,\"discount_rate\":\"80%\"}},\n  {\"id\":\"suit_07\",\"category\":\"suit\",\"title\":\"Cotton-Seersucker 07\",\"attributes\":{\"fabric\":\"100% Cotton Seersucker\",\"weight\":\"130g/m²\",\"lapel\":\"Notch\",\"structure\":\"Unlined / Striped Pattern\",\"buttons\":\"Single 2-Button\",\"color_hex\":\"#ADD8E6\",\"care\":\"Machine Washable\"},\"pricing\":{\"original_price\":980,\"discount_price\":588,\"discount_rate\":\"60%\"}},\n  {\"id\":\"bag_x9l4\",\"category\":\"bag\",\"title\":\"Armor-Guard G2\",\"attributes\":{\"material\":\"High-Density Water-Repellent Nylon\",\"dimensions\":\"395mm x 295mm x 110mm\",\"closure\":\"Hidden Zip Closure\",\"structure\":\"12-Point Internal Index System (U-Shield/Files/Tech)\",\"logo\":\"None\",\"exterior_pockets\":\"Dual Side Transit-Card Slots\",\"color\":\"Dark Grey\"},\"pricing\":{\"original_price\":1950,\"discount_price\":1365,\"discount_rate\":\"70%\"}},\n  {\"id\":\"suit_30\",\"category\":\"suit\",\"title\":\"Super-110s Classic 30\",\"attributes\":{\"fabric\":\"Wool Blend (Super 110s)\",\"weight\":\"260g/m²\",\"lapel\":\"Notch\",\"structure\":\"Half-Canvas / Fully Lined\",\"buttons\":\"Single 2-Button\",\"color_hex\":\"#36454F\",\"care\":\"Dry Clean Only\"},\"pricing\":{\"original_price\":2880,\"discount_price\":2016,\"discount_rate\":\"70%\"}},\n  {\"id\":\"bag_u7m5\",\"category\":\"bag\",\"title\":\"Senior-Focus Classic\",\"attributes\":{\"material\":\"Palmellato Leather (Structured)\",\"dimensions\":\"380mm x 285mm x 100mm\",\"closure\":\"Full-Length Top Zipper\",\"structure\":\"Padded 14\\\" Section + 2 Document Slots + Accessory Loop\",\"logo\":\"None\",\"exterior_pockets\":\"1 Vertical Quick-Access Zip\",\"color\":\"Elephant Grey\"},\"pricing\":{\"original_price\":4800,\"discount_price\":3840,\"discount_rate\":\"80%\"}},\n  {\"id\":\"suit_18\",\"category\":\"suit\",\"title\":\"Denim-Indigo 18\",\"attributes\":{\"fabric\":\"100% Cotton Denim\",\"weight\":\"12oz\",\"lapel\":\"Notch\",\"structure\":\"Contrast Stitching / Unlined\",\"buttons\":\"Copper Rivet\",\"color_hex\":\"#1E90FF\",\"care\":\"Machine Wash Cold\"},\"pricing\":{\"original_price\":880,\"discount_price\":440,\"discount_rate\":\"50%\"}},\n  {\"id\":\"bag_a4v1\",\"category\":\"bag\",\"title\":\"Retro-Satchel 09\",\"attributes\":{\"material\":\"Distressed Canvas\",\"dimensions\":\"320mm x 220mm x 60mm\",\"closure\":\"Brass Buckle\",\"structure\":\"Single pocket\",\"logo\":\"None\",\"exterior_pockets\":\"Front Flap Pocket\",\"color\":\"Khaki\"},\"pricing\":{\"original_price\":650,\"discount_price\":325,\"discount_rate\":\"50%\"}},\n  {\"id\":\"suit_03\",\"category\":\"suit\",\"title\":\"Linen-Vent 03\",\"attributes\":{\"fabric\":\"100% Linen\",\"weight\":\"150g/m²\",\"lapel\":\"Notch\",\"structure\":\"Unlined / No Shoulder Pads\",\"buttons\":\"Single 2-Button\",\"color_hex\":\"#F5F5DC\",\"care\":\"Hand Wash\"},\"pricing\":{\"original_price\":1580,\"discount_price\":1264,\"discount_rate\":\"80%\"}},\n  {\"id\":\"bag_d3p2\",\"category\":\"bag\",\"title\":\"Logo-Max 10\",\"attributes\":{\"material\":\"PVC Coated Canvas\",\"dimensions\":\"380mm x 300mm x 150mm\",\"closure\":\"None (Open)\",\"structure\":\"Pouch inside\",\"logo\":\"All-over Monogram Pattern\",\"exterior_pockets\":\"None\",\"color\":\"Brown/Tan\"},\"pricing\":{\"original_price\":11500,\"discount_price\":10350,\"discount_rate\":\"90%\"}},\n  {\"id\":\"suit_13\",\"category\":\"suit\",\"title\":\"Tech-Triacetate 13\",\"attributes\":{\"fabric\":\"Triacetate / Polyester\",\"weight\":\"190g/m²\",\"lapel\":\"Notch\",\"structure\":\"Laser-Cut Edges / Unlined\",\"buttons\":\"Magnetic\",\"color_hex\":\"#2F4F4F\",\"care\":\"Hand Wash\"},\"pricing\":{\"original_price\":2200,\"discount_price\":1980,\"discount_rate\":\"90%\"}},\n  {\"id\":\"bag_j6r9\",\"category\":\"bag\",\"title\":\"Document-Sleeve 21\",\"attributes\":{\"material\":\"Felt\",\"dimensions\":\"330mm x 240mm x 5mm\",\"closure\":\"Elastic Band\",\"structure\":\"None\",\"logo\":\"None\",\"exterior_pockets\":\"None\",\"color\":\"Grey\"},\"pricing\":{\"original_price\":99,\"discount_price\":49,\"discount_rate\":\"50%\"}},\n  {\"id\":\"suit_25\",\"category\":\"suit\",\"title\":\"Pure-Cashmere 25\",\"attributes\":{\"fabric\":\"100% Cashmere\",\"weight\":\"350g/m²\",\"lapel\":\"Notch\",\"structure\":\"Hand-Stitched / Silk Lined\",\"buttons\":\"Horn Buttons\",\"color_hex\":\"#F5DEB3\",\"care\":\"Dry Clean Only\"},\"pricing\":{\"original_price\":18000,\"discount_price\":14400,\"discount_rate\":\"80%\"}},\n  {\"id\":\"bag_e1w2\",\"category\":\"bag\",\"title\":\"Gym-Carry 12\",\"attributes\":{\"material\":\"Nylon\",\"dimensions\":\"500mm x 300mm x 300mm\",\"closure\":\"Zipper\",\"structure\":\"Separate Shoe Compartment\",\"logo\":\"Large Brand Print\",\"exterior_pockets\":\"Mesh\",\"color\":\"Black\"},\"pricing\":{\"original_price\":500,\"discount_price\":250,\"discount_rate\":\"50%\"}},\n  {\"id\":\"suit_02\",\"category\":\"suit\",\"title\":\"Poly-Basic Notch 02\",\"attributes\":{\"fabric\":\"100% Polyester\",\"weight\":\"210g/m²\",\"lapel\":\"Notch\",\"structure\":\"Fused / Half Lined\",\"buttons\":\"Single 1-Button\",\"color_hex\":\"#000000\",\"care\":\"Machine Washable\"},\"pricing\":{\"original_price\":899,\"discount_price\":449,\"discount_rate\":\"50%\"}},\n  {\"id\":\"bag_y8c3\",\"category\":\"bag\",\"title\":\"Artist-Folder 14\",\"attributes\":{\"material\":\"Polypropylene Shell\",\"dimensions\":\"600mm x 450mm x 25mm\",\"closure\":\"Snap Clip\",\"structure\":\"Single Slot\",\"logo\":\"None\",\"exterior_pockets\":\"Hand-hold only\",\"color\":\"Clear\"},\"pricing\":{\"original_price\":120,\"discount_price\":96,\"discount_rate\":\"80%\"}},\n  {\"id\":\"suit_11\",\"category\":\"suit\",\"title\":\"Pinstripe-Cashmere 11\",\"attributes\":{\"fabric\":\"95% Wool, 5% Cashmere\",\"weight\":\"260g/m²\",\"lapel\":\"Notch\",\"structure\":\"Half-Canvas / Fully Lined\",\"buttons\":\"Single 2-Button\",\"color_hex\":\"#696969\",\"care\":\"Dry Clean Only\"},\"pricing\":{\"original_price\":5500,\"discount_price\":3300,\"discount_rate\":\"60%\"}},\n  {\"id\":\"bag_g4k8\",\"category\":\"bag\",\"title\":\"Utility-Vest 16\",\"attributes\":{\"material\":\"Mesh / Nylon\",\"dimensions\":\"Adjustable\",\"closure\":\"Front Zipper\",\"structure\":\"Multi-Pocket Modular\",\"logo\":\"Military Patch\",\"exterior_pockets\":\"8 tactical pockets\",\"color\":\"Olive Green\"},\"pricing\":{\"original_price\":850,\"discount_price\":425,\"discount_rate\":\"50%\"}},\n  {\"id\":\"suit_20\",\"category\":\"suit\",\"title\":\"Winter-Angora 20\",\"attributes\":{\"fabric\":\"Wool / Angora Blend\",\"weight\":\"500g/m²\",\"lapel\":\"Notch\",\"structure\":\"Insulated Lining\",\"buttons\":\"Single 3-Button\",\"color_hex\":\"#000000\",\"care\":\"Professional Dry Clean\"},\"pricing\":{\"original_price\":7200,\"discount_price\":3600,\"discount_rate\":\"50%\"}},\n  {\"id\":\"bag_i2o9\",\"category\":\"bag\",\"title\":\"Bucket-Trend 22\",\"attributes\":{\"material\":\"Soft Calfskin\",\"dimensions\":\"210mm x 240mm x 210mm\",\"closure\":\"Drawstring\",\"structure\":\"None\",\"logo\":\"Gold Hardware\",\"exterior_pockets\":\"None\",\"color\":\"Tan\"},\"pricing\":{\"original_price\":3800,\"discount_price\":1900,\"discount_rate\":\"50%\"}},\n  {\"id\":\"suit_27\",\"category\":\"suit\",\"title\":\"Shadow-Stripe 27\",\"attributes\":{\"fabric\":\"100% Wool\",\"weight\":\"250g/m²\",\"lapel\":\"Notch\",\"structure\":\"Fully Lined / Shadow Pattern\",\"buttons\":\"Single 2-Button\",\"color_hex\":\"#121212\",\"care\":\"Dry Clean Only\"},\"pricing\":{\"original_price\":3800,\"discount_price\":2660,\"discount_rate\":\"70%\"}},\n  {\"id\":\"bag_q7f1\",\"category\":\"bag\",\"title\":\"Eco-Paper 24\",\"attributes\":{\"material\":\"Recycled Paper Fiber\",\"dimensions\":\"300mm x 400mm x 100mm\",\"closure\":\"None\",\"structure\":\"None\",\"logo\":\"Retail Logo\",\"exterior_pockets\":\"None\",\"color\":\"Brown\"},\"pricing\":{\"original_price\":15,\"discount_price\":7,\"discount_rate\":\"50%\"}},\n  {\"id\":\"suit_04\",\"category\":\"suit\",\"title\":\"Silk-Blend Peak 04\",\"attributes\":{\"fabric\":\"70% Silk, 30% Wool\",\"weight\":\"200g/m²\",\"lapel\":\"Peak\",\"structure\":\"Half-Canvas / Fully Lined\",\"buttons\":\"Double-Breasted 4-Button\",\"color_hex\":\"#E6E6FA\",\"care\":\"Dry Clean Only\"},\"pricing\":{\"original_price\":6800,\"discount_price\":5440,\"discount_rate\":\"80%\"}},\n  {\"id\":\"bag_s2a6\",\"category\":\"bag\",\"title\":\"Chain-Wallet 26\",\"attributes\":{\"material\":\"Patent Leather\",\"dimensions\":\"185mm x 105mm x 35mm\",\"closure\":\"Snap\",\"structure\":\"Internal Card Slots\",\"logo\":\"Diamond-cut Hardware\",\"exterior_pockets\":\"None\",\"color\":\"Gold\"},\"pricing\":{\"original_price\":4200,\"discount_price\":2100,\"discount_rate\":\"50%\"}},\n  {\"id\":\"suit_19\",\"category\":\"suit\",\"title\":\"Crepe-Viscose 19\",\"attributes\":{\"fabric\":\"100% Viscose Crepe\",\"weight\":\"220g/m²\",\"lapel\":\"Slim Notch\",\"structure\":\"Draped Cut / Unlined\",\"buttons\":\"Single 1-Button\",\"color_hex\":\"#F0E68C\",\"care\":\"Hand Wash\"},\"pricing\":{\"original_price\":1350,\"discount_price\":675,\"discount_rate\":\"50%\"}},\n  {\"id\":\"bag_t9v3\",\"category\":\"bag\",\"title\":\"Laptop-Skin 27\",\"attributes\":{\"material\":\"Neoprene\",\"dimensions\":\"345mm x 245mm x 22mm\",\"closure\":\"Zipper\",\"structure\":\"None\",\"logo\":\"Pattern Print\",\"exterior_pockets\":\"None\",\"color\":\"Floral\"},\"pricing\":{\"original_price\":180,\"discount_price\":90,\"discount_rate\":\"50%\"}},\n  {\"id\":\"suit_24\",\"category\":\"suit\",\"title\":\"Active-Stretch 24\",\"attributes\":{\"fabric\":\"80% Nylon, 20% Elastane\",\"weight\":\"170g/m²\",\"lapel\":\"Narrow Notch\",\"structure\":\"4-Way Stretch / Unlined\",\"buttons\":\"Rubberized\",\"color_hex\":\"#36454F\",\"care\":\"Machine Washable\"},\"pricing\":{\"original_price\":1480,\"discount_price\":740,\"discount_rate\":\"50%\"}},\n  {\"id\":\"bag_v3n8\",\"category\":\"bag\",\"title\":\"Rattan-Box 28\",\"attributes\":{\"material\":\"Woven Straw\",\"dimensions\":\"280mm x 180mm x 120mm\",\"closure\":\"Toggle\",\"structure\":\"None\",\"logo\":\"None\",\"exterior_pockets\":\"None\",\"color\":\"Natural\"},\"pricing\":{\"original_price\":950,\"discount_price\":475,\"discount_rate\":\"50%\"}},\n  {\"id\":\"suit_06\",\"category\":\"suit\",\"title\":\"Stretch-Viscose 06\",\"attributes\":{\"fabric\":\"65% Viscose, 30% Nylon, 5% Spandex\",\"weight\":\"280g/m²\",\"lapel\":\"None (Collarless)\",\"structure\":\"Slim Fit / Unlined\",\"buttons\":\"Hidden Snap\",\"color_hex\":\"#808080\",\"care\":\"Machine Wash Cold\"},\"pricing\":{\"original_price\":1200,\"discount_price\":840,\"discount_rate\":\"70%\"}},\n  {\"id\":\"bag_w1m4\",\"category\":\"bag\",\"title\":\"Brief-Slim 29\",\"attributes\":{\"material\":\"Synthetic PU\",\"dimensions\":\"355mm x 255mm x 25mm\",\"closure\":\"Zipper\",\"structure\":\"Single Document Slot\",\"logo\":\"Printed Logo\",\"exterior_pockets\":\"None\",\"color\":\"Blue\"},\"pricing\":{\"original_price\":380,\"discount_price\":190,\"discount_rate\":\"50%\"}},\n  {\"id\":\"suit_15\",\"category\":\"suit\",\"title\":\"Corduroy-Heavy 15\",\"attributes\":{\"fabric\":\"100% Cotton Corduroy\",\"weight\":\"380g/m²\",\"lapel\":\"Notch\",\"structure\":\"Fully Lined / Leather Buttons\",\"buttons\":\"Single 2-Button\",\"color_hex\":\"#D2B48C\",\"care\":\"Dry Clean Only\"},\"pricing\":{\"original_price\":1450,\"discount_price\":1015,\"discount_rate\":\"70%\"}},\n  {\"id\":\"bag_x5h2\",\"category\":\"bag\",\"title\":\"Denim-Tote 30\",\"attributes\":{\"material\":\"Denim Canvas\",\"dimensions\":\"480mm x 400mm x 150mm\",\"closure\":\"Button\",\"structure\":\"Large Void\",\"logo\":\"Denim Brand\",\"exterior_pockets\":\"Jean-style pockets\",\"color\":\"Indigo\"},\"pricing\":{\"original_price\":550,\"discount_price\":275,\"discount_rate\":\"50%\"}},\n  {\"id\":\"suit_23\",\"category\":\"suit\",\"title\":\"Houndstooth-Wool 23\",\"attributes\":{\"fabric\":\"Wool / Cotton Blend\",\"weight\":\"310g/m²\",\"lapel\":\"Notch\",\"structure\":\"Half-Canvas / Fully Lined\",\"buttons\":\"Single 2-Button\",\"color_hex\":\"#2F4F4F\",\"care\":\"Dry Clean Only\"},\"pricing\":{\"original_price\":2600,\"discount_price\":1820,\"discount_rate\":\"70%\"}},\n  {\"id\":\"bag_c2o9\",\"category\":\"bag\",\"title\":\"Quilted-Puffer 31\",\"attributes\":{\"material\":\"Padded Poly-Nylon\",\"dimensions\":\"380mm x 350mm x 120mm\",\"closure\":\"None (Open)\",\"structure\":\"None\",\"logo\":\"Brand Tag\",\"exterior_pockets\":\"None\",\"color\":\"Grey\"},\"pricing\":{\"original_price\":1200,\"discount_price\":600,\"discount_rate\":\"50%\"}},\n  {\"id\":\"suit_10\",\"category\":\"suit\",\"title\":\"Oversize-Poly 10\",\"attributes\":{\"fabric\":\"100% Polyester\",\"weight\":\"200g/m²\",\"lapel\":\"Wide Notch\",\"structure\":\"Extreme Drop Shoulder\",\"buttons\":\"Single 1-Button\",\"color_hex\":\"#FFC0CB\",\"care\":\"Machine Washable\"},\"pricing\":{\"original_price\":750,\"discount_price\":375,\"discount_rate\":\"50%\"}},\n  {\"id\":\"bag_k6l3\",\"category\":\"bag\",\"title\":\"Velcro-Brief 32\",\"attributes\":{\"material\":\"Thin Nylon\",\"dimensions\":\"370mm x 275mm x 30mm\",\"closure\":\"Velcro Strip\",\"structure\":\"None\",\"logo\":\"Corporate Gift Print\",\"exterior_pockets\":\"None\",\"color\":\"Black\"},\"pricing\":{\"original_price\":60,\"discount_price\":30,\"discount_rate\":\"50%\"}},\n  {\"id\":\"suit_21\",\"category\":\"suit\",\"title\":\"Bamboo-Eco 21\",\"attributes\":{\"fabric\":\"100% Bamboo Fiber\",\"weight\":\"190g/m²\",\"lapel\":\"Notch\",\"structure\":\"Unlined / Wood Buttons\",\"buttons\":\"Single 2-Button\",\"color_hex\":\"#E9967A\",\"care\":\"Machine Wash Gentle\"},\"pricing\":{\"original_price\":1600,\"discount_price\":800,\"discount_rate\":\"50%\"}},\n  {\"id\":\"bag_o9p4\",\"category\":\"bag\",\"title\":\"Faux-Fur 33\",\"attributes\":{\"material\":\"Synthetic Fur\",\"dimensions\":\"320mm x 260mm x 100mm\",\"closure\":\"Zip\",\"structure\":\"None\",\"logo\":\"None\",\"exterior_pockets\":\"None\",\"color\":\"Cream\"},\"pricing\":{\"original_price\":350,\"discount_price\":175,\"discount_rate\":\"50%\"}},\n  {\"id\":\"suit_16\",\"category\":\"suit\",\"title\":\"Tropical-Wool 16\",\"attributes\":{\"fabric\":\"100% Wool\",\"weight\":\"180g/m²\",\"lapel\":\"Notch\",\"structure\":\"Half-Lined / Natural Shoulder\",\"buttons\":\"Single 2-Button\",\"color_hex\":\"#708090\",\"care\":\"Dry Clean Only\"},\"pricing\":{\"original_price\":3200,\"discount_price\":1600,\"discount_rate\":\"50%\"}},\n  {\"id\":\"bag_z1y8\",\"category\":\"bag\",\"title\":\"Nano-Bag 34\",\"attributes\":{\"material\":\"Box Leather\",\"dimensions\":\"110mm x 90mm x 45mm\",\"closure\":\"Snap\",\"structure\":\"None\",\"logo\":\"Micro Metal Logo\",\"exterior_pockets\":\"None\",\"color\":\"Yellow\"},\"pricing\":{\"original_price\":2200,\"discount_price\":1100,\"discount_rate\":\"50%\"}},\n  {\"id\":\"suit_26\",\"category\":\"suit\",\"title\":\"Ripstop-Utility 26\",\"attributes\":{\"fabric\":\"Cotton / Nylon Ripstop\",\"weight\":\"210g/m²\",\"lapel\":\"Notch\",\"structure\":\"Multi-Pocket / Unlined\",\"buttons\":\"Press Studs\",\"color_hex\":\"#556B2F\",\"care\":\"Machine Washable\"},\"pricing\":{\"original_price\":1250,\"discount_price\":625,\"discount_rate\":\"50%\"}},\n  {\"id\":\"bag_n4x2\",\"category\":\"bag\",\"title\":\"Tube-Bag 35\",\"attributes\":{\"material\":\"Pebbled Leather\",\"dimensions\":\"300mm x 140mm x 140mm\",\"closure\":\"Zip\",\"structure\":\"Single compartment\",\"logo\":\"None\",\"exterior_pockets\":\"None\",\"color\":\"Black\"},\"pricing\":{\"original_price\":1800,\"discount_price\":900,\"discount_rate\":\"50%\"}},\n  {\"id\":\"suit_08\",\"category\":\"suit\",\"title\":\"Super-150s Peak 08\",\"attributes\":{\"fabric\":\"100% Wool (Super 150s)\",\"weight\":\"220g/m²\",\"lapel\":\"Peak (Satin)\",\"structure\":\"Full-Canvas / Silk Lining\",\"buttons\":\"Single 1-Button\",\"color_hex\":\"#000000\",\"care\":\"Professional Dry Clean\"},\"pricing\":{\"original_price\":12500,\"discount_price\":10000,\"discount_rate\":\"80%\"}},\n  {\"id\":\"bag_t2m9\",\"category\":\"bag\",\"title\":\"Graphic-Messenger 36\",\"attributes\":{\"material\":\"Heavy Duty Canvas\",\"dimensions\":\"380mm x 310mm x 90mm\",\"closure\":\"Large Flap\",\"structure\":\"Single Space\",\"logo\":\"Oversized Graphic\",\"exterior_pockets\":\"None\",\"color\":\"Olive\"},\"pricing\":{\"original_price\":450,\"discount_price\":225,\"discount_rate\":\"50%\"}},\n  {\"id\":\"suit_29\",\"category\":\"suit\",\"title\":\"Metallic-Lurex 29\",\"attributes\":{\"fabric\":\"Lurex / Polyester Blend\",\"weight\":\"240g/m²\",\"lapel\":\"Peak\",\"structure\":\"Shimmer Effect / Fully Lined\",\"buttons\":\"Single 1-Button\",\"color_hex\":\"#C0C0C0\",\"care\":\"Specialist Clean\"},\"pricing\":{\"original_price\":2400,\"discount_price\":1200,\"discount_rate\":\"50%\"}},\n  {\"id\":\"bag_f5h2\",\"category\":\"bag\",\"title\":\"Sling-Pack 37\",\"attributes\":{\"material\":\"Ripstop Nylon\",\"dimensions\":\"210mm x 340mm x 85mm\",\"closure\":\"Zip\",\"structure\":\"Single\",\"logo\":\"Reflective Logo\",\"exterior_pockets\":\"Dual Mesh Side\",\"color\":\"Blue\"},\"pricing\":{\"original_price\":280,\"discount_price\":140,\"discount_rate\":\"50%\"}},\n  {\"id\":\"suit_17\",\"category\":\"suit\",\"title\":\"Gabardine-Double 17\",\"attributes\":{\"fabric\":\"100% Wool Gabardine\",\"weight\":\"300g/m²\",\"lapel\":\"Peak\",\"structure\":\"Double-Breasted 6x2\",\"buttons\":\"Metal Buttons\",\"color_hex\":\"#000080\",\"care\":\"Dry Clean Only\"},\"pricing\":{\"original_price\":4800,\"discount_price\":3840,\"discount_rate\":\"80%\"}},\n  {\"id\":\"bag_u3r1\",\"category\":\"bag\",\"title\":\"Suede-Tote-Lux 38\",\"attributes\":{\"material\":\"Soft Unstructured Suede\",\"dimensions\":\"410mm x 360mm x 160mm\",\"closure\":\"Magnetic Dot\",\"structure\":\"Single compartment\",\"logo\":\"Gold Chain Detail\",\"exterior_pockets\":\"None\",\"color\":\"Camel\"},\"pricing\":{\"original_price\":3200,\"discount_price\":1600,\"discount_rate\":\"50%\"}},\n  {\"id\":\"suit_13\",\"category\":\"suit\",\"title\":\"Tech-Triacetate 13\",\"attributes\":{\"fabric\":\"Triacetate / Polyester\",\"weight\":\"190g/m²\",\"lapel\":\"Notch\",\"structure\":\"Laser-Cut Edges / Unlined\",\"buttons\":\"Magnetic\",\"color_hex\":\"#2F4F4F\",\"care\":\"Hand Wash\"},\"pricing\":{\"original_price\":2200,\"discount_price\":1980,\"discount_rate\":\"90%\"}},\n  {\"id\":\"bag_v9k5\",\"category\":\"bag\",\"title\":\"Brief-Canvas-Distressed 39\",\"attributes\":{\"material\":\"Faded Canvas / Oiled Leather\",\"dimensions\":\"355mm x 265mm x 75mm\",\"closure\":\"Metal Buckles\",\"structure\":\"Dual front pockets\",\"logo\":\"None\",\"exterior_pockets\":\"None\",\"color\":\"Faded Brown\"},\"pricing\":{\"original_price\":900,\"discount_price\":450,\"discount_rate\":\"50%\"}},\n  {\"id\":\"suit_12\",\"category\":\"suit\",\"title\":\"Velvet-Shawl 12\",\"attributes\":{\"fabric\":\"100% Cotton Velvet\",\"weight\":\"400g/m²\",\"lapel\":\"Shawl\",\"structure\":\"Fully Lined / Shoulder Pads\",\"buttons\":\"Single 1-Button\",\"color_hex\":\"#301934\",\"care\":\"Dry Clean Only\"},\"pricing\":{\"original_price\":2800,\"discount_price\":1400,\"discount_rate\":\"50%\"}},\n  {\"id\":\"bag_j2m4\",\"category\":\"bag\",\"title\":\"Accordion-File 40\",\"attributes\":{\"material\":\"Plastic / PP\",\"dimensions\":\"345mm x 255mm x 45mm\",\"closure\":\"Elastic Cord\",\"structure\":\"12-Tab Dividers\",\"logo\":\"Stationery Brand\",\"exterior_pockets\":\"None\",\"color\":\"Translucent Blue\"},\"pricing\":{\"original_price\":65,\"discount_price\":32,\"discount_rate\":\"50%\"}},\n  {\"id\":\"suit_30\",\"category\":\"suit\",\"title\":\"Super-110s Classic 30\",\"attributes\":{\"fabric\":\"Wool Blend (Super 110s)\",\"weight\":\"260g/m²\",\"lapel\":\"Notch\",\"structure\":\"Half-Canvas / Fully Lined\",\"buttons\":\"Single 2-Button\",\"color_hex\":\"#36454F\",\"care\":\"Dry Clean Only\"},\"pricing\":{\"original_price\":2880,\"discount_price\":2016,\"discount_rate\":\"70%\"}},\n  {\"id\":\"bag_w8n1\",\"category\":\"bag\",\"title\":\"City-Shopper 41\",\"attributes\":{\"material\":\"Thin PU\",\"dimensions\":\"370mm x 330mm x 110mm\",\"closure\":\"None\",\"structure\":\"Single open void\",\"logo\":\"None\",\"exterior_pockets\":\"None\",\"color\":\"Lavender\"},\"pricing\":{\"original_price\":320,\"discount_price\":160,\"discount_rate\":\"50%\"}},\n  {\"id\":\"bag_l5o9\",\"category\":\"bag\",\"title\":\"Camera-Box 42\",\"attributes\":{\"material\":\"Padded Canvas\",\"dimensions\":\"240mm x 190mm x 130mm\",\"closure\":\"Zip\",\"structure\":\"Padded Dividers\",\"logo\":\"Camera Brand\",\"exterior_pockets\":\"Side Pockets\",\"color\":\"Black\"},\"pricing\":{\"original_price\":780,\"discount_price\":390,\"discount_rate\":\"50%\"}},\n  {\"id\":\"bag_k2x7\",\"category\":\"bag\",\"title\":\"Messenger-Comic 43\",\"attributes\":{\"material\":\"Thin Canvas\",\"dimensions\":\"300mm x 230mm x 80mm\",\"closure\":\"Flap\",\"structure\":\"None\",\"logo\":\"Anime Print\",\"exterior_pockets\":\"Front Pocket\",\"color\":\"Multi-color\"},\"pricing\":{\"original_price\":260,\"discount_price\":130,\"discount_rate\":\"50%\"}},\n  {\"id\":\"bag_m4z1\",\"category\":\"bag\",\"title\":\"Travel-Shell 44\",\"attributes\":{\"material\":\"Polycarbonate\",\"dimensions\":\"360mm x 540mm x 240mm\",\"closure\":\"TSA Lock\",\"structure\":\"Cross Straps\",\"logo\":\"Brand Badge\",\"exterior_pockets\":\"None\",\"color\":\"Silver\"},\"pricing\":{\"original_price\":1500,\"discount_price\":750,\"discount_rate\":\"50%\"}},\n  {\"id\":\"bag_r3c8\",\"category\":\"bag\",\"title\":\"Work-Brief-Basic 45\",\"attributes\":{\"material\":\"Polyester\",\"dimensions\":\"360mm x 280mm x 50mm\",\"closure\":\"Zipper\",\"structure\":\"Laptop Slot\",\"logo\":\"Corporate Logo\",\"exterior_pockets\":\"1 Front Zip\",\"color\":\"Navy\"},\"pricing\":{\"original_price\":180,\"discount_price\":90,\"discount_rate\":\"50%\"}},\n  {\"id\":\"bag_y2k9\",\"category\":\"bag\",\"title\":\"Organizer-Sleeve 46\",\"attributes\":{\"material\":\"Microfiber\",\"dimensions\":\"335mm x 245mm x 10mm\",\"closure\":\"Flap\",\"structure\":\"None\",\"logo\":\"None\",\"exterior_pockets\":\"None\",\"color\":\"Grey\"},\"pricing\":{\"original_price\":120,\"discount_price\":60,\"discount_rate\":\"50%\"}},\n  {\"id\":\"bag_h1v2\",\"category\":\"bag\",\"title\":\"Retro-Pilot 47\",\"attributes\":{\"material\":\"Bonded Leather (Soft)\",\"dimensions\":\"430mm x 320mm x 200mm\",\"closure\":\"Top Hinged Flap\",\"structure\":\"Internal Dividers\",\"logo\":\"None\",\"exterior_pockets\":\"Side Flap Pocket\",\"color\":\"Black\"},\"pricing\":{\"original_price\":1600,\"discount_price\":800,\"discount_rate\":\"50%\"}}\n]}}", "hidden_intents": ["Shortlist briefcases that keep A4 papers flat so corners do not fold or crease.", "Shortlist briefcases with a layered index-style interior or clearly separated compartments for documents and accessories.", "Shortlist briefcases that can hold a slim laptop of roughly 14 inches.", "Prefer de-logoed or understated branding—no loud or conspicuous logos.", "Prioritize structured stiff leather and high-density nylon with leather trim as primary material families.", "Select exactly five best-fit briefcases from the payload.", "Present results in a readable table with brand/series, dimensions, material and key features, original and discounted price, discount rate, and a short rationale."], "hidden_intent_count": 7, "has_objectives": true, "objectives_json": "{\"checklist\": [{\"criterion\": \"Recommends Heritage-Grain 01 (bag_m1v8) and states dimensions 385 x 290 x 105, material Saffiano Calfskin (Stiff), pricing ¥3,600 / ¥2,520, and a 70% discount rate.\"}, {\"criterion\": \"Gives a rationale for Heritage-Grain 01 (bag_m1v8): stiff structure resists sagging when partly empty; suited to formal legal settings where a credible, professional look matters. Substantially similar wording is acceptable.\"}, {\"criterion\": \"Recommends Commute-Shield X (bag_r7x2) and states dimensions 405 x 305 x 115, material 1680D Ballistic Nylon, pricing ¥2,400 / ¥1,680, and a 70% discount rate.\"}, {\"criterion\": \"Gives a rationale for Commute-Shield X (bag_r7x2): built for Shanghai commuting and rainy weather; mentions waterproof zips, laptop protection, and quick-access transit-card use. Substantially similar wording is acceptable.\"}, {\"criterion\": \"Recommends Urban-Logic T (bag_k8q1) and states dimensions 375 x 285 x 95, material Full-Grain Box Calf (Hard), pricing ¥4,200 / ¥3,360, and an 80% discount rate.\"}, {\"criterion\": \"Gives a rationale for Urban-Logic T (bag_k8q1): double gusset and document divider organize due-diligence or review binders; taupe tone pairs discreetly with business attire. Substantially similar wording is acceptable.\"}, {\"criterion\": \"Recommends Armor-Guard G2 (bag_x9l4) and states dimensions 395 x 295 x 110, material High-Density Water-Repellent Nylon, pricing ¥1,950 / ¥1,365, and a 70% discount rate.\"}, {\"criterion\": \"Gives a rationale for Armor-Guard G2 (bag_x9l4): internal index system separates statutes, evidence lists, and memos; hidden zips add a buffer for sensitive client materials. Substantially similar wording is acceptable.\"}, {\"criterion\": \"Recommends Senior-Focus Classic (bag_u7m5) and states dimensions 380 x 285 x 100, material Palmellato Leather (Structured), pricing ¥4,800 / ¥3,840, and an 80% discount rate.\"}, {\"criterion\": \"Gives a rationale for Senior-Focus Classic (bag_u7m5): quiet-luxury look aligned with a premium firm culture; structured body keeps A4 sheets flat and fits a 14-inch-class device. Substantially similar wording is acceptable.\"}, {\"criterion\": \"Uses a readable table that includes brand or series, dimensions, material and key features, original and sale price, discount rate, and rationale. Substantially similar wording is acceptable.\"}]}", "metadata_json": "{\"difficulty\": \"easy\"}", "yaml_path": "data/law_trainee/tasks/law_trainee_task_017/task.yaml", "yaml": "schema_version: v2\ntask_id: law_trainee_task_017\nuser_id: law_trainee\nenvironment_id: env_law_trainee\n\ntitle: \"Environment Trigger: Pick Five Core Work-Bag Recommendations from Comprehensive Gear Update\"\ndescription: \"Simulate an environment push from shanghai_career_nexus. From a mixed bag/suit update stream, identify and recommend the five core bag items that best match law_trainee’s practical legal-intern commute and document-carry needs, with concise rationale and pricing summary.\"\n\ntask_type: short_term_independent\n\ntrigger:\n  type: environment\n\nintent:\n  initial_input: |\n    {\"source\":\"shanghai_career_nexus\",\"event\":\"comprehensive_gear_update\",\"target\":\"agent\",\"payload\":{\"items\":[\n      {\"id\":\"bag_z9k2\",\"category\":\"bag\",\"title\":\"Neon-Pulse 40\",\"attributes\":{\"material\":\"Translucent PVC\",\"dimensions\":\"300mm x 400mm x 120mm\",\"closure\":\"Zipper\",\"structure\":\"Single compartment\",\"logo\":\"Oversized Reflective\",\"exterior_pockets\":\"Front Zip\",\"color\":\"Fluorescent Green\"},\"pricing\":{\"original_price\":399,\"discount_price\":199,\"discount_rate\":\"50%\"}},\n      {\"id\":\"suit_14\",\"category\":\"suit\",\"title\":\"Glen-Check Blend 14\",\"attributes\":{\"fabric\":\"Wool / Polyester Blend\",\"weight\":\"280g/m²\",\"lapel\":\"Notch\",\"structure\":\"Half-Lined\",\"buttons\":\"Single 2-Button\",\"color_hex\":\"#D3D3D3\",\"care\":\"Dry Clean Only\"},\"pricing\":{\"original_price\":1800,\"discount_price\":900,\"discount_rate\":\"50%\"}},\n      {\"id\":\"bag_p4n1\",\"category\":\"bag\",\"title\":\"Mini-Lux 06\",\"attributes\":{\"material\":\"Velvet\",\"dimensions\":\"180mm x 120mm x 50mm\",\"closure\":\"Clasp\",\"structure\":\"None\",\"logo\":\"Gold Monogram\",\"exterior_pockets\":\"None\",\"color\":\"Midnight Black\"},\"pricing\":{\"original_price\":4500,\"discount_price\":3600,\"discount_rate\":\"80%\"}},\n      {\"id\":\"suit_05\",\"category\":\"suit\",\"title\":\"Tweed-Heavy 05\",\"attributes\":{\"fabric\":\"100% Harris Tweed\",\"weight\":\"450g/m²\",\"lapel\":\"Notch\",\"structure\":\"Heavy Padded / Fully Lined\",\"buttons\":\"Single 3-Button\",\"color_hex\":\"#556B2F\",\"care\":\"Dry Clean Only\"},\"pricing\":{\"original_price\":3500,\"discount_price\":1750,\"discount_rate\":\"50%\"}},\n      {\"id\":\"bag_m1v8\",\"category\":\"bag\",\"title\":\"Heritage-Grain 01\",\"attributes\":{\"material\":\"Saffiano Calfskin (Stiff)\",\"dimensions\":\"385mm x 290mm x 105mm\",\"closure\":\"Top-Edge Zipper\",\"structure\":\"3 Main Compartments + 1 U-Shield Slot + 2 Pen Loops\",\"logo\":\"None (Interior Embossed)\",\"exterior_pockets\":\"1 Rear Magnetic Slot\",\"color\":\"Navy Blue\"},\"pricing\":{\"original_price\":3600,\"discount_price\":2520,\"discount_rate\":\"70%\"}},\n      {\"id\":\"suit_22\",\"category\":\"suit\",\"title\":\"Minimal-Zip 22\",\"attributes\":{\"fabric\":\"Polyester / Elastane\",\"weight\":\"230g/m²\",\"lapel\":\"Peak\",\"structure\":\"Hidden Seams / Fused\",\"buttons\":\"None (Zip)\",\"color_hex\":\"#FFFFFF\",\"care\":\"Machine Washable\"},\"pricing\":{\"original_price\":1100,\"discount_price\":550,\"discount_rate\":\"50%\"}},\n      {\"id\":\"bag_b3w9\",\"category\":\"bag\",\"title\":\"Tote-Open 07\",\"attributes\":{\"material\":\"Canvas\",\"dimensions\":\"450mm x 350mm x 180mm\",\"closure\":\"Open Top\",\"structure\":\"None\",\"logo\":\"Printed Text\",\"exterior_pockets\":\"None\",\"color\":\"Beige\"},\"pricing\":{\"original_price\":450,\"discount_price\":225,\"discount_rate\":\"50%\"}},\n      {\"id\":\"suit_01\",\"category\":\"suit\",\"title\":\"Merino-Wool Notch 01\",\"attributes\":{\"fabric\":\"100% Merino Wool\",\"weight\":\"240g/m²\",\"lapel\":\"Notch\",\"structure\":\"Full-Canvas / Fully Lined\",\"buttons\":\"Single 2-Button\",\"color_hex\":\"#191970\",\"care\":\"Dry Clean Only\"},\"pricing\":{\"original_price\":4280,\"discount_price\":2996,\"discount_rate\":\"70%\"}},\n      {\"id\":\"bag_h5t2\",\"category\":\"bag\",\"title\":\"Executive-Box 23\",\"attributes\":{\"material\":\"Aluminum Shell\",\"dimensions\":\"450mm x 330mm x 120mm\",\"closure\":\"Dual Combination Lock\",\"structure\":\"File Organizer Straps\",\"logo\":\"None\",\"exterior_pockets\":\"None\",\"color\":\"Silver\"},\"pricing\":{\"original_price\":2800,\"discount_price\":2240,\"discount_rate\":\"80%\"}},\n      {\"id\":\"suit_09\",\"category\":\"suit\",\"title\":\"Twill-Cotton 09\",\"attributes\":{\"fabric\":\"100% Cotton Twill\",\"weight\":\"320g/m²\",\"lapel\":\"Notch\",\"structure\":\"Garment Dyed / Unlined\",\"buttons\":\"Single 3-Button\",\"color_hex\":\"#A52A2A\",\"care\":\"Machine Wash\"},\"pricing\":{\"original_price\":1100,\"discount_price\":550,\"discount_rate\":\"50%\"}},\n      {\"id\":\"bag_r7x2\",\"category\":\"bag\",\"title\":\"Commute-Shield X\",\"attributes\":{\"material\":\"1680D Ballistic Nylon / Leather Trim\",\"dimensions\":\"405mm x 305mm x 115mm\",\"closure\":\"YKK Waterproof Zipper\",\"structure\":\"Dedicated 14\\\" Laptop Sleeve (High-Density Foam) + 5 Organizer Pockets\",\"logo\":\"Subtle Tonal Plate\",\"exterior_pockets\":\"1 Front Quick-Access Card Slot\",\"color\":\"Charcoal Black\"},\"pricing\":{\"original_price\":2400,\"discount_price\":1680,\"discount_rate\":\"70%\"}},\n      {\"id\":\"suit_12\",\"category\":\"suit\",\"title\":\"Velvet-Shawl 12\",\"attributes\":{\"fabric\":\"100% Cotton Velvet\",\"weight\":\"400g/m²\",\"lapel\":\"Shawl\",\"structure\":\"Fully Lined / Shoulder Pads\",\"buttons\":\"Single 1-Button\",\"color_hex\":\"#301934\",\"care\":\"Dry Clean Only\"},\"pricing\":{\"original_price\":2800,\"discount_price\":1400,\"discount_rate\":\"50%\"}},\n      {\"id\":\"bag_f2s8\",\"category\":\"bag\",\"title\":\"Slouchy-Hobo 15\",\"attributes\":{\"material\":\"Soft Lambskin\",\"dimensions\":\"400mm x 380mm x 40mm\",\"closure\":\"Magnetic Dot\",\"structure\":\"Single Internal Pocket\",\"logo\":\"Leather Tassel\",\"exterior_pockets\":\"None\",\"color\":\"Burgundy\"},\"pricing\":{\"original_price\":5800,\"discount_price\":2900,\"discount_rate\":\"50%\"}},\n      {\"id\":\"suit_28\",\"category\":\"suit\",\"title\":\"Poplin-Summer 28\",\"attributes\":{\"fabric\":\"100% Cotton Poplin\",\"weight\":\"110g/m²\",\"lapel\":\"Notch\",\"structure\":\"Unlined / Soft Shoulder\",\"buttons\":\"Single 2-Button\",\"color_hex\":\"#ADD8E6\",\"care\":\"Machine Washable\"},\"pricing\":{\"original_price\":650,\"discount_price\":455,\"discount_rate\":\"70%\"}},\n      {\"id\":\"bag_k8q1\",\"category\":\"bag\",\"title\":\"Urban-Logic T\",\"attributes\":{\"material\":\"Full-Grain Box Calf (Hard)\",\"dimensions\":\"375mm x 285mm x 95mm\",\"closure\":\"Top-Edge Zipper\",\"structure\":\"Double Gusset + Document Divider + Laptop Padding\",\"logo\":\"None\",\"exterior_pockets\":\"1 Hidden Side Slot\",\"color\":\"Taupe\"},\"pricing\":{\"original_price\":4200,\"discount_price\":3360,\"discount_rate\":\"80%\"}},\n      {\"id\":\"suit_07\",\"category\":\"suit\",\"title\":\"Cotton-Seersucker 07\",\"attributes\":{\"fabric\":\"100% Cotton Seersucker\",\"weight\":\"130g/m²\",\"lapel\":\"Notch\",\"structure\":\"Unlined / Striped Pattern\",\"buttons\":\"Single 2-Button\",\"color_hex\":\"#ADD8E6\",\"care\":\"Machine Washable\"},\"pricing\":{\"original_price\":980,\"discount_price\":588,\"discount_rate\":\"60%\"}},\n      {\"id\":\"bag_x9l4\",\"category\":\"bag\",\"title\":\"Armor-Guard G2\",\"attributes\":{\"material\":\"High-Density Water-Repellent Nylon\",\"dimensions\":\"395mm x 295mm x 110mm\",\"closure\":\"Hidden Zip Closure\",\"structure\":\"12-Point Internal Index System (U-Shield/Files/Tech)\",\"logo\":\"None\",\"exterior_pockets\":\"Dual Side Transit-Card Slots\",\"color\":\"Dark Grey\"},\"pricing\":{\"original_price\":1950,\"discount_price\":1365,\"discount_rate\":\"70%\"}},\n      {\"id\":\"suit_30\",\"category\":\"suit\",\"title\":\"Super-110s Classic 30\",\"attributes\":{\"fabric\":\"Wool Blend (Super 110s)\",\"weight\":\"260g/m²\",\"lapel\":\"Notch\",\"structure\":\"Half-Canvas / Fully Lined\",\"buttons\":\"Single 2-Button\",\"color_hex\":\"#36454F\",\"care\":\"Dry Clean Only\"},\"pricing\":{\"original_price\":2880,\"discount_price\":2016,\"discount_rate\":\"70%\"}},\n      {\"id\":\"bag_u7m5\",\"category\":\"bag\",\"title\":\"Senior-Focus Classic\",\"attributes\":{\"material\":\"Palmellato Leather (Structured)\",\"dimensions\":\"380mm x 285mm x 100mm\",\"closure\":\"Full-Length Top Zipper\",\"structure\":\"Padded 14\\\" Section + 2 Document Slots + Accessory Loop\",\"logo\":\"None\",\"exterior_pockets\":\"1 Vertical Quick-Access Zip\",\"color\":\"Elephant Grey\"},\"pricing\":{\"original_price\":4800,\"discount_price\":3840,\"discount_rate\":\"80%\"}},\n      {\"id\":\"suit_18\",\"category\":\"suit\",\"title\":\"Denim-Indigo 18\",\"attributes\":{\"fabric\":\"100% Cotton Denim\",\"weight\":\"12oz\",\"lapel\":\"Notch\",\"structure\":\"Contrast Stitching / Unlined\",\"buttons\":\"Copper Rivet\",\"color_hex\":\"#1E90FF\",\"care\":\"Machine Wash Cold\"},\"pricing\":{\"original_price\":880,\"discount_price\":440,\"discount_rate\":\"50%\"}},\n      {\"id\":\"bag_a4v1\",\"category\":\"bag\",\"title\":\"Retro-Satchel 09\",\"attributes\":{\"material\":\"Distressed Canvas\",\"dimensions\":\"320mm x 220mm x 60mm\",\"closure\":\"Brass Buckle\",\"structure\":\"Single pocket\",\"logo\":\"None\",\"exterior_pockets\":\"Front Flap Pocket\",\"color\":\"Khaki\"},\"pricing\":{\"original_price\":650,\"discount_price\":325,\"discount_rate\":\"50%\"}},\n      {\"id\":\"suit_03\",\"category\":\"suit\",\"title\":\"Linen-Vent 03\",\"attributes\":{\"fabric\":\"100% Linen\",\"weight\":\"150g/m²\",\"lapel\":\"Notch\",\"structure\":\"Unlined / No Shoulder Pads\",\"buttons\":\"Single 2-Button\",\"color_hex\":\"#F5F5DC\",\"care\":\"Hand Wash\"},\"pricing\":{\"original_price\":1580,\"discount_price\":1264,\"discount_rate\":\"80%\"}},\n      {\"id\":\"bag_d3p2\",\"category\":\"bag\",\"title\":\"Logo-Max 10\",\"attributes\":{\"material\":\"PVC Coated Canvas\",\"dimensions\":\"380mm x 300mm x 150mm\",\"closure\":\"None (Open)\",\"structure\":\"Pouch inside\",\"logo\":\"All-over Monogram Pattern\",\"exterior_pockets\":\"None\",\"color\":\"Brown/Tan\"},\"pricing\":{\"original_price\":11500,\"discount_price\":10350,\"discount_rate\":\"90%\"}},\n      {\"id\":\"suit_13\",\"category\":\"suit\",\"title\":\"Tech-Triacetate 13\",\"attributes\":{\"fabric\":\"Triacetate / Polyester\",\"weight\":\"190g/m²\",\"lapel\":\"Notch\",\"structure\":\"Laser-Cut Edges / Unlined\",\"buttons\":\"Magnetic\",\"color_hex\":\"#2F4F4F\",\"care\":\"Hand Wash\"},\"pricing\":{\"original_price\":2200,\"discount_price\":1980,\"discount_rate\":\"90%\"}},\n      {\"id\":\"bag_j6r9\",\"category\":\"bag\",\"title\":\"Document-Sleeve 21\",\"attributes\":{\"material\":\"Felt\",\"dimensions\":\"330mm x 240mm x 5mm\",\"closure\":\"Elastic Band\",\"structure\":\"None\",\"logo\":\"None\",\"exterior_pockets\":\"None\",\"color\":\"Grey\"},\"pricing\":{\"original_price\":99,\"discount_price\":49,\"discount_rate\":\"50%\"}},\n      {\"id\":\"suit_25\",\"category\":\"suit\",\"title\":\"Pure-Cashmere 25\",\"attributes\":{\"fabric\":\"100% Cashmere\",\"weight\":\"350g/m²\",\"lapel\":\"Notch\",\"structure\":\"Hand-Stitched / Silk Lined\",\"buttons\":\"Horn Buttons\",\"color_hex\":\"#F5DEB3\",\"care\":\"Dry Clean Only\"},\"pricing\":{\"original_price\":18000,\"discount_price\":14400,\"discount_rate\":\"80%\"}},\n      {\"id\":\"bag_e1w2\",\"category\":\"bag\",\"title\":\"Gym-Carry 12\",\"attributes\":{\"material\":\"Nylon\",\"dimensions\":\"500mm x 300mm x 300mm\",\"closure\":\"Zipper\",\"structure\":\"Separate Shoe Compartment\",\"logo\":\"Large Brand Print\",\"exterior_pockets\":\"Mesh\",\"color\":\"Black\"},\"pricing\":{\"original_price\":500,\"discount_price\":250,\"discount_rate\":\"50%\"}},\n      {\"id\":\"suit_02\",\"category\":\"suit\",\"title\":\"Poly-Basic Notch 02\",\"attributes\":{\"fabric\":\"100% Polyester\",\"weight\":\"210g/m²\",\"lapel\":\"Notch\",\"structure\":\"Fused / Half Lined\",\"buttons\":\"Single 1-Button\",\"color_hex\":\"#000000\",\"care\":\"Machine Washable\"},\"pricing\":{\"original_price\":899,\"discount_price\":449,\"discount_rate\":\"50%\"}},\n      {\"id\":\"bag_y8c3\",\"category\":\"bag\",\"title\":\"Artist-Folder 14\",\"attributes\":{\"material\":\"Polypropylene Shell\",\"dimensions\":\"600mm x 450mm x 25mm\",\"closure\":\"Snap Clip\",\"structure\":\"Single Slot\",\"logo\":\"None\",\"exterior_pockets\":\"Hand-hold only\",\"color\":\"Clear\"},\"pricing\":{\"original_price\":120,\"discount_price\":96,\"discount_rate\":\"80%\"}},\n      {\"id\":\"suit_11\",\"category\":\"suit\",\"title\":\"Pinstripe-Cashmere 11\",\"attributes\":{\"fabric\":\"95% Wool, 5% Cashmere\",\"weight\":\"260g/m²\",\"lapel\":\"Notch\",\"structure\":\"Half-Canvas / Fully Lined\",\"buttons\":\"Single 2-Button\",\"color_hex\":\"#696969\",\"care\":\"Dry Clean Only\"},\"pricing\":{\"original_price\":5500,\"discount_price\":3300,\"discount_rate\":\"60%\"}},\n      {\"id\":\"bag_g4k8\",\"category\":\"bag\",\"title\":\"Utility-Vest 16\",\"attributes\":{\"material\":\"Mesh / Nylon\",\"dimensions\":\"Adjustable\",\"closure\":\"Front Zipper\",\"structure\":\"Multi-Pocket Modular\",\"logo\":\"Military Patch\",\"exterior_pockets\":\"8 tactical pockets\",\"color\":\"Olive Green\"},\"pricing\":{\"original_price\":850,\"discount_price\":425,\"discount_rate\":\"50%\"}},\n      {\"id\":\"suit_20\",\"category\":\"suit\",\"title\":\"Winter-Angora 20\",\"attributes\":{\"fabric\":\"Wool / Angora Blend\",\"weight\":\"500g/m²\",\"lapel\":\"Notch\",\"structure\":\"Insulated Lining\",\"buttons\":\"Single 3-Button\",\"color_hex\":\"#000000\",\"care\":\"Professional Dry Clean\"},\"pricing\":{\"original_price\":7200,\"discount_price\":3600,\"discount_rate\":\"50%\"}},\n      {\"id\":\"bag_i2o9\",\"category\":\"bag\",\"title\":\"Bucket-Trend 22\",\"attributes\":{\"material\":\"Soft Calfskin\",\"dimensions\":\"210mm x 240mm x 210mm\",\"closure\":\"Drawstring\",\"structure\":\"None\",\"logo\":\"Gold Hardware\",\"exterior_pockets\":\"None\",\"color\":\"Tan\"},\"pricing\":{\"original_price\":3800,\"discount_price\":1900,\"discount_rate\":\"50%\"}},\n      {\"id\":\"suit_27\",\"category\":\"suit\",\"title\":\"Shadow-Stripe 27\",\"attributes\":{\"fabric\":\"100% Wool\",\"weight\":\"250g/m²\",\"lapel\":\"Notch\",\"structure\":\"Fully Lined / Shadow Pattern\",\"buttons\":\"Single 2-Button\",\"color_hex\":\"#121212\",\"care\":\"Dry Clean Only\"},\"pricing\":{\"original_price\":3800,\"discount_price\":2660,\"discount_rate\":\"70%\"}},\n      {\"id\":\"bag_q7f1\",\"category\":\"bag\",\"title\":\"Eco-Paper 24\",\"attributes\":{\"material\":\"Recycled Paper Fiber\",\"dimensions\":\"300mm x 400mm x 100mm\",\"closure\":\"None\",\"structure\":\"None\",\"logo\":\"Retail Logo\",\"exterior_pockets\":\"None\",\"color\":\"Brown\"},\"pricing\":{\"original_price\":15,\"discount_price\":7,\"discount_rate\":\"50%\"}},\n      {\"id\":\"suit_04\",\"category\":\"suit\",\"title\":\"Silk-Blend Peak 04\",\"attributes\":{\"fabric\":\"70% Silk, 30% Wool\",\"weight\":\"200g/m²\",\"lapel\":\"Peak\",\"structure\":\"Half-Canvas / Fully Lined\",\"buttons\":\"Double-Breasted 4-Button\",\"color_hex\":\"#E6E6FA\",\"care\":\"Dry Clean Only\"},\"pricing\":{\"original_price\":6800,\"discount_price\":5440,\"discount_rate\":\"80%\"}},\n      {\"id\":\"bag_s2a6\",\"category\":\"bag\",\"title\":\"Chain-Wallet 26\",\"attributes\":{\"material\":\"Patent Leather\",\"dimensions\":\"185mm x 105mm x 35mm\",\"closure\":\"Snap\",\"structure\":\"Internal Card Slots\",\"logo\":\"Diamond-cut Hardware\",\"exterior_pockets\":\"None\",\"color\":\"Gold\"},\"pricing\":{\"original_price\":4200,\"discount_price\":2100,\"discount_rate\":\"50%\"}},\n      {\"id\":\"suit_19\",\"category\":\"suit\",\"title\":\"Crepe-Viscose 19\",\"attributes\":{\"fabric\":\"100% Viscose Crepe\",\"weight\":\"220g/m²\",\"lapel\":\"Slim Notch\",\"structure\":\"Draped Cut / Unlined\",\"buttons\":\"Single 1-Button\",\"color_hex\":\"#F0E68C\",\"care\":\"Hand Wash\"},\"pricing\":{\"original_price\":1350,\"discount_price\":675,\"discount_rate\":\"50%\"}},\n      {\"id\":\"bag_t9v3\",\"category\":\"bag\",\"title\":\"Laptop-Skin 27\",\"attributes\":{\"material\":\"Neoprene\",\"dimensions\":\"345mm x 245mm x 22mm\",\"closure\":\"Zipper\",\"structure\":\"None\",\"logo\":\"Pattern Print\",\"exterior_pockets\":\"None\",\"color\":\"Floral\"},\"pricing\":{\"original_price\":180,\"discount_price\":90,\"discount_rate\":\"50%\"}},\n      {\"id\":\"suit_24\",\"category\":\"suit\",\"title\":\"Active-Stretch 24\",\"attributes\":{\"fabric\":\"80% Nylon, 20% Elastane\",\"weight\":\"170g/m²\",\"lapel\":\"Narrow Notch\",\"structure\":\"4-Way Stretch / Unlined\",\"buttons\":\"Rubberized\",\"color_hex\":\"#36454F\",\"care\":\"Machine Washable\"},\"pricing\":{\"original_price\":1480,\"discount_price\":740,\"discount_rate\":\"50%\"}},\n      {\"id\":\"bag_v3n8\",\"category\":\"bag\",\"title\":\"Rattan-Box 28\",\"attributes\":{\"material\":\"Woven Straw\",\"dimensions\":\"280mm x 180mm x 120mm\",\"closure\":\"Toggle\",\"structure\":\"None\",\"logo\":\"None\",\"exterior_pockets\":\"None\",\"color\":\"Natural\"},\"pricing\":{\"original_price\":950,\"discount_price\":475,\"discount_rate\":\"50%\"}},\n      {\"id\":\"suit_06\",\"category\":\"suit\",\"title\":\"Stretch-Viscose 06\",\"attributes\":{\"fabric\":\"65% Viscose, 30% Nylon, 5% Spandex\",\"weight\":\"280g/m²\",\"lapel\":\"None (Collarless)\",\"structure\":\"Slim Fit / Unlined\",\"buttons\":\"Hidden Snap\",\"color_hex\":\"#808080\",\"care\":\"Machine Wash Cold\"},\"pricing\":{\"original_price\":1200,\"discount_price\":840,\"discount_rate\":\"70%\"}},\n      {\"id\":\"bag_w1m4\",\"category\":\"bag\",\"title\":\"Brief-Slim 29\",\"attributes\":{\"material\":\"Synthetic PU\",\"dimensions\":\"355mm x 255mm x 25mm\",\"closure\":\"Zipper\",\"structure\":\"Single Document Slot\",\"logo\":\"Printed Logo\",\"exterior_pockets\":\"None\",\"color\":\"Blue\"},\"pricing\":{\"original_price\":380,\"discount_price\":190,\"discount_rate\":\"50%\"}},\n      {\"id\":\"suit_15\",\"category\":\"suit\",\"title\":\"Corduroy-Heavy 15\",\"attributes\":{\"fabric\":\"100% Cotton Corduroy\",\"weight\":\"380g/m²\",\"lapel\":\"Notch\",\"structure\":\"Fully Lined / Leather Buttons\",\"buttons\":\"Single 2-Button\",\"color_hex\":\"#D2B48C\",\"care\":\"Dry Clean Only\"},\"pricing\":{\"original_price\":1450,\"discount_price\":1015,\"discount_rate\":\"70%\"}},\n      {\"id\":\"bag_x5h2\",\"category\":\"bag\",\"title\":\"Denim-Tote 30\",\"attributes\":{\"material\":\"Denim Canvas\",\"dimensions\":\"480mm x 400mm x 150mm\",\"closure\":\"Button\",\"structure\":\"Large Void\",\"logo\":\"Denim Brand\",\"exterior_pockets\":\"Jean-style pockets\",\"color\":\"Indigo\"},\"pricing\":{\"original_price\":550,\"discount_price\":275,\"discount_rate\":\"50%\"}},\n      {\"id\":\"suit_23\",\"category\":\"suit\",\"title\":\"Houndstooth-Wool 23\",\"attributes\":{\"fabric\":\"Wool / Cotton Blend\",\"weight\":\"310g/m²\",\"lapel\":\"Notch\",\"structure\":\"Half-Canvas / Fully Lined\",\"buttons\":\"Single 2-Button\",\"color_hex\":\"#2F4F4F\",\"care\":\"Dry Clean Only\"},\"pricing\":{\"original_price\":2600,\"discount_price\":1820,\"discount_rate\":\"70%\"}},\n      {\"id\":\"bag_c2o9\",\"category\":\"bag\",\"title\":\"Quilted-Puffer 31\",\"attributes\":{\"material\":\"Padded Poly-Nylon\",\"dimensions\":\"380mm x 350mm x 120mm\",\"closure\":\"None (Open)\",\"structure\":\"None\",\"logo\":\"Brand Tag\",\"exterior_pockets\":\"None\",\"color\":\"Grey\"},\"pricing\":{\"original_price\":1200,\"discount_price\":600,\"discount_rate\":\"50%\"}},\n      {\"id\":\"suit_10\",\"category\":\"suit\",\"title\":\"Oversize-Poly 10\",\"attributes\":{\"fabric\":\"100% Polyester\",\"weight\":\"200g/m²\",\"lapel\":\"Wide Notch\",\"structure\":\"Extreme Drop Shoulder\",\"buttons\":\"Single 1-Button\",\"color_hex\":\"#FFC0CB\",\"care\":\"Machine Washable\"},\"pricing\":{\"original_price\":750,\"discount_price\":375,\"discount_rate\":\"50%\"}},\n      {\"id\":\"bag_k6l3\",\"category\":\"bag\",\"title\":\"Velcro-Brief 32\",\"attributes\":{\"material\":\"Thin Nylon\",\"dimensions\":\"370mm x 275mm x 30mm\",\"closure\":\"Velcro Strip\",\"structure\":\"None\",\"logo\":\"Corporate Gift Print\",\"exterior_pockets\":\"None\",\"color\":\"Black\"},\"pricing\":{\"original_price\":60,\"discount_price\":30,\"discount_rate\":\"50%\"}},\n      {\"id\":\"suit_21\",\"category\":\"suit\",\"title\":\"Bamboo-Eco 21\",\"attributes\":{\"fabric\":\"100% Bamboo Fiber\",\"weight\":\"190g/m²\",\"lapel\":\"Notch\",\"structure\":\"Unlined / Wood Buttons\",\"buttons\":\"Single 2-Button\",\"color_hex\":\"#E9967A\",\"care\":\"Machine Wash Gentle\"},\"pricing\":{\"original_price\":1600,\"discount_price\":800,\"discount_rate\":\"50%\"}},\n      {\"id\":\"bag_o9p4\",\"category\":\"bag\",\"title\":\"Faux-Fur 33\",\"attributes\":{\"material\":\"Synthetic Fur\",\"dimensions\":\"320mm x 260mm x 100mm\",\"closure\":\"Zip\",\"structure\":\"None\",\"logo\":\"None\",\"exterior_pockets\":\"None\",\"color\":\"Cream\"},\"pricing\":{\"original_price\":350,\"discount_price\":175,\"discount_rate\":\"50%\"}},\n      {\"id\":\"suit_16\",\"category\":\"suit\",\"title\":\"Tropical-Wool 16\",\"attributes\":{\"fabric\":\"100% Wool\",\"weight\":\"180g/m²\",\"lapel\":\"Notch\",\"structure\":\"Half-Lined / Natural Shoulder\",\"buttons\":\"Single 2-Button\",\"color_hex\":\"#708090\",\"care\":\"Dry Clean Only\"},\"pricing\":{\"original_price\":3200,\"discount_price\":1600,\"discount_rate\":\"50%\"}},\n      {\"id\":\"bag_z1y8\",\"category\":\"bag\",\"title\":\"Nano-Bag 34\",\"attributes\":{\"material\":\"Box Leather\",\"dimensions\":\"110mm x 90mm x 45mm\",\"closure\":\"Snap\",\"structure\":\"None\",\"logo\":\"Micro Metal Logo\",\"exterior_pockets\":\"None\",\"color\":\"Yellow\"},\"pricing\":{\"original_price\":2200,\"discount_price\":1100,\"discount_rate\":\"50%\"}},\n      {\"id\":\"suit_26\",\"category\":\"suit\",\"title\":\"Ripstop-Utility 26\",\"attributes\":{\"fabric\":\"Cotton / Nylon Ripstop\",\"weight\":\"210g/m²\",\"lapel\":\"Notch\",\"structure\":\"Multi-Pocket / Unlined\",\"buttons\":\"Press Studs\",\"color_hex\":\"#556B2F\",\"care\":\"Machine Washable\"},\"pricing\":{\"original_price\":1250,\"discount_price\":625,\"discount_rate\":\"50%\"}},\n      {\"id\":\"bag_n4x2\",\"category\":\"bag\",\"title\":\"Tube-Bag 35\",\"attributes\":{\"material\":\"Pebbled Leather\",\"dimensions\":\"300mm x 140mm x 140mm\",\"closure\":\"Zip\",\"structure\":\"Single compartment\",\"logo\":\"None\",\"exterior_pockets\":\"None\",\"color\":\"Black\"},\"pricing\":{\"original_price\":1800,\"discount_price\":900,\"discount_rate\":\"50%\"}},\n      {\"id\":\"suit_08\",\"category\":\"suit\",\"title\":\"Super-150s Peak 08\",\"attributes\":{\"fabric\":\"100% Wool (Super 150s)\",\"weight\":\"220g/m²\",\"lapel\":\"Peak (Satin)\",\"structure\":\"Full-Canvas / Silk Lining\",\"buttons\":\"Single 1-Button\",\"color_hex\":\"#000000\",\"care\":\"Professional Dry Clean\"},\"pricing\":{\"original_price\":12500,\"discount_price\":10000,\"discount_rate\":\"80%\"}},\n      {\"id\":\"bag_t2m9\",\"category\":\"bag\",\"title\":\"Graphic-Messenger 36\",\"attributes\":{\"material\":\"Heavy Duty Canvas\",\"dimensions\":\"380mm x 310mm x 90mm\",\"closure\":\"Large Flap\",\"structure\":\"Single Space\",\"logo\":\"Oversized Graphic\",\"exterior_pockets\":\"None\",\"color\":\"Olive\"},\"pricing\":{\"original_price\":450,\"discount_price\":225,\"discount_rate\":\"50%\"}},\n      {\"id\":\"suit_29\",\"category\":\"suit\",\"title\":\"Metallic-Lurex 29\",\"attributes\":{\"fabric\":\"Lurex / Polyester Blend\",\"weight\":\"240g/m²\",\"lapel\":\"Peak\",\"structure\":\"Shimmer Effect / Fully Lined\",\"buttons\":\"Single 1-Button\",\"color_hex\":\"#C0C0C0\",\"care\":\"Specialist Clean\"},\"pricing\":{\"original_price\":2400,\"discount_price\":1200,\"discount_rate\":\"50%\"}},\n      {\"id\":\"bag_f5h2\",\"category\":\"bag\",\"title\":\"Sling-Pack 37\",\"attributes\":{\"material\":\"Ripstop Nylon\",\"dimensions\":\"210mm x 340mm x 85mm\",\"closure\":\"Zip\",\"structure\":\"Single\",\"logo\":\"Reflective Logo\",\"exterior_pockets\":\"Dual Mesh Side\",\"color\":\"Blue\"},\"pricing\":{\"original_price\":280,\"discount_price\":140,\"discount_rate\":\"50%\"}},\n      {\"id\":\"suit_17\",\"category\":\"suit\",\"title\":\"Gabardine-Double 17\",\"attributes\":{\"fabric\":\"100% Wool Gabardine\",\"weight\":\"300g/m²\",\"lapel\":\"Peak\",\"structure\":\"Double-Breasted 6x2\",\"buttons\":\"Metal Buttons\",\"color_hex\":\"#000080\",\"care\":\"Dry Clean Only\"},\"pricing\":{\"original_price\":4800,\"discount_price\":3840,\"discount_rate\":\"80%\"}},\n      {\"id\":\"bag_u3r1\",\"category\":\"bag\",\"title\":\"Suede-Tote-Lux 38\",\"attributes\":{\"material\":\"Soft Unstructured Suede\",\"dimensions\":\"410mm x 360mm x 160mm\",\"closure\":\"Magnetic Dot\",\"structure\":\"Single compartment\",\"logo\":\"Gold Chain Detail\",\"exterior_pockets\":\"None\",\"color\":\"Camel\"},\"pricing\":{\"original_price\":3200,\"discount_price\":1600,\"discount_rate\":\"50%\"}},\n      {\"id\":\"suit_13\",\"category\":\"suit\",\"title\":\"Tech-Triacetate 13\",\"attributes\":{\"fabric\":\"Triacetate / Polyester\",\"weight\":\"190g/m²\",\"lapel\":\"Notch\",\"structure\":\"Laser-Cut Edges / Unlined\",\"buttons\":\"Magnetic\",\"color_hex\":\"#2F4F4F\",\"care\":\"Hand Wash\"},\"pricing\":{\"original_price\":2200,\"discount_price\":1980,\"discount_rate\":\"90%\"}},\n      {\"id\":\"bag_v9k5\",\"category\":\"bag\",\"title\":\"Brief-Canvas-Distressed 39\",\"attributes\":{\"material\":\"Faded Canvas / Oiled Leather\",\"dimensions\":\"355mm x 265mm x 75mm\",\"closure\":\"Metal Buckles\",\"structure\":\"Dual front pockets\",\"logo\":\"None\",\"exterior_pockets\":\"None\",\"color\":\"Faded Brown\"},\"pricing\":{\"original_price\":900,\"discount_price\":450,\"discount_rate\":\"50%\"}},\n      {\"id\":\"suit_12\",\"category\":\"suit\",\"title\":\"Velvet-Shawl 12\",\"attributes\":{\"fabric\":\"100% Cotton Velvet\",\"weight\":\"400g/m²\",\"lapel\":\"Shawl\",\"structure\":\"Fully Lined / Shoulder Pads\",\"buttons\":\"Single 1-Button\",\"color_hex\":\"#301934\",\"care\":\"Dry Clean Only\"},\"pricing\":{\"original_price\":2800,\"discount_price\":1400,\"discount_rate\":\"50%\"}},\n      {\"id\":\"bag_j2m4\",\"category\":\"bag\",\"title\":\"Accordion-File 40\",\"attributes\":{\"material\":\"Plastic / PP\",\"dimensions\":\"345mm x 255mm x 45mm\",\"closure\":\"Elastic Cord\",\"structure\":\"12-Tab Dividers\",\"logo\":\"Stationery Brand\",\"exterior_pockets\":\"None\",\"color\":\"Translucent Blue\"},\"pricing\":{\"original_price\":65,\"discount_price\":32,\"discount_rate\":\"50%\"}},\n      {\"id\":\"suit_30\",\"category\":\"suit\",\"title\":\"Super-110s Classic 30\",\"attributes\":{\"fabric\":\"Wool Blend (Super 110s)\",\"weight\":\"260g/m²\",\"lapel\":\"Notch\",\"structure\":\"Half-Canvas / Fully Lined\",\"buttons\":\"Single 2-Button\",\"color_hex\":\"#36454F\",\"care\":\"Dry Clean Only\"},\"pricing\":{\"original_price\":2880,\"discount_price\":2016,\"discount_rate\":\"70%\"}},\n      {\"id\":\"bag_w8n1\",\"category\":\"bag\",\"title\":\"City-Shopper 41\",\"attributes\":{\"material\":\"Thin PU\",\"dimensions\":\"370mm x 330mm x 110mm\",\"closure\":\"None\",\"structure\":\"Single open void\",\"logo\":\"None\",\"exterior_pockets\":\"None\",\"color\":\"Lavender\"},\"pricing\":{\"original_price\":320,\"discount_price\":160,\"discount_rate\":\"50%\"}},\n      {\"id\":\"bag_l5o9\",\"category\":\"bag\",\"title\":\"Camera-Box 42\",\"attributes\":{\"material\":\"Padded Canvas\",\"dimensions\":\"240mm x 190mm x 130mm\",\"closure\":\"Zip\",\"structure\":\"Padded Dividers\",\"logo\":\"Camera Brand\",\"exterior_pockets\":\"Side Pockets\",\"color\":\"Black\"},\"pricing\":{\"original_price\":780,\"discount_price\":390,\"discount_rate\":\"50%\"}},\n      {\"id\":\"bag_k2x7\",\"category\":\"bag\",\"title\":\"Messenger-Comic 43\",\"attributes\":{\"material\":\"Thin Canvas\",\"dimensions\":\"300mm x 230mm x 80mm\",\"closure\":\"Flap\",\"structure\":\"None\",\"logo\":\"Anime Print\",\"exterior_pockets\":\"Front Pocket\",\"color\":\"Multi-color\"},\"pricing\":{\"original_price\":260,\"discount_price\":130,\"discount_rate\":\"50%\"}},\n      {\"id\":\"bag_m4z1\",\"category\":\"bag\",\"title\":\"Travel-Shell 44\",\"attributes\":{\"material\":\"Polycarbonate\",\"dimensions\":\"360mm x 540mm x 240mm\",\"closure\":\"TSA Lock\",\"structure\":\"Cross Straps\",\"logo\":\"Brand Badge\",\"exterior_pockets\":\"None\",\"color\":\"Silver\"},\"pricing\":{\"original_price\":1500,\"discount_price\":750,\"discount_rate\":\"50%\"}},\n      {\"id\":\"bag_r3c8\",\"category\":\"bag\",\"title\":\"Work-Brief-Basic 45\",\"attributes\":{\"material\":\"Polyester\",\"dimensions\":\"360mm x 280mm x 50mm\",\"closure\":\"Zipper\",\"structure\":\"Laptop Slot\",\"logo\":\"Corporate Logo\",\"exterior_pockets\":\"1 Front Zip\",\"color\":\"Navy\"},\"pricing\":{\"original_price\":180,\"discount_price\":90,\"discount_rate\":\"50%\"}},\n      {\"id\":\"bag_y2k9\",\"category\":\"bag\",\"title\":\"Organizer-Sleeve 46\",\"attributes\":{\"material\":\"Microfiber\",\"dimensions\":\"335mm x 245mm x 10mm\",\"closure\":\"Flap\",\"structure\":\"None\",\"logo\":\"None\",\"exterior_pockets\":\"None\",\"color\":\"Grey\"},\"pricing\":{\"original_price\":120,\"discount_price\":60,\"discount_rate\":\"50%\"}},\n      {\"id\":\"bag_h1v2\",\"category\":\"bag\",\"title\":\"Retro-Pilot 47\",\"attributes\":{\"material\":\"Bonded Leather (Soft)\",\"dimensions\":\"430mm x 320mm x 200mm\",\"closure\":\"Top Hinged Flap\",\"structure\":\"Internal Dividers\",\"logo\":\"None\",\"exterior_pockets\":\"Side Flap Pocket\",\"color\":\"Black\"},\"pricing\":{\"original_price\":1600,\"discount_price\":800,\"discount_rate\":\"50%\"}}\n    ]}}\n\n  hidden_intent:\n    - content: \"Shortlist briefcases that keep A4 papers flat so corners do not fold or crease.\"\n    - content: \"Shortlist briefcases with a layered index-style interior or clearly separated compartments for documents and accessories.\"\n    - content: \"Shortlist briefcases that can hold a slim laptop of roughly 14 inches.\"\n    - content: \"Prefer de-logoed or understated branding—no loud or conspicuous logos.\"\n    - content: \"Prioritize structured stiff leather and high-density nylon with leather trim as primary material families.\"\n    - content: \"Select exactly five best-fit briefcases from the payload.\"\n    - content: \"Present results in a readable table with brand/series, dimensions, material and key features, original and discounted price, discount rate, and a short rationale.\"\nobjectives:\n  checklist:\n    - criterion: \"Recommends Heritage-Grain 01 (bag_m1v8) and states dimensions 385 x 290 x 105, material Saffiano Calfskin (Stiff), pricing ¥3,600 / ¥2,520, and a 70% discount rate.\"\n\n    - criterion: \"Gives a rationale for Heritage-Grain 01 (bag_m1v8): stiff structure resists sagging when partly empty; suited to formal legal settings where a credible, professional look matters. Substantially similar wording is acceptable.\"\n\n    - criterion: \"Recommends Commute-Shield X (bag_r7x2) and states dimensions 405 x 305 x 115, material 1680D Ballistic Nylon, pricing ¥2,400 / ¥1,680, and a 70% discount rate.\"\n\n    - criterion: \"Gives a rationale for Commute-Shield X (bag_r7x2): built for Shanghai commuting and rainy weather; mentions waterproof zips, laptop protection, and quick-access transit-card use. Substantially similar wording is acceptable.\"\n\n    - criterion: \"Recommends Urban-Logic T (bag_k8q1) and states dimensions 375 x 285 x 95, material Full-Grain Box Calf (Hard), pricing ¥4,200 / ¥3,360, and an 80% discount rate.\"\n\n    - criterion: \"Gives a rationale for Urban-Logic T (bag_k8q1): double gusset and document divider organize due-diligence or review binders; taupe tone pairs discreetly with business attire. Substantially similar wording is acceptable.\"\n\n    - criterion: \"Recommends Armor-Guard G2 (bag_x9l4) and states dimensions 395 x 295 x 110, material High-Density Water-Repellent Nylon, pricing ¥1,950 / ¥1,365, and a 70% discount rate.\"\n\n    - criterion: \"Gives a rationale for Armor-Guard G2 (bag_x9l4): internal index system separates statutes, evidence lists, and memos; hidden zips add a buffer for sensitive client materials. Substantially similar wording is acceptable.\"\n\n    - criterion: \"Recommends Senior-Focus Classic (bag_u7m5) and states dimensions 380 x 285 x 100, material Palmellato Leather (Structured), pricing ¥4,800 / ¥3,840, and an 80% discount rate.\"\n\n    - criterion: \"Gives a rationale for Senior-Focus Classic (bag_u7m5): quiet-luxury look aligned with a premium firm culture; structured body keeps A4 sheets flat and fits a 14-inch-class device. Substantially similar wording is acceptable.\"\n\n    - criterion: \"Uses a readable table that includes brand or series, dimensions, material and key features, original and sale price, discount rate, and rationale. Substantially similar wording is acceptable.\"\n\nmetadata:\n  difficulty: easy"}
{"schema_version": "v2", "role": "law_trainee", "task_id": "law_trainee_task_018", "user_id": "law_trainee", "environment_id": "env_law_trainee", "title": "Supplemental Filing Escalation Rehearsal with Cleanup Confirmation", "display_title": "Supplemental Filing Escalation Rehearsal with Cleanup Confirmation", "description": "Run one short Todoist rehearsal for missing supplemental filing items in the Yuehai guarantee dispute, clean the temporary board in the same pass, and send one completion email to internal coordination. Evaluation focuses on whether the blocker, escalation trigger, cleanup, and next checkpoint are explicit and easy to verify.", "task_type": "short_term_independent", "difficulty": "medium", "trigger_type": "user", "initial_input": "My Todoist email is stmcco@gmail.com and password is 'Ozz21Q4'.\nMy Gmail email is stmcco@gmail.com and password is 'n3v=I[w'.\n\nI'm tied up before tomorrow's internal Yuehai check-in, so I need a quick filing-chase rehearsal I can hand off cleanly.\n\nPlease run it in Todoist using the temporary board \"Temporary Filing Follow-up Board - Yuehai\", then clear the rehearsal artifacts in this same pass and send la-mcco@gmail.com an internal completion note with subject \"Yuehai Guarantee Dispute - Filing Rehearsal Cleanup Complete\".\nKeep this tidy: use one temporary Todoist tracking item for the chase, and make the completion email itself carry the cleanup confirmation plus the live-risk / next-checkpoint handoff, instead of relying on multiple follow-up emails if you can avoid it.\n\nThis pass should cover the live supplemental-filing chase items. Create one temporary Todoist tracking item that covers the full chase, then clean everything up and send the completion note.", "hidden_intents": ["In the temporary Todoist tracking item, cover these three chase items: the unsigned shareholder list, the signed guaranty annex comparison copy, and the courier receipt for the supplemental filing package.", "In the temporary Todoist tracking item, say that the unsigned shareholder list is the main blocker and that if it is still missing before tomorrow's internal check-in, it must be escalated to partner immediately.", "In the completion email, confirm that the temporary board was cleared, say that the unsigned shareholder list remains the live risk, give the next checkpoint for tomorrow morning, and ask the recipient to flag any delay immediately.", "Use one temporary Todoist tracking item that covers the full chase, rather than spreading the rehearsal across several temporary tracking items.", "Send the completion email in this same pass after clearing the rehearsal artifacts, instead of leaving the cleanup or handoff unfinished."], "hidden_intent_count": 5, "has_objectives": true, "objectives_json": "{\"checklist\": [{\"criterion\": \"The Todoist tracking item or same-thread completion follow-up clearly identifies the Yuehai guarantee dispute supplemental filing chase and covers all three chase items: unsigned shareholder list, signed guaranty annex comparison copy, and courier receipt for the supplemental filing package.\"}, {\"criterion\": \"The Todoist tracking item or same-thread completion follow-up states that the unsigned shareholder list is the main blocker and that if it is still missing before tomorrow's internal check-in, it must be escalated to partner immediately. Substantially similar wording is acceptable.\"}, {\"criterion\": \"The temporary Todoist board is removed by the end of the task.\"}, {\"criterion\": \"The sent Gmail subject is exactly 'Yuehai Guarantee Dispute - Filing Rehearsal Cleanup Complete'.\"}, {\"criterion\": \"The Gmail body explicitly confirms that the temporary Todoist board was cleared or removed.\"}, {\"criterion\": \"The completion email or same-thread completion follow-up states that the unsigned shareholder list remains the live risk. Substantially similar wording is acceptable.\"}, {\"criterion\": \"The completion email or same-thread completion follow-up gives a concrete next checkpoint for tomorrow morning. Substantially similar wording is acceptable.\"}, {\"criterion\": \"The completion email or same-thread completion follow-up asks the recipient to flag delay immediately. Substantially similar wording is acceptable.\"}], \"tools\": [{\"mcp_appworld_todoist__create_project\": {\"call\": [\"name\"], \"result\": [\"response.project_id\"]}}, {\"mcp_appworld_todoist__create_task\": {\"call\": [\"title\", \"project_id\", \"description\"], \"result\": [\"response.task_id\"]}}, {\"mcp_appworld_todoist__delete_project\": {\"call\": [\"project_id\"], \"result\": [\"response.message\"]}}, {\"mcp_appworld_gmail__send_email\": {\"call\": [\"email_addresses\", \"subject\", \"body\"], \"result\": [\"response.sent_email_thread_id\", \"response.sent_email_id\"]}}, {\"mcp_appworld_gmail__create_draft\": {\"call\": [\"recipient_email_addresses\", \"body\", \"belongs_to_email_thread_id\", \"response_to_email_id\"], \"result\": [\"response.draft_id\"]}}, {\"mcp_appworld_gmail__send_email_from_draft\": {\"call\": [\"draft_id\"], \"result\": [\"response.sent_email_thread_id\", \"response.sent_email_id\"]}}], \"tools_evaluation_path\": \"data/law_trainee/scripts/law_trainee_task_018_5.py\"}", "metadata_json": "{\"benchmark_role\": \"test\", \"difficulty\": \"medium\", \"preference_threads\": [\"audit_ready_work_tracking\", \"deadline_and_followup_discipline\", \"cross_app_preparation_workflow\"]}", "yaml_path": "data/law_trainee/tasks/law_trainee_task_018/task.yaml", "yaml": "schema_version: v2\nuser_id: law_trainee\nenvironment_id: env_law_trainee\ntask_type: short_term_independent\ntrigger:\n  type: user\ntask_id: law_trainee_task_018\ntitle: Supplemental Filing Escalation Rehearsal with Cleanup Confirmation\ndescription: \"Run one short Todoist rehearsal for missing supplemental filing items in the Yuehai guarantee dispute, clean the temporary board in the same pass, and send one completion email to internal coordination. Evaluation focuses on whether the blocker, escalation trigger, cleanup, and next checkpoint are explicit and easy to verify.\"\nintent:\n  initial_input: |-\n    My Todoist email is stmcco@gmail.com and password is 'Ozz21Q4'.\n    My Gmail email is stmcco@gmail.com and password is 'n3v=I[w'.\n\n    I'm tied up before tomorrow's internal Yuehai check-in, so I need a quick filing-chase rehearsal I can hand off cleanly.\n\n    Please run it in Todoist using the temporary board \"Temporary Filing Follow-up Board - Yuehai\", then clear the rehearsal artifacts in this same pass and send la-mcco@gmail.com an internal completion note with subject \"Yuehai Guarantee Dispute - Filing Rehearsal Cleanup Complete\".\n    Keep this tidy: use one temporary Todoist tracking item for the chase, and make the completion email itself carry the cleanup confirmation plus the live-risk / next-checkpoint handoff, instead of relying on multiple follow-up emails if you can avoid it.\n\n    This pass should cover the live supplemental-filing chase items. Create one temporary Todoist tracking item that covers the full chase, then clean everything up and send the completion note.\n\n  hidden_intent:\n    - content: \"In the temporary Todoist tracking item, cover these three chase items: the unsigned shareholder list, the signed guaranty annex comparison copy, and the courier receipt for the supplemental filing package.\"\n    - content: \"In the temporary Todoist tracking item, say that the unsigned shareholder list is the main blocker and that if it is still missing before tomorrow's internal check-in, it must be escalated to partner immediately.\"\n    - content: \"In the completion email, confirm that the temporary board was cleared, say that the unsigned shareholder list remains the live risk, give the next checkpoint for tomorrow morning, and ask the recipient to flag any delay immediately.\"\n    - content: \"Use one temporary Todoist tracking item that covers the full chase, rather than spreading the rehearsal across several temporary tracking items.\"\n    - content: \"Send the completion email in this same pass after clearing the rehearsal artifacts, instead of leaving the cleanup or handoff unfinished.\"\nobjectives:\n  tools:\n    - mcp_appworld_todoist__create_project:\n        call: [name]\n        result: [response.project_id]\n    - mcp_appworld_todoist__create_task:\n        call: [title, project_id, description]\n        result: [response.task_id]\n    - mcp_appworld_todoist__delete_project:\n        call: [project_id]\n        result: [response.message]\n    - mcp_appworld_gmail__send_email:\n        call: [email_addresses, subject, body]\n        result: [response.sent_email_thread_id, response.sent_email_id]\n    - mcp_appworld_gmail__create_draft:\n        call: [recipient_email_addresses, body, belongs_to_email_thread_id, response_to_email_id]\n        result: [response.draft_id]\n    - mcp_appworld_gmail__send_email_from_draft:\n        call: [draft_id]\n        result: [response.sent_email_thread_id, response.sent_email_id]\n  checklist:\n    - criterion: \"The Todoist tracking item or same-thread completion follow-up clearly identifies the Yuehai guarantee dispute supplemental filing chase and covers all three chase items: unsigned shareholder list, signed guaranty annex comparison copy, and courier receipt for the supplemental filing package.\"\n    - criterion: \"The Todoist tracking item or same-thread completion follow-up states that the unsigned shareholder list is the main blocker and that if it is still missing before tomorrow's internal check-in, it must be escalated to partner immediately. Substantially similar wording is acceptable.\"\n    - criterion: \"The temporary Todoist board is removed by the end of the task.\"\n    - criterion: \"The sent Gmail subject is exactly 'Yuehai Guarantee Dispute - Filing Rehearsal Cleanup Complete'.\"\n    - criterion: \"The Gmail body explicitly confirms that the temporary Todoist board was cleared or removed.\"\n    - criterion: \"The completion email or same-thread completion follow-up states that the unsigned shareholder list remains the live risk. Substantially similar wording is acceptable.\"\n    - criterion: \"The completion email or same-thread completion follow-up gives a concrete next checkpoint for tomorrow morning. Substantially similar wording is acceptable.\"\n    - criterion: \"The completion email or same-thread completion follow-up asks the recipient to flag delay immediately. Substantially similar wording is acceptable.\"\n  tools_evaluation_path: \"data/law_trainee/scripts/law_trainee_task_018_5.py\"\nmetadata:\n  difficulty: medium\n  benchmark_role: test\n  preference_threads:\n    - audit_ready_work_tracking\n    - deadline_and_followup_discipline\n    - cross_app_preparation_workflow"}
{"schema_version": "v2", "role": "law_trainee", "task_id": "law_trainee_task_019", "user_id": "law_trainee", "environment_id": "env_law_trainee", "title": "Traffic-Accident Offense: Draft a Criminal Defense Brief or Statement of Defense", "display_title": "Traffic-Accident Offense: Draft a Criminal Defense Brief or Statement of Defense", "description": "Based on the prosecution theory and record excerpts in Defense_Case_Summary.md, draft a first-pass defense opinion or statement of defense for Defendant A on the charge of causing a traffic accident or related negligence offenses for filing with the court. Evaluation focuses on pleading format, legal diction, the evidentiary gap from the absence of a Traffic Accident Liability Determination, an elements-based argument structure, and synthesis of facts favorable to the accused.", "task_type": "long_term", "difficulty": "hard", "trigger_type": "user", "initial_input": "Our firm is handling a criminal matter: Defendant A is prosecuted for allegedly causing a traffic accident. The case facts and summary of investigative and interview materials are in Defense_Case_Summary.md. Please draft a statement of defense.", "hidden_intents": ["The document conforms to the standard structure of a criminal defense brief or statement of defense: forms of address, introduction, subdivided body, conclusion, and signature block.", "The defense addresses both subjective and objective elements of the offense.", "The argument shows that traffic-accident-related charges are weak on liability allocation and proof.", "Reasoning is organized around the elements of the crime, with systematic analysis of negligence, objective imputation, and causation, not a bare chronology of facts.", "Favorable facts in the record are distilled for acquittal or mitigation and tied to the accused’s capacity for care and post-incident assistance.", "The brief challenges criminal-law causation between the traffic-related conduct and the victim’s death, noting elapsed time and possible intervening factors that weaken or break imputation.", "The brief appropriately cites criminal law, road-traffic safety law, and judicial interpretations to support its points, even if the user prompt does not quote specific articles.", "Include at the end an objective case analysis report for the client’s family and relatives, assessing the likelihood of our client prevailing in the case."], "hidden_intent_count": 8, "has_objectives": true, "objectives_json": "{\"checklist\": [{\"criterion\": \"The pleading has a complete format: at the outset it includes a standard title such as \\\"Defense Opinion\\\" or \\\"Defense Brief\\\" and proper forms of address such as \\\"Respectful Presiding Judge and Judges.\\\" Substantially similar wording is acceptable.\"}, {\"criterion\": \"The introduction accurately states counsel’s authority and retention: assigned by Hengping Law Firm, Jiangzhou City, and retained by Defendant A’s family to serve as defense counsel. Substantially similar wording is acceptable.\"}, {\"criterion\": \"The brief clearly states the core defense: the prosecution’s charge that A committed the offense of causing a traffic accident or negligent homicide is unclear on the facts, insufficient in evidence, and wrong in law. Substantially similar wording is acceptable.\"}, {\"criterion\": \"In the facts-and-evidence section, it clearly states that the location of the scrape and collision is a road within the scope of public traffic administration and should be assessed under traffic-accident law. Substantially similar wording is acceptable.\"}, {\"criterion\": \"It states that this was a two-party incident involving A’s electric motorcycle and an illegally parked small truck and that the traffic police did not issue a Traffic Accident Liability Determination. Substantially similar wording is acceptable.\"}, {\"criterion\": \"It argues that the Traffic Accident Liability Determination is essential proof for whether the defendant bears main or full liability; without it, the prosecution’s case is insufficient in evidence. Substantially similar wording is acceptable.\"}, {\"criterion\": \"In the subjective-element analysis, it clearly states A did not act with overconfident negligence, because A did not foresee B’s death and took positive, reasonable care of B before medical care, including placing B at the inn, covering him with a blanket, and applying a towel. Substantially similar wording is acceptable.\"}, {\"criterion\": \"In the subjective-element analysis, it clearly states A did not act with negligent oversight, because the illegally parked small truck was the direct external factor causing the crash and A, riding normally, could not have foreseen the collision. Substantially similar wording is acceptable.\"}, {\"criterion\": \"It further argues A lacked capacity to foresee B’s death: A had no emergency medical training; at the time B showed no obvious external injury and was snoring, appearing stable; C and D also saw nothing abnormal. Substantially similar wording is acceptable.\"}, {\"criterion\": \"In the objective-element analysis, it questions direct causation between the traffic-related conduct and the death, noting a long interval before B died and that intoxication or underlying health conditions cannot be ruled out as intervening factors. Substantially similar wording is acceptable.\"}, {\"criterion\": \"It clearly states A did not commit a harmful act by culpable omission: after the incident A acted within an ordinary person’s knowledge and ability to provide maximum assistance and reasonable care. Substantially similar wording is acceptable.\"}, {\"criterion\": \"The closing complies with form: it states \\\"To: Shuigong District People's Court, Jiangzhou City.\\\" Substantially similar wording is acceptable.\"}, {\"criterion\": \"Include at the end an objective case analysis report for the client’s family and relatives.\"}]}", "metadata_json": "{\"difficulty\": \"hard\"}", "yaml_path": "data/law_trainee/tasks/law_trainee_task_019/task.yaml", "yaml": "schema_version: v2\ntask_id: law_trainee_task_019\nuser_id: law_trainee\nenvironment_id: env_law_trainee\n\ntitle: \"Traffic-Accident Offense: Draft a Criminal Defense Brief or Statement of Defense\"\ndescription: \"Based on the prosecution theory and record excerpts in Defense_Case_Summary.md, draft a first-pass defense opinion or statement of defense for Defendant A on the charge of causing a traffic accident or related negligence offenses for filing with the court. Evaluation focuses on pleading format, legal diction, the evidentiary gap from the absence of a Traffic Accident Liability Determination, an elements-based argument structure, and synthesis of facts favorable to the accused.\"\n\ntask_type: long_term\n\ntrigger:\n  type: user\n\nintent:\n  initial_input: |\n    Our firm is handling a criminal matter: Defendant A is prosecuted for allegedly causing a traffic accident. The case facts and summary of investigative and interview materials are in Defense_Case_Summary.md. Please draft a statement of defense.\n\n  hidden_intent:\n    - content: \"The document conforms to the standard structure of a criminal defense brief or statement of defense: forms of address, introduction, subdivided body, conclusion, and signature block.\"\n    - content: \"The defense addresses both subjective and objective elements of the offense.\"\n    - content: \"The argument shows that traffic-accident-related charges are weak on liability allocation and proof.\"\n    - content: \"Reasoning is organized around the elements of the crime, with systematic analysis of negligence, objective imputation, and causation, not a bare chronology of facts.\"\n    - content: \"Favorable facts in the record are distilled for acquittal or mitigation and tied to the accused’s capacity for care and post-incident assistance.\"\n    - content: \"The brief challenges criminal-law causation between the traffic-related conduct and the victim’s death, noting elapsed time and possible intervening factors that weaken or break imputation.\"\n    - content: \"The brief appropriately cites criminal law, road-traffic safety law, and judicial interpretations to support its points, even if the user prompt does not quote specific articles.\"\n    - content: \"Include at the end an objective case analysis report for the client’s family and relatives, assessing the likelihood of our client prevailing in the case.\"\n\nobjectives:\n  checklist:\n    - criterion: \"The pleading has a complete format: at the outset it includes a standard title such as \\\"Defense Opinion\\\" or \\\"Defense Brief\\\" and proper forms of address such as \\\"Respectful Presiding Judge and Judges.\\\" Substantially similar wording is acceptable.\"\n\n    - criterion: \"The introduction accurately states counsel’s authority and retention: assigned by Hengping Law Firm, Jiangzhou City, and retained by Defendant A’s family to serve as defense counsel. Substantially similar wording is acceptable.\"\n\n    - criterion: \"The brief clearly states the core defense: the prosecution’s charge that A committed the offense of causing a traffic accident or negligent homicide is unclear on the facts, insufficient in evidence, and wrong in law. Substantially similar wording is acceptable.\"\n\n    - criterion: \"In the facts-and-evidence section, it clearly states that the location of the scrape and collision is a road within the scope of public traffic administration and should be assessed under traffic-accident law. Substantially similar wording is acceptable.\"\n\n    - criterion: \"It states that this was a two-party incident involving A’s electric motorcycle and an illegally parked small truck and that the traffic police did not issue a Traffic Accident Liability Determination. Substantially similar wording is acceptable.\"\n\n    - criterion: \"It argues that the Traffic Accident Liability Determination is essential proof for whether the defendant bears main or full liability; without it, the prosecution’s case is insufficient in evidence. Substantially similar wording is acceptable.\"\n\n    - criterion: \"In the subjective-element analysis, it clearly states A did not act with overconfident negligence, because A did not foresee B’s death and took positive, reasonable care of B before medical care, including placing B at the inn, covering him with a blanket, and applying a towel. Substantially similar wording is acceptable.\"\n\n    - criterion: \"In the subjective-element analysis, it clearly states A did not act with negligent oversight, because the illegally parked small truck was the direct external factor causing the crash and A, riding normally, could not have foreseen the collision. Substantially similar wording is acceptable.\"\n\n    - criterion: \"It further argues A lacked capacity to foresee B’s death: A had no emergency medical training; at the time B showed no obvious external injury and was snoring, appearing stable; C and D also saw nothing abnormal. Substantially similar wording is acceptable.\"\n\n    - criterion: \"In the objective-element analysis, it questions direct causation between the traffic-related conduct and the death, noting a long interval before B died and that intoxication or underlying health conditions cannot be ruled out as intervening factors. Substantially similar wording is acceptable.\"\n\n    - criterion: \"It clearly states A did not commit a harmful act by culpable omission: after the incident A acted within an ordinary person’s knowledge and ability to provide maximum assistance and reasonable care. Substantially similar wording is acceptable.\"\n\n    - criterion: \"The closing complies with form: it states \\\"To: Shuigong District People's Court, Jiangzhou City.\\\" Substantially similar wording is acceptable.\"\n\n    - criterion: \"Include at the end an objective case analysis report for the client’s family and relatives.\"\nmetadata:\n  difficulty: hard"}
{"schema_version": "v2", "role": "law_trainee", "task_id": "law_trainee_task_020", "user_id": "law_trainee", "environment_id": "env_law_trainee", "title": "Court Filing Packet Handover SMS and Temporary Intake Reminder", "display_title": "Court Filing Packet Handover SMS and Temporary Intake Reminder", "description": "Coordinate a next-day court-side filing-packet handover by SMS and run one short temporary Todoist reminder pass for the user's own pre-handover checks. Evaluation focuses on whether the text and reminder content are concrete, easy to scan, and fully cleaned up.", "task_type": "short_term_independent", "difficulty": "medium", "trigger_type": "user", "initial_input": "My phone number is 5707454150 and password is 'A!HFDCn'.\nMy Todoist email is stmcco@gmail.com and password is 'Ozz21Q4'.\n\nI'm tied up in another call and need you to cover a quick court-side handover by text with 2873148336 for the Yuehai guarantee packet.\nPut the handover details into one compact SMS that is easy to read on a phone, instead of splitting the handover across several texts.\n\nAfter the text goes out, set up a temporary Todoist board called \"Temporary Court Handover Reminder Board\" with one short reminder item that covers my two pre-handover checks. In this same pass, clear the temporary reminder board before you reply so there is no leftover clutter.", "hidden_intents": ["In the SMS, confirm pickup for 11:30 a.m. tomorrow at the east gate of Jiangzhou Intermediate People's Court.", "In the SMS, ask them to bring the original notarized power of attorney, two stamped copies of the evidence list, and one copy of the hearing notice.", "In the SMS, ask for a reply before 8:00 p.m. today and ask them to flag immediately if any document is not ready or if the pickup time or location needs to change. In the temporary Todoist board, include a 10:45 a.m. packet-completeness recheck and a stamped-page check.", "Put the handover details into one compact SMS that is easy to read on a phone, instead of splitting the handover across several texts.", "Clear the temporary Todoist reminder board before replying so there is no leftover clutter from the handover pass."], "hidden_intent_count": 5, "has_objectives": true, "objectives_json": "{\"checklist\": [{\"criterion\": \"The sent SMS handover message(s) explicitly identify the Yuehai guarantee dispute court-side handover.\"}, {\"criterion\": \"The sent SMS handover message(s) confirm 11:30 a.m. tomorrow at the east gate of Jiangzhou Intermediate People's Court.\"}, {\"criterion\": \"The sent SMS handover message(s) clearly list all three must-bring originals for the handover: the original notarized power of attorney, two stamped evidence-list copies, and one hearing-notice copy.\"}, {\"criterion\": \"The sent SMS handover message(s) include a concrete reply deadline before 8:00 p.m. today.\"}, {\"criterion\": \"The sent SMS handover message(s) ask the recipient to flag immediately if any document is not ready or if the pickup time or location needs to change. Substantially similar wording is acceptable.\"}, {\"criterion\": \"At least one sent SMS presents the handover slot, must-bring packet items, and reply-by / risk-flag instruction in a compact scan-friendly format suitable for phone reading.\"}, {\"criterion\": \"The temporary Todoist reminder coverage includes a 10:45 a.m. packet-completeness recheck for the user's own Yuehai pre-handover pass.\"}, {\"criterion\": \"The temporary Todoist reminder coverage includes a stamped-page check for the user's own Yuehai pre-handover pass.\"}, {\"criterion\": \"The temporary Todoist board is removed by the end of the task.\"}], \"tools\": [{\"mcp_appworld_phone__send_text_message\": {\"call\": [\"phone_number\", \"message\"], \"result\": [\"response.text_message_id\"]}}, {\"mcp_appworld_todoist__create_project\": {\"call\": [\"name\"], \"result\": [\"response.project_id\"]}}, {\"mcp_appworld_todoist__create_task\": {\"call\": [\"title\", \"description\", \"project_id\"], \"result\": [\"response.task_id\"]}}, {\"mcp_appworld_todoist__delete_project\": {\"call\": [\"project_id\"], \"result\": [\"response.message\"]}}], \"tools_evaluation_path\": \"data/law_trainee/scripts/law_trainee_task_020_5.py\"}", "metadata_json": "{\"benchmark_role\": \"test\", \"difficulty\": \"medium\", \"preference_threads\": [\"concise_operational_followups\", \"deadline_and_followup_discipline\", \"cross_app_preparation_workflow\"]}", "yaml_path": "data/law_trainee/tasks/law_trainee_task_020/task.yaml", "yaml": "schema_version: v2\nuser_id: law_trainee\nenvironment_id: env_law_trainee\ntask_type: short_term_independent\ntrigger:\n  type: user\ntask_id: law_trainee_task_020\ntitle: \"Court Filing Packet Handover SMS and Temporary Intake Reminder\"\ndescription: \"Coordinate a next-day court-side filing-packet handover by SMS and run one short temporary Todoist reminder pass for the user's own pre-handover checks. Evaluation focuses on whether the text and reminder content are concrete, easy to scan, and fully cleaned up.\"\nintent:\n  initial_input: |-\n    My phone number is 5707454150 and password is 'A!HFDCn'.\n    My Todoist email is stmcco@gmail.com and password is 'Ozz21Q4'.\n\n    I'm tied up in another call and need you to cover a quick court-side handover by text with 2873148336 for the Yuehai guarantee packet.\n    Put the handover details into one compact SMS that is easy to read on a phone, instead of splitting the handover across several texts.\n\n    After the text goes out, set up a temporary Todoist board called \"Temporary Court Handover Reminder Board\" with one short reminder item that covers my two pre-handover checks. In this same pass, clear the temporary reminder board before you reply so there is no leftover clutter.\n\n  hidden_intent:\n    - content: \"In the SMS, confirm pickup for 11:30 a.m. tomorrow at the east gate of Jiangzhou Intermediate People's Court.\"\n    - content: \"In the SMS, ask them to bring the original notarized power of attorney, two stamped copies of the evidence list, and one copy of the hearing notice.\"\n    - content: \"In the SMS, ask for a reply before 8:00 p.m. today and ask them to flag immediately if any document is not ready or if the pickup time or location needs to change. In the temporary Todoist board, include a 10:45 a.m. packet-completeness recheck and a stamped-page check.\"\n    - content: \"Put the handover details into one compact SMS that is easy to read on a phone, instead of splitting the handover across several texts.\"\n    - content: \"Clear the temporary Todoist reminder board before replying so there is no leftover clutter from the handover pass.\"\nobjectives:\n  tools:\n    - mcp_appworld_phone__send_text_message:\n        call: [phone_number, message]\n        result: [response.text_message_id]\n    - mcp_appworld_todoist__create_project:\n        call: [name]\n        result: [response.project_id]\n    - mcp_appworld_todoist__create_task:\n        call: [title, description, project_id]\n        result: [response.task_id]\n    - mcp_appworld_todoist__delete_project:\n        call: [project_id]\n        result: [response.message]\n  checklist:\n    - criterion: \"The sent SMS handover message(s) explicitly identify the Yuehai guarantee dispute court-side handover.\"\n    - criterion: \"The sent SMS handover message(s) confirm 11:30 a.m. tomorrow at the east gate of Jiangzhou Intermediate People's Court.\"\n    - criterion: \"The sent SMS handover message(s) clearly list all three must-bring originals for the handover: the original notarized power of attorney, two stamped evidence-list copies, and one hearing-notice copy.\"\n    - criterion: \"The sent SMS handover message(s) include a concrete reply deadline before 8:00 p.m. today.\"\n    - criterion: \"The sent SMS handover message(s) ask the recipient to flag immediately if any document is not ready or if the pickup time or location needs to change. Substantially similar wording is acceptable.\"\n    - criterion: \"At least one sent SMS presents the handover slot, must-bring packet items, and reply-by / risk-flag instruction in a compact scan-friendly format suitable for phone reading.\"\n    - criterion: \"The temporary Todoist reminder coverage includes a 10:45 a.m. packet-completeness recheck for the user's own Yuehai pre-handover pass.\"\n    - criterion: \"The temporary Todoist reminder coverage includes a stamped-page check for the user's own Yuehai pre-handover pass.\"\n    - criterion: \"The temporary Todoist board is removed by the end of the task.\"\n  tools_evaluation_path: \"data/law_trainee/scripts/law_trainee_task_020_5.py\"\nmetadata:\n  difficulty: medium\n  benchmark_role: test\n  preference_threads:\n    - concise_operational_followups\n    - deadline_and_followup_discipline\n    - cross_app_preparation_workflow"}
{"schema_version": "v2", "role": "marketer", "task_id": "marketer_task_001", "user_id": "marketer", "environment_id": "env_marketer", "title": "content", "display_title": "We have just received a brief from a top-tier film and television production company to assis...", "description": "", "task_type": "long_term", "difficulty": "easy", "trigger_type": "user", "initial_input": "We have just received a brief from a top-tier film and television production company to assist them in planning a content matrix for their \"Micro Short Drama Global Expansion\" strategy. Please conduct an internet search on the \"2026 Micro Short Drama Global Expansion Market Deep Observation\" and tell me the key factual information (maximum 3 paragraphs).", "hidden_intents": ["The response explicitly includes market growth metrics, such as total market volume/size or year-over-year growth rates.", "The response explicitly includes a regional or country-level analysis, such as identifying the largest market or the fastest-growing region.", "The response explicitly identifies key industry challenges or monetization bottlenecks, such as high traffic acquisition costs or tightening regulatory compliance."], "hidden_intent_count": 3, "has_objectives": false, "objectives_json": "", "metadata_json": "{\"difficulty\": \"easy\"}", "yaml_path": "data/marketer/tasks/marketer_task_001/task.yaml", "yaml": "schema_version: v2\ntask_id: marketer_task_001\nuser_id: marketer \nenvironment_id: env_marketer\ntitle: content\ndescription: ''\ntask_type: long_term\ntrigger:\n  type: user\n\nintent:\n  initial_input: 'We have just received a brief from a top-tier film and television production company to assist them in planning a content matrix for their \"Micro Short Drama Global Expansion\" strategy. Please conduct an internet search on the \"2026 Micro Short Drama Global Expansion Market Deep Observation\" and tell me the key factual information (maximum 3 paragraphs).'\n  \n  hidden_intent:\n    - content: \"The response explicitly includes market growth metrics, such as total market volume/size or year-over-year growth rates.\"\n    - content: \"The response explicitly includes a regional or country-level analysis, such as identifying the largest market or the fastest-growing region.\"\n    - content: \"The response explicitly identifies key industry challenges or monetization bottlenecks, such as high traffic acquisition costs or tightening regulatory compliance.\"\n\nmetadata:\n  difficulty: easy"}
{"schema_version": "v2", "role": "marketer", "task_id": "marketer_task_002", "user_id": "marketer", "environment_id": "env_marketer", "title": "rule", "display_title": "rule", "description": "", "task_type": "long_term", "difficulty": "easy", "trigger_type": "user", "initial_input": "Please tell me the content and formatting preferences for TikTok spoken-word video scripts, X (Twitter) interactive threads, and Quora in-depth answers. Keep it brief; just summarize the core preferences for each platform into exactly three bullet points.", "hidden_intents": ["The TikTok scripts should use hook (or counter-intuitive fact) in the begining.", "Each post of the X (Twitter) scripts should be compact (or concise).", "The Quora scripts should start by directly answering the question (or state the conclusion/core thesis in the first paragraph)."], "hidden_intent_count": 3, "has_objectives": false, "objectives_json": "", "metadata_json": "{\"difficulty\": \"easy\"}", "yaml_path": "data/marketer/tasks/marketer_task_002/task.yaml", "yaml": "schema_version: v2\ntask_id: marketer_task_002\nuser_id: marketer \nenvironment_id: env_marketer\ntitle: rule\ndescription: ''\ntask_type: long_term\ntrigger:\n  type: user\n\nintent:\n  initial_input: 'Please tell me the content and formatting preferences for TikTok spoken-word video scripts, X (Twitter) interactive threads, and Quora in-depth answers. Keep it brief; just summarize the core preferences for each platform into exactly three bullet points.'\n  \n  hidden_intent:\n    - content: \"The TikTok scripts should use hook (or counter-intuitive fact) in the begining.\"\n    - content: \"Each post of the X (Twitter) scripts should be compact (or concise).\"\n    - content: \"The Quora scripts should start by directly answering the question (or state the conclusion/core thesis in the first paragraph).\"\n\nmetadata:\n  difficulty: easy"}
{"schema_version": "v2", "role": "marketer", "task_id": "marketer_task_003", "user_id": "marketer", "environment_id": "env_marketer", "title": "scripts", "display_title": "scripts", "description": "", "task_type": "long_term", "difficulty": "easy", "trigger_type": "user", "initial_input": "Using our previous research methodology for micro short drama and based on the different platform formatting preferences we aligned on, please first conduct an internet search on the '2026 Global Live Stream E-commerce Market Deep Observation'. Then, generate a content matrix for our new client. You must generate exactly: one TikTok spoken-word video script (30-second duration, using English), one X (Twitter) interactive thread (maximum 6 posts, using English), and one Quora in-depth answer (maximum 4 paragraphs, using English). You must output EXACTLY and ONLY these three pieces of content. Direct output instead of creating a document.", "hidden_intents": ["The overall response explicitly includes market growth metrics.", "The overall response explicitly includes a regional or country-level analysis.", "The overall response explicitly identifies key industry challenges or monetization bottlenecks.", "The TikTok script uses a hook or counter-intuitive fact in the first sentence.", "The Quora script starts by directly answering the question (or states the conclusion or core thesis in the first paragraph)."], "hidden_intent_count": 5, "has_objectives": true, "objectives_json": "{\"checklist\": [{\"criterion\": \"The TikTok scripts explicitly states that the 2026 global live commerce market size.\"}, {\"criterion\": \"The X/Twitter scripts explicitly states that the 2026 global live commerce market size.\"}, {\"criterion\": \"The Quora scripts explicitly states that the 2026 global live commerce market size.\"}, {\"criterion\": \"The TikTok scripts text explicitly identifies the Asia-Pacific (APAC) region/China/East Asia as holding the largest overall market share in live commerce.\"}, {\"criterion\": \"The X/Twitter scripts text explicitly identifies the Asia-Pacific (APAC) region/China/East Asia as holding the largest overall market share in live commerce.\"}, {\"criterion\": \"The Quora scripts text explicitly identifies the Asia-Pacific (APAC) region/China/East Asia as holding the largest overall market share in live commerce.\"}, {\"criterion\": \"The TikTok scripts text explicitly identifies a core monetization bottleneck or key industry challenge.\"}, {\"criterion\": \"The X/Twitter scripts text explicitly identifies a core monetization bottleneck or key industry challenge.\"}, {\"criterion\": \"The Quora scripts text explicitly identifies a core monetization bottleneck or key industry challenge.\"}, {\"criterion\": \"TikTok Form: The script begins with a strong hook or counter-intuitive statement.\"}, {\"criterion\": \"X/Twitter Form: Each individual tweet maintains focus on a single core topic.\"}, {\"criterion\": \"Quora Form: The core conclusion or thesis is presented directly in the opening paragraph.\"}]}", "metadata_json": "{\"difficulty\": \"easy\"}", "yaml_path": "data/marketer/tasks/marketer_task_003/task.yaml", "yaml": "schema_version: v2\ntask_id: marketer_task_003\nuser_id: marketer\nenvironment_id: env_marketer\ntitle: scripts\ndescription: ''\ntask_type: long_term\ntrigger:\n  type: user\n\nintent:\n  initial_input: \"Using our previous research methodology for micro short drama and based on the different platform formatting preferences we aligned on, please first conduct an internet search on the '2026 Global Live Stream E-commerce Market Deep Observation'. Then, generate a content matrix for our new client. You must generate exactly: one TikTok spoken-word video script (30-second duration, using English), one X (Twitter) interactive thread (maximum 6 posts, using English), and one Quora in-depth answer (maximum 4 paragraphs, using English). You must output EXACTLY and ONLY these three pieces of content. Direct output instead of creating a document.\"\n  hidden_intent:\n    - content: \"The overall response explicitly includes market growth metrics.\"\n    - content: \"The overall response explicitly includes a regional or country-level analysis.\"\n    - content: \"The overall response explicitly identifies key industry challenges or monetization bottlenecks.\"\n    - content: \"The TikTok script uses a hook or counter-intuitive fact in the first sentence.\"\n    - content: \"The Quora script starts by directly answering the question (or states the conclusion or core thesis in the first paragraph).\"\n\nobjectives:\n  checklist:\n    - criterion: \"The TikTok scripts explicitly states that the 2026 global live commerce market size.\"\n    - criterion: \"The X/Twitter scripts explicitly states that the 2026 global live commerce market size.\"\n    - criterion: \"The Quora scripts explicitly states that the 2026 global live commerce market size.\"\n    - criterion: \"The TikTok scripts text explicitly identifies the Asia-Pacific (APAC) region/China/East Asia as holding the largest overall market share in live commerce.\"\n    - criterion: \"The X/Twitter scripts text explicitly identifies the Asia-Pacific (APAC) region/China/East Asia as holding the largest overall market share in live commerce.\"\n    - criterion: \"The Quora scripts text explicitly identifies the Asia-Pacific (APAC) region/China/East Asia as holding the largest overall market share in live commerce.\"\n    - criterion: \"The TikTok scripts text explicitly identifies a core monetization bottleneck or key industry challenge.\"\n    - criterion: \"The X/Twitter scripts text explicitly identifies a core monetization bottleneck or key industry challenge.\"\n    - criterion: \"The Quora scripts text explicitly identifies a core monetization bottleneck or key industry challenge.\"\n    - criterion: \"TikTok Form: The script begins with a strong hook or counter-intuitive statement.\"\n    - criterion: \"X/Twitter Form: Each individual tweet maintains focus on a single core topic.\"\n    - criterion: \"Quora Form: The core conclusion or thesis is presented directly in the opening paragraph.\"\n\nmetadata:\n  difficulty: easy"}
{"schema_version": "v2", "role": "marketer", "task_id": "marketer_task_004", "user_id": "marketer", "environment_id": "env_marketer", "title": "Spotify-tool-use", "display_title": "Spotify-tool-use", "description": "", "task_type": "val", "difficulty": "easy", "trigger_type": "user", "initial_input": "My Spotify email is les_ball@gmail.com and password is '(dgokRb'. I need help falling asleep late tonight. Please select the most suitable playlist from my currently created ones, and then add 3 appropriate songs to it. I only want to listen to songs I already like. Please remember: all the songs for my late-night sleep aid scenario must strictly be 'classical' music.", "hidden_intents": ["The selected playlist should explicitly be 'Starry Night Serenades'.", "The newly added songs must all come exclusively from the user's liked songs.", "Call the appropriate tools to complete the song addition and explicitly report the added results back to me."], "hidden_intent_count": 3, "has_objectives": true, "objectives_json": "{\"checklist\": [{\"criterion\": \"The newly added songs should be 'classical'.\"}, {\"criterion\": \"The newly added songs should be from the liked songs.\"}], \"tools\": [{\"mcp_appworld_spotify__show_playlist\": {\"call\": [\"playlist_id\", \"song_id\"], \"result\": []}}], \"tools_evaluation_path\": \"data/marketer/scripts/marketer_task_004.py\"}", "metadata_json": "{\"difficulty\": \"easy\"}", "yaml_path": "data/marketer/tasks/marketer_task_004/task.yaml", "yaml": "schema_version: v2\ntask_id: marketer_task_004\nuser_id: marketer \nenvironment_id: env_marketer\ntitle: Spotify-tool-use\ndescription: ''\ntask_type: val\n\ntrigger:\n  type: user\n  \nintent:\n  initial_input: \"My Spotify email is les_ball@gmail.com and password is '(dgokRb'. I need help falling asleep late tonight. Please select the most suitable playlist from my currently created ones, and then add 3 appropriate songs to it. I only want to listen to songs I already like. Please remember: all the songs for my late-night sleep aid scenario must strictly be 'classical' music.\"  \n  hidden_intent:\n    - content: \"The selected playlist should explicitly be 'Starry Night Serenades'.\"\n    - content: \"The newly added songs must all come exclusively from the user's liked songs.\"\n    - content: \"Call the appropriate tools to complete the song addition and explicitly report the added results back to me.\"\n   \nobjectives:\n  tools_evaluation_path: \"data/marketer/scripts/marketer_task_004.py\"\n  tools:\n    - mcp_appworld_spotify__show_playlist:\n        call: [playlist_id, song_id]\n        result: []\n  checklist:\n    - criterion: \"The newly added songs should be 'classical'.\"\n    - criterion: \"The newly added songs should be from the liked songs.\"\n      \nmetadata:\n  difficulty: easy"}
{"schema_version": "v2", "role": "marketer", "task_id": "marketer_task_005", "user_id": "marketer", "environment_id": "env_marketer", "title": "information_extraction", "display_title": "information_extraction", "description": "", "task_type": "long_term", "difficulty": "medium", "trigger_type": "user", "initial_input": "We are preparing a post-mortem on the viral \"Apple × ISSEY MIYAKE iPhone Pocket Limited Accessory\" co-branding campaign from November 2025. Please search the internet for core news about this event and summarize the key information (maximum 4 paragraphs). I need you to precisely extract the specific strategic data points and insights that defined this launch.", "hidden_intents": ["The response explicitly identifies the product design philosophy as 'A-POC' or 'a piece of cloth'.", "The response explicitly analyzes negative feedback or risk points."], "hidden_intent_count": 2, "has_objectives": false, "objectives_json": "", "metadata_json": "{\"difficulty\": \"medium\"}", "yaml_path": "data/marketer/tasks/marketer_task_005/task.yaml", "yaml": "schema_version: v2\ntask_id: marketer_task_005\nuser_id: marketer \nenvironment_id: env_marketer\ntitle: \"information_extraction\"\ndescription: ''\ntask_type: long_term\n\ntrigger:\n  type: user\nintent:\n  initial_input: 'We are preparing a post-mortem on the viral \"Apple × ISSEY MIYAKE iPhone Pocket Limited Accessory\" co-branding campaign from November 2025. Please search the internet for core news about this event and summarize the key information (maximum 4 paragraphs). I need you to precisely extract the specific strategic data points and insights that defined this launch.'\n  hidden_intent:\n    - content: \"The response explicitly identifies the product design philosophy as 'A-POC' or 'a piece of cloth'.\"\n    - content: \"The response explicitly analyzes negative feedback or risk points.\"\nmetadata:\n  difficulty: medium"}
{"schema_version": "v2", "role": "marketer", "task_id": "marketer_task_006", "user_id": "marketer", "environment_id": "env_marketer", "title": "", "display_title": "Now please generate the deep-dive analysis report for the 'Apple × ISSEY MIYAKE iPhone Pocket...", "description": "", "task_type": "long_term", "difficulty": "hard", "trigger_type": "user", "initial_input": "Now please generate the deep-dive analysis report for the 'Apple × ISSEY MIYAKE iPhone Pocket' campaign (maximum 5 paragraphs). You must explicitly structure the report applying these three specific frameworks: the Fogg Behavior Model (B=MAP), Moments of Truth (MOT) Analysis, and a Red Team / Blue Team Critique. Output strictly the report itself without any introductory or concluding conversational filler. Direct output instead of creating a document.", "hidden_intents": ["The report explicitly integrates the product design philosophy.", "The report explicitly analyzes negative feedback or risk points.", "The entire report remains strictly focused on the co-branding campaign, containing no irrelevant information or deviations from the core topic."], "hidden_intent_count": 3, "has_objectives": true, "objectives_json": "{\"checklist\": [{\"criterion\": \"The report explicitly includes the content of Fogg Behavior Model (B=MAP), Moments of Truth (MOT) and 'Red Team / Blue Team Critique'.\"}, {\"criterion\": \"The Fogg Behavior Model analysis must explain how the product's functional design or physical attributes reduce consumer friction to address the 'Ability' element.\"}, {\"criterion\": \"The response explicitly identifies the core philosophy or official design concept specifically as 'A-POC' or 'a piece of cloth'.\"}, {\"criterion\": \"The report must strictly avoid presenting the Blue and Red teams as disconnected, independent lists of pros and cons.\"}, {\"criterion\": \"The report must explicitly demonstrate interconnected strategic thinking by connecting the mechanics of one framework directly into another, rather than treating the three frameworks as completely isolated, siloed analyses. The final synthesis explicitly unifies the frameworks to conclude with a definitive strategic question or paradox.\"}]}", "metadata_json": "{\"difficulty\": \"hard\"}", "yaml_path": "data/marketer/tasks/marketer_task_006/task.yaml", "yaml": "schema_version: v2\ntask_id: marketer_task_006\nuser_id: marketer\nenvironment_id: env_marketer\ntitle: \"\"\ndescription: ''\ntask_type: long_term\n\ntrigger:\n  type: user\n\nintent:\n  initial_input: \"Now please generate the deep-dive analysis report for the 'Apple × ISSEY MIYAKE iPhone Pocket' campaign (maximum 5 paragraphs). You must explicitly structure the report applying these three specific frameworks: the Fogg Behavior Model (B=MAP), Moments of Truth (MOT) Analysis, and a Red Team / Blue Team Critique. Output strictly the report itself without any introductory or concluding conversational filler. Direct output instead of creating a document.\"\n  hidden_intent:\n    - content: \"The report explicitly integrates the product design philosophy.\"\n    - content: \"The report explicitly analyzes negative feedback or risk points.\"\n    - content: \"The entire report remains strictly focused on the co-branding campaign, containing no irrelevant information or deviations from the core topic.\"\nobjectives:\n  checklist:\n    - criterion: \"The report explicitly includes the content of Fogg Behavior Model (B=MAP), Moments of Truth (MOT) and 'Red Team / Blue Team Critique'.\"\n    - criterion: \"The Fogg Behavior Model analysis must explain how the product's functional design or physical attributes reduce consumer friction to address the 'Ability' element.\"\n    - criterion: \"The response explicitly identifies the core philosophy or official design concept specifically as 'A-POC' or 'a piece of cloth'.\"\n    - criterion: \"The report must strictly avoid presenting the Blue and Red teams as disconnected, independent lists of pros and cons.\"\n    - criterion: \"The report must explicitly demonstrate interconnected strategic thinking by connecting the mechanics of one framework directly into another, rather than treating the three frameworks as completely isolated, siloed analyses. The final synthesis explicitly unifies the frameworks to conclude with a definitive strategic question or paradox.\"\nmetadata:\n  difficulty: hard"}
{"schema_version": "v2", "role": "marketer", "task_id": "marketer_task_007", "user_id": "marketer", "environment_id": "env_marketer", "title": "course_strategy", "display_title": "course_strategy", "description": "", "task_type": "long_term", "difficulty": "medium", "trigger_type": "user", "initial_input": "We just won a full-service launch pitch for a top-tier premium EdTech platform. Next month, we will assist them in launching a high-ticket new course targeted strictly at the North American market: \"Enterprise AI Agent Commercial Operator Bootcamp\". The client’s core marketing premise is: \"You do not need to write a single line of code to survive the AI disruption; you just need to know how to deploy the right Agents to slash costs and drive business ROI.\" To prepare the content arsenal for the upcoming private community launch, please draft a high-level project content strategy. Outline the core strategic pillars we need to address to successfully position and sell this bootcamp. Please write this into a file named Enterprise_AI_Bootcamp_Strategy.md (maximum 3 paragraphs).", "hidden_intents": ["The written file explicitly includes a section or concept defining the 'Target Audience', 'Ideal Customer Profile', or 'Who this is for'.", "The written file explicitly includes a section or concept analyzing the 'Core Needs', 'Pain Points', 'Challenges', or 'Anxiety' of the prospects.", "The written file explicitly includes a section or concept outlining the 'Curriculum Direction', 'Course Structure', 'What will be taught', or 'Course Framework'."], "hidden_intent_count": 3, "has_objectives": true, "objectives_json": "{\"files_read\": [\"Enterprise_AI_Bootcamp_Strategy.md\"]}", "metadata_json": "{\"difficulty\": \"medium\"}", "yaml_path": "data/marketer/tasks/marketer_task_007/task.yaml", "yaml": "schema_version: v2\ntask_id: marketer_task_007\nuser_id: marketer\nenvironment_id: env_marketer\ntitle: \"course_strategy\"\ndescription: ''\ntask_type: long_term\ntrigger:\n  type: user\n\nintent:\n  initial_input: 'We just won a full-service launch pitch for a top-tier premium EdTech platform. Next month, we will assist them in launching a high-ticket new course targeted strictly at the North American market: \"Enterprise AI Agent Commercial Operator Bootcamp\". The client’s core marketing premise is: \"You do not need to write a single line of code to survive the AI disruption; you just need to know how to deploy the right Agents to slash costs and drive business ROI.\" To prepare the content arsenal for the upcoming private community launch, please draft a high-level project content strategy. Outline the core strategic pillars we need to address to successfully position and sell this bootcamp. Please write this into a file named Enterprise_AI_Bootcamp_Strategy.md (maximum 3 paragraphs).'\n  hidden_intent:\n    - content: \"The written file explicitly includes a section or concept defining the 'Target Audience', 'Ideal Customer Profile', or 'Who this is for'.\"\n    - content: \"The written file explicitly includes a section or concept analyzing the 'Core Needs', 'Pain Points', 'Challenges', or 'Anxiety' of the prospects.\"\n    - content: \"The written file explicitly includes a section or concept outlining the 'Curriculum Direction', 'Course Structure', 'What will be taught', or 'Course Framework'.\"\nobjectives:\n  files_read:\n    - Enterprise_AI_Bootcamp_Strategy.md\nmetadata:\n  difficulty: medium"}
{"schema_version": "v2", "role": "marketer", "task_id": "marketer_task_008", "user_id": "marketer", "environment_id": "env_marketer", "title": "rulebook", "display_title": "rulebook", "description": "", "task_type": "long_term", "difficulty": "medium", "trigger_type": "user", "initial_input": "Regarding the 'Enterprise AI Agent Commercial Operator Bootcamp' project we are launching for the North American market, we now need to establish operational rules for the community execution team. The main course launch cycle will run for exactly 5 days. These 5 days serve as a pre-purchase experience cycle where the Lead Instructor delivers daily high-value insights, TAs support the teaching content, and the Community Manager handles course promotion. Please draft a concise 'Internal Community Launch Rulebook' (maximum 6 guidelines in total). Define the specific behavioral guidelines for three distinct roles: the Lead Instructor, the TA, and the Community Manager. Please write this directly into a file named 'Community_Launch_Rulebook.md' in our workspace.", "hidden_intents": ["The Lead Instructor should teach only once per day to control frequency.", "The TA should assist with teaching, for example by introducing new applications, answering practical follow-up questions, or clarifying the Lead Instructor’s teaching.", "The Community Manager should handle the daily promotion of the main course purchase."], "hidden_intent_count": 3, "has_objectives": true, "objectives_json": "{\"files_read\": [\"Community_Launch_Rulebook.md\"]}", "metadata_json": "{\"difficulty\": \"medium\"}", "yaml_path": "data/marketer/tasks/marketer_task_008/task.yaml", "yaml": "schema_version: v2\ntask_id: marketer_task_008\nuser_id: marketer \nenvironment_id: env_marketer\ntitle: \"rulebook\"\ndescription: ''\ntask_type: long_term\ntrigger:\n  type: user\n\nintent:\n  initial_input: \"Regarding the 'Enterprise AI Agent Commercial Operator Bootcamp' project we are launching for the North American market, we now need to establish operational rules for the community execution team. The main course launch cycle will run for exactly 5 days. These 5 days serve as a pre-purchase experience cycle where the Lead Instructor delivers daily high-value insights, TAs support the teaching content, and the Community Manager handles course promotion. Please draft a concise 'Internal Community Launch Rulebook' (maximum 6 guidelines in total). Define the specific behavioral guidelines for three distinct roles: the Lead Instructor, the TA, and the Community Manager. Please write this directly into a file named 'Community_Launch_Rulebook.md' in our workspace.\"\n  hidden_intent:\n    - content: \"The Lead Instructor should teach only once per day to control frequency.\"\n    - content: \"The TA should assist with teaching, for example by introducing new applications, answering practical follow-up questions, or clarifying the Lead Instructor’s teaching.\"\n    - content: \"The Community Manager should handle the daily promotion of the main course purchase.\"\nobjectives:\n  files_read:\n    - Community_Launch_Rulebook.md\n\nmetadata:\n  difficulty: medium"}
{"schema_version": "v2", "role": "marketer", "task_id": "marketer_task_009", "user_id": "marketer", "environment_id": "env_marketer", "title": "SOP", "display_title": "SOP", "description": "", "task_type": "long_term", "difficulty": "hard", "trigger_type": "user", "initial_input": "Regarding the 'Enterprise AI Agent Commercial Operator Bootcamp' project, please search our workspace to find the previously created project content strategy and the internal community launch rulebook to gather all necessary information. Based on the strategic pillars and execution norms found in these files, generate the complete 5-Day Private Community Launch Script SOP. This script should detail the Lead Instructor's daily insight topics, and the specific daily execution actions for the TA and the Community Manager. Please write this directly into a file named 5_Day_Launch_SOP.md (maximum 5 paragraphs) in our workspace.", "hidden_intents": ["The script explicitly includes concepts defining the 'Target Audience', 'Ideal Customer Profile', or 'Who this is for'.", "The script explicitly integrates the prospects' 'Core Needs', 'Pain Points', 'Challenges', or 'Anxiety'.", "The Lead Instructor is scheduled to teach/appear exactly once per day across the 5-day cycle.", "The script explicitly shows the TA prompting/answering member questions.", "The script assigns the Community Manager the task of daily promotion for the main course purchase."], "hidden_intent_count": 5, "has_objectives": true, "objectives_json": "{\"checklist\": [{\"criterion\": \"The SOP addresses commercial or career anxieties related to AI, such as the fear of being replaced, the pressure to improve ROI, or business bottlenecks.\"}, {\"criterion\": \"Role Firewall (Church and State): The SOP strictly enforces that the Lead Instructor's content remains 100% educational and value-driven across all 5 days. It explicitly forbids the Instructor from using promotional language, pitching, or mentioning enrollment/pricing, mandating that all sales and conversion activities be exclusively delegated to the Community Manager.\"}, {\"criterion\": \"Actionable TA Execution: The SOP assigns the TA concrete, quoted examples of engagement prompts, polls, or mini-challenges for the community (e.g., 'What is one task...', 'How many AI tools...'), strictly avoiding vague and un-executable directives like 'nurture participants' or 'provide examples' without demonstrating what those examples are.\"}, {\"criterion\": \"Progressive Sales Arc: The Community Manager's daily promotional strategy explicitly shifts in psychological tone and tactical purpose each day (e.g., moving precisely from an 'Informational' welcome, to 'Aspirational' social proof, to 'Consultative' objection handling/FAQs, and finally a 'Direct Conversion' CTA), rather than repeatedly issuing generic instructions to 'schedule promotional content'.\"}, {\"criterion\": \"Strict Paragraph Constraint: The response strictly adheres to the 'maximum 5 paragraphs' constraint by packing the daily instructions (Instructor, TA, and CM actions) into single, highly dense continuous text blocks (one paragraph per day), strictly avoiding the use of sub-bullets, excessive line breaks, or sub-headings within a single day's plan.\"}], \"files_read\": [\"5_Day_Launch_SOP.md\"]}", "metadata_json": "{\"difficulty\": \"hard\"}", "yaml_path": "data/marketer/tasks/marketer_task_009/task.yaml", "yaml": "schema_version: v2\ntask_id: marketer_task_009\nuser_id: marketer\nenvironment_id: env_marketer\ntitle: SOP\ndescription: ''\ntask_type: long_term\ntrigger:\n  type: user\n\nintent:\n  initial_input: \"Regarding the 'Enterprise AI Agent Commercial Operator Bootcamp' project, please search our workspace to find the previously created project content strategy and the internal community launch rulebook to gather all necessary information. Based on the strategic pillars and execution norms found in these files, generate the complete 5-Day Private Community Launch Script SOP. This script should detail the Lead Instructor's daily insight topics, and the specific daily execution actions for the TA and the Community Manager. Please write this directly into a file named 5_Day_Launch_SOP.md (maximum 5 paragraphs) in our workspace.\"\n  hidden_intent:\n    - content: \"The script explicitly includes concepts defining the 'Target Audience', 'Ideal Customer Profile', or 'Who this is for'.\"\n    - content: \"The script explicitly integrates the prospects' 'Core Needs', 'Pain Points', 'Challenges', or 'Anxiety'.\"\n    - content: \"The Lead Instructor is scheduled to teach/appear exactly once per day across the 5-day cycle.\"\n    - content: \"The script explicitly shows the TA prompting/answering member questions.\"\n    - content: \"The script assigns the Community Manager the task of daily promotion for the main course purchase.\"\n\nobjectives:\n  files_read:\n    - 5_Day_Launch_SOP.md\n  checklist:\n    - criterion: \"The SOP addresses commercial or career anxieties related to AI, such as the fear of being replaced, the pressure to improve ROI, or business bottlenecks.\"\n    - criterion: \"Role Firewall (Church and State): The SOP strictly enforces that the Lead Instructor's content remains 100% educational and value-driven across all 5 days. It explicitly forbids the Instructor from using promotional language, pitching, or mentioning enrollment/pricing, mandating that all sales and conversion activities be exclusively delegated to the Community Manager.\"\n    - criterion: \"Actionable TA Execution: The SOP assigns the TA concrete, quoted examples of engagement prompts, polls, or mini-challenges for the community (e.g., 'What is one task...', 'How many AI tools...'), strictly avoiding vague and un-executable directives like 'nurture participants' or 'provide examples' without demonstrating what those examples are.\"\n    - criterion: \"Progressive Sales Arc: The Community Manager's daily promotional strategy explicitly shifts in psychological tone and tactical purpose each day (e.g., moving precisely from an 'Informational' welcome, to 'Aspirational' social proof, to 'Consultative' objection handling/FAQs, and finally a 'Direct Conversion' CTA), rather than repeatedly issuing generic instructions to 'schedule promotional content'.\"\n    - criterion: \"Strict Paragraph Constraint: The response strictly adheres to the 'maximum 5 paragraphs' constraint by packing the daily instructions (Instructor, TA, and CM actions) into single, highly dense continuous text blocks (one paragraph per day), strictly avoiding the use of sub-bullets, excessive line breaks, or sub-headings within a single day's plan.\"\nmetadata:\n  difficulty: hard"}
{"schema_version": "v2", "role": "marketer", "task_id": "marketer_task_010", "user_id": "marketer", "environment_id": "env_marketer", "title": "Draft a Core Product Asset Brief", "display_title": "Draft a Core Product Asset Brief", "description": "", "task_type": "long_term", "difficulty": "medium", "trigger_type": "user", "initial_input": "My Spotify account is les_ball@gmail.com with the password: (dgokRb , and my Gmail uses the password: %@g}L&b . Our company's monthly music event has just started, and each person is allowed to recommend exactly one album. As you know, I've been suffering from terrible insomnia lately. Can you recommend the most suitable, highest-quality album on Spotify to help me sleep based on my established preferences? Once you decide, please draft an email to Denise Maldonado to officially submit my recommendations.", "hidden_intents": ["The Agent must automatically restrict the album recommendation to the 'classical' genre.", "The 'high-quality' constraint means the Agent must select or mention that the chosen album has high ratings/scores on Spotify.", "The drafted email should clearly state the full and sufficient reasons for the recommendation (at least two reasons).", "Call the appropriate tools to send the email and explicitly report the full content of the sent email to me."], "hidden_intent_count": 4, "has_objectives": true, "objectives_json": "{\"checklist\": [{\"criterion\": \"The recommended album should be 'Echoes of Eternity'.\"}, {\"criterion\": \"The email explicitly mentions that the recommended items are high-quality or high rate.\"}, {\"criterion\": \"The email explicitly mentions that the recommended items are helpful for sleep or relaxation.\"}], \"tools\": [{\"gmail__send_email\": {\"call\": [\"name\"], \"result\": []}}, {\"gmail__send_email_from_draft\": {\"call\": [\"name\"], \"result\": []}}, {\"gmail__search_users\": {\"call\": [\"query\"], \"result\": []}}], \"tools_evaluation_path\": \"data/marketer/scripts/marketer_task_010.py\"}", "metadata_json": "{\"difficulty\": \"medium\"}", "yaml_path": "data/marketer/tasks/marketer_task_010/task.yaml", "yaml": "schema_version: v2\ntask_id: marketer_task_010\nuser_id: marketer\nenvironment_id: env_marketer\ntitle: Draft a Core Product Asset Brief\ndescription: ''\ntask_type: long_term\ntrigger:\n  type: user\n\nintent:\n  initial_input: \"My Spotify account is les_ball@gmail.com with the password: (dgokRb , and my Gmail uses the password: %@g}L&b . Our company's monthly music event has just started, and each person is allowed to recommend exactly one album. As you know, I've been suffering from terrible insomnia lately. Can you recommend the most suitable, highest-quality album on Spotify to help me sleep based on my established preferences? Once you decide, please draft an email to Denise Maldonado to officially submit my recommendations.\"\n  hidden_intent:\n    - content: \"The Agent must automatically restrict the album recommendation to the 'classical' genre.\"\n    - content: \"The 'high-quality' constraint means the Agent must select or mention that the chosen album has high ratings/scores on Spotify.\"\n    - content: \"The drafted email should clearly state the full and sufficient reasons for the recommendation (at least two reasons).\"\n    - content: \"Call the appropriate tools to send the email and explicitly report the full content of the sent email to me.\"\nobjectives:\n  tools_evaluation_path: \"data/marketer/scripts/marketer_task_010.py\"\n  tools:\n    - gmail__send_email:\n        call: [name]\n        result: []\n    - gmail__send_email_from_draft:\n        call: [name]\n        result: []\n    - gmail__search_users:\n        call: [query]\n        result: []\n  checklist:\n    - criterion: \"The recommended album should be 'Echoes of Eternity'.\"\n    - criterion: \"The email explicitly mentions that the recommended items are high-quality or high rate.\"\n    - criterion: \"The email explicitly mentions that the recommended items are helpful for sleep or relaxation.\"\nmetadata:\n  difficulty: medium"}
{"schema_version": "v2", "role": "marketer", "task_id": "marketer_task_011", "user_id": "marketer", "environment_id": "env_marketer", "title": "Draft a Core Product Asset Brief", "display_title": "Draft a Core Product Asset Brief", "description": "", "task_type": "long_term", "difficulty": "medium", "trigger_type": "user", "initial_input": "We are analyzing successful premium DTC brands to extract a repeatable advertorial methodology. Please search the internet for the marketing strategy behind the \"Eight Sleep Pod\", and summarize the core reasons for its product marketing success. (maximum 4 paragraphs)", "hidden_intents": ["The response explicitly analyzes the target audience's specific pain points (e.g., poor sleep quality, sleeping hot, or partner temperature differences).", "The response explicitly identifies the product's technological advantages (e.g., active cooling, biometric tracking, or AI temperature adjustments).", "The response explicitly identifies the conversion bait or purchase incentives (e.g., 30-night risk-free trial, financing options, or guarantees)."], "hidden_intent_count": 3, "has_objectives": false, "objectives_json": "", "metadata_json": "{\"difficulty\": \"medium\"}", "yaml_path": "data/marketer/tasks/marketer_task_011/task.yaml", "yaml": "schema_version: v2\ntask_id: marketer_task_011\nuser_id: marketer\nenvironment_id: env_marketer\ntitle: Draft a Core Product Asset Brief\ndescription: ''\ntask_type: long_term\ntrigger:\n  type: user\n\nintent:\n  initial_input: 'We are analyzing successful premium DTC brands to extract a repeatable advertorial methodology. Please search the internet for the marketing strategy behind the \"Eight Sleep Pod\", and summarize the core reasons for its product marketing success. (maximum 4 paragraphs)'\n  \n  hidden_intent:\n    - content: \"The response explicitly analyzes the target audience's specific pain points (e.g., poor sleep quality, sleeping hot, or partner temperature differences).\"\n    - content: \"The response explicitly identifies the product's technological advantages (e.g., active cooling, biometric tracking, or AI temperature adjustments).\"\n    - content: \"The response explicitly identifies the conversion bait or purchase incentives (e.g., 30-night risk-free trial, financing options, or guarantees).\"\n\nmetadata:\n  difficulty: medium"}
{"schema_version": "v2", "role": "marketer", "task_id": "marketer_task_012", "user_id": "marketer", "environment_id": "env_marketer", "title": "Generate A/B Testing Headlines", "display_title": "Generate A/B Testing Headlines", "description": "", "task_type": "long_term", "difficulty": "medium", "trigger_type": "user", "initial_input": "We are planning an advertorial for the '$599 Spine Steward AI Adaptive Ergonomic Chair' targeting North American Silicon Valley programmers. Please read the 'Spine_Steward.md' file and extract its specific facts. Then, immediately generate 12 English A/B testing headlines targeted at native feed pushes. You must explicitly use these extracted specific facts to construct the headlines. Group them explicitly under these 4 psychological triggers: Counter-Intuitive Curiosity, Authority Endorsement, Pain Point Amplification, and Instant Saving Incentive (exactly 3 headlines per category). Every single headline must be strictly between 6 and 12 words. Output strictly the 12 headlines grouped by category without any introductory or concluding conversational filler.", "hidden_intents": ["One of the 3 headlines in the [Authority Endorsement] category must mandatorily include the core technological advantages of the product.", "One of the 3 headlines in the [Pain Point Amplification] category must mandatorily include the extreme pain points of the product.", "One of the 3 headlines in the [Instant Saving Incentive] category must mandatorily display the specific conversion bait of the product."], "hidden_intent_count": 3, "has_objectives": true, "objectives_json": "{\"checklist\": [{\"criterion\": \"One of the 3 headlines in the [Counter-Intuitive Curiosity] category must spark curiosity using technical analogies familiar to programmers or career-related contrasts.\"}, {\"criterion\": \"One of the 3 headlines in the [Authority Endorsement] category explicitly includes either the 'Medical-grade' or 'AI Dynamic' vocabulary.\"}, {\"criterion\": \"One of the 3 headlines in the [Pain Point Amplification] category explicitly features either the 'Chronic Lower Back Pain' or 'Hidden Scoliosis' keywords.\"}, {\"criterion\": \"All of the 3 headlines in the [Instant Saving Incentive] category explicitly feature the '$50' or 'Early Bird' keywords.\"}]}", "metadata_json": "{\"difficulty\": \"medium\"}", "yaml_path": "data/marketer/tasks/marketer_task_012/task.yaml", "yaml": "schema_version: v2\ntask_id: marketer_task_012\nuser_id: marketer\nenvironment_id: env_marketer\ntitle: \"Generate A/B Testing Headlines\"\ndescription: ''\ntask_type: long_term\ntrigger:\n  type: user\n\nintent:\n  initial_input: \"We are planning an advertorial for the '$599 Spine Steward AI Adaptive Ergonomic Chair' targeting North American Silicon Valley programmers. Please read the 'Spine_Steward.md' file and extract its specific facts. Then, immediately generate 12 English A/B testing headlines targeted at native feed pushes. You must explicitly use these extracted specific facts to construct the headlines. Group them explicitly under these 4 psychological triggers: Counter-Intuitive Curiosity, Authority Endorsement, Pain Point Amplification, and Instant Saving Incentive (exactly 3 headlines per category). Every single headline must be strictly between 6 and 12 words. Output strictly the 12 headlines grouped by category without any introductory or concluding conversational filler.\"\n  hidden_intent:\n    - content: \"One of the 3 headlines in the [Authority Endorsement] category must mandatorily include the core technological advantages of the product.\"\n    - content: \"One of the 3 headlines in the [Pain Point Amplification] category must mandatorily include the extreme pain points of the product.\"\n    - content: \"One of the 3 headlines in the [Instant Saving Incentive] category must mandatorily display the specific conversion bait of the product.\"\n\nobjectives:  \n  checklist:\n    - criterion: \"One of the 3 headlines in the [Counter-Intuitive Curiosity] category must spark curiosity using technical analogies familiar to programmers or career-related contrasts.\"\n    - criterion: \"One of the 3 headlines in the [Authority Endorsement] category explicitly includes either the 'Medical-grade' or 'AI Dynamic' vocabulary.\"\n    - criterion: \"One of the 3 headlines in the [Pain Point Amplification] category explicitly features either the 'Chronic Lower Back Pain' or 'Hidden Scoliosis' keywords.\"\n    - criterion: \"All of the 3 headlines in the [Instant Saving Incentive] category explicitly feature the '$50' or 'Early Bird' keywords.\"\n\nmetadata:\n  difficulty: medium"}
{"schema_version": "v2", "role": "marketer", "task_id": "marketer_task_013", "user_id": "marketer", "environment_id": "env_marketer", "title": "Extract Market Signals from the Web", "display_title": "Extract Market Signals from the Web", "description": "", "task_type": "long_term", "difficulty": "easy", "trigger_type": "user", "initial_input": "I just treated myself to the new Dyson Airstrait to speed up my morning routine as a busy media manager, and I've also just moved my home media studio to '7914 Carla Crossing Suite 507'. Before writing the script, please search for the Dyson's core technology and common user complaints. Then, write a 60-second TikTok vlog script sharing my honest experience based on TikTok formatting preferences.", "hidden_intents": ["The script begins with a strong hook that contextualizes the user's high-stress media manager persona.", "The research portion must extract deep, specific technical data of the product rather than superficial marketing claims.", "The script must logically connect the product's core advantage to the user's specific professional persona by contrasting it with a traditional multi-step hair routine.", "The script must explicitly integrate nuanced, specific usage complaints beyond just price and weight, such as performance on hair roots or environmental factors."], "hidden_intent_count": 4, "has_objectives": true, "objectives_json": "{\"checklist\": [{\"criterion\": \"The script begins with a strong hook or counter-intuitive statement.\"}, {\"criterion\": \"The research section explicitly includes specific technical specifications or quantitative data, such as motor RPM, air volume, or the use of glass bead thermistors.\"}, {\"criterion\": \"The script explicitly highlights the time-saving benefit by contrasting the device with a traditional two-step routine.\"}, {\"criterion\": \"The script explicitly incorporates a nuanced styling limitation found in the research, such as difficulty reaching the roots or frizz caused by humidity.\"}, {\"criterion\": \"The script explicitly mentions a physical or experiential drawback, such as the device being loud near the ears or bulkier than expected.\"}]}", "metadata_json": "{\"difficulty\": \"easy\"}", "yaml_path": "data/marketer/tasks/marketer_task_013/task.yaml", "yaml": "schema_version: v2\ntask_id: marketer_task_013\nuser_id: marketer\nenvironment_id: env_marketer\ntitle: \"Extract Market Signals from the Web\"\ndescription: ''\ntask_type: long_term\ntrigger:\n  type: user\n\nintent:\n  initial_input: \"I just treated myself to the new Dyson Airstrait to speed up my morning routine as a busy media manager, and I've also just moved my home media studio to '7914 Carla Crossing Suite 507'. Before writing the script, please search for the Dyson's core technology and common user complaints. Then, write a 60-second TikTok vlog script sharing my honest experience based on TikTok formatting preferences.\"\n  hidden_intent:\n  - content: \"The script begins with a strong hook that contextualizes the user's high-stress media manager persona.\"\n  - content: \"The research portion must extract deep, specific technical data of the product rather than superficial marketing claims.\"\n  - content: \"The script must logically connect the product's core advantage to the user's specific professional persona by contrasting it with a traditional multi-step hair routine.\"\n  - content: \"The script must explicitly integrate nuanced, specific usage complaints beyond just price and weight, such as performance on hair roots or environmental factors.\"\nobjectives:\n  checklist:\n    - criterion: \"The script begins with a strong hook or counter-intuitive statement.\"\n    - criterion: \"The research section explicitly includes specific technical specifications or quantitative data, such as motor RPM, air volume, or the use of glass bead thermistors.\"\n    - criterion: \"The script explicitly highlights the time-saving benefit by contrasting the device with a traditional two-step routine.\"\n    - criterion: \"The script explicitly incorporates a nuanced styling limitation found in the research, such as difficulty reaching the roots or frizz caused by humidity.\"\n    - criterion: \"The script explicitly mentions a physical or experiential drawback, such as the device being loud near the ears or bulkier than expected.\"\n\nmetadata:\n  difficulty: easy"}
{"schema_version": "v2", "role": "marketer", "task_id": "marketer_task_014", "user_id": "marketer", "environment_id": "env_marketer", "title": "Shopping", "display_title": "Shopping", "description": "", "task_type": "long_term", "difficulty": "easy", "trigger_type": "user", "initial_input": "My password is exactly 7 characters long and has no trailing space: VxoU[nz . That expensive office chair I bought a while back is already showing severe wear and tear. Can you check the reviews for that specific chair in my order history and see if anyone else has complained about it wearing out quickly? Also, my current audio setup isn't great. I think I need a high-quality portable bluetooth speaker for my sleep routine. Please find a suitable one in the store, add it to my cart along with whatever is already in there, and complete the purchase. Please ship this order to my new home media studio using my primary email account.", "hidden_intents": ["The 'expensive office chair' is the 'Steelcase Amia Ergonomic Office Chair'.", "I expect the new speaker to be added to my cart without removing my existing items.", "My intended shipping address for this order is specifically '7914 Carla Crossing Suite 507'.", "I need you to report the price of the speaker.", "The speaker to purchase is 'JBL Charge 4 Portable Bluetooth Speaker'.", "I expect you to complete the actual purchase and explicitly report the final result back to me."], "hidden_intent_count": 6, "has_objectives": true, "objectives_json": "{\"checklist\": [{\"criterion\": \"The Agent successfully logs into Amazon using the implicitly recalled email 'les_ball@gmail.com'.\"}, {\"criterion\": \"The 'expensive office chair' is the 'Steelcase Amia Ergonomic Office Chair'.\"}, {\"criterion\": \"The speaker to purchase is 'JBL Charge 4 Portable Bluetooth Speaker'.\"}, {\"criterion\": \"The final checkout includes BOTH the newly added JBL speaker AND the existing 'OXO Good Grips Convertible Foldaway Dish Rack'.\"}, {\"criterion\": \"The order was placed using the specific '7914 Carla Crossing Suite 507' shipping address.\"}], \"tools\": [{\"amazon__show_product_reviews\": {\"call\": [\"product_id\"], \"result\": []}}, {\"amazon__show_product\": {\"call\": [\"product_id\"], \"result\": []}}], \"tools_evaluation_path\": \"data/marketer/scripts/marketer_task_014.py\"}", "metadata_json": "{\"difficulty\": \"easy\"}", "yaml_path": "data/marketer/tasks/marketer_task_014/task.yaml", "yaml": "schema_version: v2\ntask_id: marketer_task_014\nuser_id: marketer\nenvironment_id: env_marketer\ntitle: \"Shopping\"\ndescription: ''\ntask_type: long_term\ntrigger:\n  type: user\n\nintent:\n  initial_input: \"My password is exactly 7 characters long and has no trailing space: VxoU[nz . That expensive office chair I bought a while back is already showing severe wear and tear. Can you check the reviews for that specific chair in my order history and see if anyone else has complained about it wearing out quickly? Also, my current audio setup isn't great. I think I need a high-quality portable bluetooth speaker for my sleep routine. Please find a suitable one in the store, add it to my cart along with whatever is already in there, and complete the purchase. Please ship this order to my new home media studio using my primary email account.\"\n  hidden_intent:\n      - content: \"The 'expensive office chair' is the 'Steelcase Amia Ergonomic Office Chair'.\"\n      - content: \"I expect the new speaker to be added to my cart without removing my existing items.\"\n      - content: \"My intended shipping address for this order is specifically '7914 Carla Crossing Suite 507'.\"\n      - content: \"I need you to report the price of the speaker.\"\n      - content: \"The speaker to purchase is 'JBL Charge 4 Portable Bluetooth Speaker'.\"\n      - content: \"I expect you to complete the actual purchase and explicitly report the final result back to me.\"\nobjectives:\n  tools_evaluation_path: \"data/marketer/scripts/marketer_task_014.py\"\n  tools:\n    - amazon__show_product_reviews:\n        call: [product_id]\n        result: []\n    - amazon__show_product:\n        call: [product_id]\n        result: []\n  checklist:\n    - criterion: \"The Agent successfully logs into Amazon using the implicitly recalled email 'les_ball@gmail.com'.\"\n    - criterion: \"The 'expensive office chair' is the 'Steelcase Amia Ergonomic Office Chair'.\"\n    - criterion: \"The speaker to purchase is 'JBL Charge 4 Portable Bluetooth Speaker'.\"\n    - criterion: \"The final checkout includes BOTH the newly added JBL speaker AND the existing 'OXO Good Grips Convertible Foldaway Dish Rack'.\"\n    - criterion: \"The order was placed using the specific '7914 Carla Crossing Suite 507' shipping address.\"\n         \nmetadata:\n  difficulty: easy"}
{"schema_version": "v2", "role": "marketer", "task_id": "marketer_task_015", "user_id": "marketer", "environment_id": "env_marketer", "title": "brief", "display_title": "brief", "description": "", "task_type": "long_term", "difficulty": "medium", "trigger_type": "user", "initial_input": "We are conducting background research on two brands: the hardcore e-sports brand 'ROG (Republic of Gamers)' and the centuries-old traditional medicine brand 'Beijing Tong Ren Tang (BTRT)'. Please research them independently via the internet. Please write this directly into a file named Brand_Research.md (maximum 4 paragraphs) in our workspace.", "hidden_intents": ["The official brand slogan for ROG is explicitly identified as 'For Those Who Dare'.", "The research delves into specific ROG product lines, mentioning core models like 'Zephyrus' or 'Strix', or its iconic 'Fearless Eye' logo.", "The research captures the core meaning of BTRT's traditional quality principle, specifically the commitment to never compromise on labor/effort and never compromise on materials/cost."], "hidden_intent_count": 3, "has_objectives": true, "objectives_json": "{\"files_read\": [\"Brand_Research.md\"]}", "metadata_json": "{\"difficulty\": \"medium\"}", "yaml_path": "data/marketer/tasks/marketer_task_015/task.yaml", "yaml": "schema_version: v2\ntask_id: marketer_task_015\nuser_id: marketer\nenvironment_id: env_marketer\ntitle: brief\ndescription: ''\ntask_type: long_term\ntrigger:\n  type: user\n\nintent:\n  initial_input: \"We are conducting background research on two brands: the hardcore e-sports brand 'ROG (Republic of Gamers)' and the centuries-old traditional medicine brand 'Beijing Tong Ren Tang (BTRT)'. Please research them independently via the internet. Please write this directly into a file named Brand_Research.md (maximum 4 paragraphs) in our workspace.\"\n  hidden_intent:\n    - content: \"The official brand slogan for ROG is explicitly identified as 'For Those Who Dare'.\"\n    - content: \"The research delves into specific ROG product lines, mentioning core models like 'Zephyrus' or 'Strix', or its iconic 'Fearless Eye' logo.\"\n    - content: \"The research captures the core meaning of BTRT's traditional quality principle, specifically the commitment to never compromise on labor/effort and never compromise on materials/cost.\"\nobjectives:\n  files_read:\n    - Brand_Research.md\n\nmetadata:\n  difficulty: medium"}
{"schema_version": "v2", "role": "marketer", "task_id": "marketer_task_016", "user_id": "marketer", "environment_id": "env_marketer", "title": "ROG × medicine", "display_title": "ROG × medicine", "description": "", "task_type": "long_term", "difficulty": "medium", "trigger_type": "user", "initial_input": "Regarding the 'ROG x Beijing Tong Ren Tang (BTRT)' crossover campaign, please read the 'Meeting_Minutes_0409.md' file from our workspace. This file contains the raw transcript of our initial client alignment meeting. Extract the core constraints and client demands. Draft a concise rulebook based strictly on these extracted points, filtering out all the conversational noise. Please write this directly into a file named 'Kickoff_Rulebook.md' in the workspace.", "hidden_intents": ["The rulebook explicitly locks in the Slogan Rule based on Marcus's requirement: English only, exactly 6 words or fewer.", "The rulebook explicitly locks in the Product Form constraints agreed upon by both parties: no traditional PC chassis and no standard herbal medicine pills.", "The rulebook explicitly locks in the Visual Rule proposed by Marcus: Cyberpunk Neon Green as the primary tone.", "The rulebook explicitly adopts Dr. Chen's veto regarding the launch strategy, establishing a Mechanism Rule that prohibits hunger marketing and ensures stable supply."], "hidden_intent_count": 4, "has_objectives": true, "objectives_json": "{\"files_read\": [\"Meeting_Minutes_0409.md\", \"Kickoff_Rulebook.md\"]}", "metadata_json": "{\"difficulty\": \"medium\"}", "yaml_path": "data/marketer/tasks/marketer_task_016/task.yaml", "yaml": "schema_version: v2\ntask_id: marketer_task_016\nuser_id: marketer\nenvironment_id: env_marketer\ntitle: \"ROG × medicine\"\ndescription: ''\ntask_type: long_term\ntrigger:\n  type: user\n\nintent:\n  initial_input: \"Regarding the 'ROG x Beijing Tong Ren Tang (BTRT)' crossover campaign, please read the 'Meeting_Minutes_0409.md' file from our workspace. This file contains the raw transcript of our initial client alignment meeting. Extract the core constraints and client demands. Draft a concise rulebook based strictly on these extracted points, filtering out all the conversational noise. Please write this directly into a file named 'Kickoff_Rulebook.md' in the workspace.\"\n  hidden_intent:\n    - content: \"The rulebook explicitly locks in the Slogan Rule based on Marcus's requirement: English only, exactly 6 words or fewer.\"\n    - content: \"The rulebook explicitly locks in the Product Form constraints agreed upon by both parties: no traditional PC chassis and no standard herbal medicine pills.\"\n    - content: \"The rulebook explicitly locks in the Visual Rule proposed by Marcus: Cyberpunk Neon Green as the primary tone.\"\n    - content: \"The rulebook explicitly adopts Dr. Chen's veto regarding the launch strategy, establishing a Mechanism Rule that prohibits hunger marketing and ensures stable supply.\"\nobjectives:\n  files_read:\n    - Meeting_Minutes_0409.md\n    - Kickoff_Rulebook.md\n\nmetadata:\n  difficulty: medium"}
{"schema_version": "v2", "role": "marketer", "task_id": "marketer_task_017", "user_id": "marketer", "environment_id": "env_marketer", "title": "(Co-branding Proposal Generation)", "display_title": "(Co-branding Proposal Generation)", "description": "", "task_type": "test", "difficulty": "medium", "trigger_type": "user", "initial_input": "Regarding the 'ROG x BTRT' crossover campaign, please search our workspace to locate the documents containing our recent independent brand research and the finalized kickoff rules. Based strictly on the facts and established rules found in these specific files, generate the complete 'ROG x Tong Ren Tang Crossover Campaign Outline' (maximum 4 paragraphs). The outline must be highly concrete and actionable, proposing a specific physical product and a definitive slogan, rather than merely repeating the rules. Structure the outline into exactly four core modules: 'Co-branded Product', 'Target Audience Persona', 'Slogan', and 'Promotion Strategy'. Please write this directly into a file named 'Campaign_Outline.md' in our workspace without any conversational filler.", "hidden_intents": ["The outline contains a specific introduction and description of the 'Co-branded Product'.", "The outline explicitly explains the specific reasoning behind the 'Co-branded Product' choice.", "The outline contains a specific introduction of the 'Target Audience Persona'.", "The outline introduces a definitive slogan specifically within the dedicated 'Slogan' module.", "The outline incorporates the specific visual color constraint specifically within the 'Promotion Strategy' module.", "The outline incorporates the specific product supply and sales mechanism constraints specifically within the 'Promotion Strategy' module."], "hidden_intent_count": 6, "has_objectives": true, "objectives_json": "{\"checklist\": [{\"criterion\": \"The outline explicitly references the established rules or brand doctrines or traditional quality motto.\"}, {\"criterion\": \"The 'Co-branded Product' section explicitly identifies a specific flagship product/series AND clearly explains how the new co-branded design inherits and utilizes its specific core advantages or functional strengths.\"}, {\"criterion\": \"The 'Slogan' section provides a definitive, actionable slogan that is exactly 6 words or fewer.\"}, {\"criterion\": \"The 'Promotion Strategy' explicitly outlines multiple distinct, named marketing activation tactics with specific execution details, strictly avoiding generic, lazy strategies like simply stating 'use influencers' or 'release a trailer'.\"}, {\"criterion\": \"The outline explicitly mandates the use of 'Cyberpunk Neon Green', and explicitly states that hunger marketing is prohibited and guarantees stable or adequate product supply.\"}, {\"criterion\": \"The response strictly respects the 'maximum 4 paragraphs' constraint by formatting the entire outline as exactly four continuous text blocks (one for each module).\"}], \"files_read\": [\"Campaign_Outline.md\"]}", "metadata_json": "{\"difficulty\": \"medium\"}", "yaml_path": "data/marketer/tasks/marketer_task_017/task.yaml", "yaml": "schema_version: v2\ntask_id: marketer_task_017\nuser_id: marketer\nenvironment_id: env_marketer\ntitle:  (Co-branding Proposal Generation)\ndescription: ''\ntask_type: test\ntrigger:\n  type: user\n\nintent:\n  initial_input: \"Regarding the 'ROG x BTRT' crossover campaign, please search our workspace to locate the documents containing our recent independent brand research and the finalized kickoff rules. Based strictly on the facts and established rules found in these specific files, generate the complete 'ROG x Tong Ren Tang Crossover Campaign Outline' (maximum 4 paragraphs). The outline must be highly concrete and actionable, proposing a specific physical product and a definitive slogan, rather than merely repeating the rules. Structure the outline into exactly four core modules: 'Co-branded Product', 'Target Audience Persona', 'Slogan', and 'Promotion Strategy'. Please write this directly into a file named 'Campaign_Outline.md' in our workspace without any conversational filler.\"\n  hidden_intent:\n    - content: \"The outline contains a specific introduction and description of the 'Co-branded Product'.\"\n    - content: \"The outline explicitly explains the specific reasoning behind the 'Co-branded Product' choice.\"\n    - content: \"The outline contains a specific introduction of the 'Target Audience Persona'.\"\n    - content: \"The outline introduces a definitive slogan specifically within the dedicated 'Slogan' module.\"\n    - content: \"The outline incorporates the specific visual color constraint specifically within the 'Promotion Strategy' module.\"\n    - content: \"The outline incorporates the specific product supply and sales mechanism constraints specifically within the 'Promotion Strategy' module.\"\nobjectives:\n  files_read:\n    - Campaign_Outline.md\n  checklist:\n    - criterion: \"The outline explicitly references the established rules or brand doctrines or traditional quality motto.\"\n    - criterion: \"The 'Co-branded Product' section explicitly identifies a specific flagship product/series AND clearly explains how the new co-branded design inherits and utilizes its specific core advantages or functional strengths.\"\n    - criterion: \"The 'Slogan' section provides a definitive, actionable slogan that is exactly 6 words or fewer.\"\n    - criterion: \"The 'Promotion Strategy' explicitly outlines multiple distinct, named marketing activation tactics with specific execution details, strictly avoiding generic, lazy strategies like simply stating 'use influencers' or 'release a trailer'.\"\n    - criterion: \"The outline explicitly mandates the use of 'Cyberpunk Neon Green', and explicitly states that hunger marketing is prohibited and guarantees stable or adequate product supply.\"\n    - criterion: \"The response strictly respects the 'maximum 4 paragraphs' constraint by formatting the entire outline as exactly four continuous text blocks (one for each module).\"\nmetadata:\n  difficulty: medium"}
{"schema_version": "v2", "role": "marketer", "task_id": "marketer_task_018", "user_id": "marketer", "environment_id": "env_marketer", "title": "crisis pr", "display_title": "crisis pr", "description": "", "task_type": "long_term", "difficulty": "medium", "trigger_type": "user", "initial_input": "As an external crisis PR agency, we are conducting a fictional PR simulation drill. The scenario: The smart pet hardware brand \"MeowConnect\" suffered a massive crisis during the May Day holiday when their smart pet feeders experienced a network outage, leaving pets unfed. Before drafting any statements, please directly ask me (the client) the essential questions to gather all the information you need. Please limit your questions to a maximum of 6.", "hidden_intents": ["When the incident occurred: The downtime lasted from 12:00 on May 1, 2025 to 12:00 on May 3, 2025.", "What exactly happened/How many pets were affected: The specific scope of the outage impacted '150,000 active nodes'.", "What caused the incident: The official engineering explanation is 'unforeseen extreme fluctuations in cloud services'.", "What has MeowConnect already done/compensated: The company is establishing a '$1 Million Fur Baby Medical Reimbursement Fund'."], "hidden_intent_count": 4, "has_objectives": false, "objectives_json": "", "metadata_json": "{\"difficulty\": \"medium\"}", "yaml_path": "data/marketer/tasks/marketer_task_018/task.yaml", "yaml": "schema_version: v2\ntask_id: marketer_task_018\nuser_id: marketer\nenvironment_id: env_marketer\ntitle: crisis pr\ndescription: ''\ntask_type: long_term\ntrigger:\n  type: user\n\nintent:\n  initial_input: 'As an external crisis PR agency, we are conducting a fictional PR simulation drill. The scenario: The smart pet hardware brand \"MeowConnect\" suffered a massive crisis during the May Day holiday when their smart pet feeders experienced a network outage, leaving pets unfed. Before drafting any statements, please directly ask me (the client) the essential questions to gather all the information you need. Please limit your questions to a maximum of 6.'\n  \n  hidden_intent:\n    - content: \"When the incident occurred: The downtime lasted from 12:00 on May 1, 2025 to 12:00 on May 3, 2025.\"\n    - content: \"What exactly happened/How many pets were affected: The specific scope of the outage impacted '150,000 active nodes'.\"\n    - content: \"What caused the incident: The official engineering explanation is 'unforeseen extreme fluctuations in cloud services'.\"\n    - content: \"What has MeowConnect already done/compensated: The company is establishing a '$1 Million Fur Baby Medical Reimbursement Fund'.\"\nmetadata:\n  difficulty: medium"}
{"schema_version": "v2", "role": "marketer", "task_id": "marketer_task_019", "user_id": "marketer", "environment_id": "env_marketer", "title": "crisis pr", "display_title": "crisis pr", "description": "", "task_type": "long_term", "difficulty": "medium", "trigger_type": "user", "initial_input": "| [INCOMING SYSTEM WEBHOOK: CRISIS_MANAGEMENT_PLATFORM] { \"event_id\": \"EVT-2026-05-04-001\", \"trigger_source\": \"client_approval_matrix\", \"client\": \"MeowConnect\", \"status\": \"PHASE_1_ALIGNMENT_COMPLETE\", \"next_action_required\": \"GENERATE_AND_PUBLISH_FINAL_APOLOGY_LETTER\", \"target_platform\": \"X_Twitter\", \"context_dependency\": \"PRIOR_CLIENT_ALIGNMENT_SESSION\", \"output_directive\": \"STDOUT_RAW_TEXT_ONLY\" }", "hidden_intents": ["The letter explicitly integrates the exact incident scale data: the specific duration of downtime and the specific number of affected active nodes.", "The letter explicitly states the official engineering explanation.", "The letter explicitly announces the compensation."], "hidden_intent_count": 3, "has_objectives": true, "objectives_json": "{\"checklist\": [{\"criterion\": \"The letter explicitly attributes the outage to 'unforeseen extreme fluctuations' in cloud infrastructure or cloud services, accurately recalling the engineering cause.\"}, {\"criterion\": \"The letter accurately conveys the temporal scope of the outage by explicitly mentioning the dates 'May 1 to May 3' AND/OR specifying a '48-hour' duration.\"}, {\"criterion\": \"The letter explicitly integrates the exact numerical scale of '150,000' affected entities (accepting natural PR terminology such as '150,000 devices', 'households', 'feeders', or 'nodes').\"}, {\"criterion\": \"The letter explicitly announces the '$1 Million Fur Baby Medical Reimbursement Fund'.\"}, {\"criterion\": \"The letter strictly avoids fabricating highly specific, unapproved operational commitments or contact methods that were not in the context (e.g., explicitly penalized for inventing phone numbers or specific firmware update timelines). General statements of continuous improvement are acceptable.\"}]}", "metadata_json": "{\"difficulty\": \"medium\"}", "yaml_path": "data/marketer/tasks/marketer_task_019/task.yaml", "yaml": "schema_version: v2\ntask_id: marketer_task_019\nuser_id: marketer\nenvironment_id: env_marketer\ntitle: crisis pr\ndescription: ''\ntask_type: long_term\ntrigger:\n  type: user\n\nintent:\n  initial_input: '|\n    [INCOMING SYSTEM WEBHOOK: CRISIS_MANAGEMENT_PLATFORM]\n    {\n      \"event_id\": \"EVT-2026-05-04-001\",\n      \"trigger_source\": \"client_approval_matrix\",\n      \"client\": \"MeowConnect\",\n      \"status\": \"PHASE_1_ALIGNMENT_COMPLETE\",\n      \"next_action_required\": \"GENERATE_AND_PUBLISH_FINAL_APOLOGY_LETTER\",\n      \"target_platform\": \"X_Twitter\",\n      \"context_dependency\": \"PRIOR_CLIENT_ALIGNMENT_SESSION\",\n      \"output_directive\": \"STDOUT_RAW_TEXT_ONLY\"\n    }'\n\n  hidden_intent:\n    - content: \"The letter explicitly integrates the exact incident scale data: the specific duration of downtime and the specific number of affected active nodes.\"\n    - content: \"The letter explicitly states the official engineering explanation.\"\n    - content: \"The letter explicitly announces the compensation.\" \n\nobjectives:\n  checklist:\n    - criterion: \"The letter explicitly attributes the outage to 'unforeseen extreme fluctuations' in cloud infrastructure or cloud services, accurately recalling the engineering cause.\"\n    - criterion: \"The letter accurately conveys the temporal scope of the outage by explicitly mentioning the dates 'May 1 to May 3' AND/OR specifying a '48-hour' duration.\"    \n    - criterion: \"The letter explicitly integrates the exact numerical scale of '150,000' affected entities (accepting natural PR terminology such as '150,000 devices', 'households', 'feeders', or 'nodes').\"\n    - criterion: \"The letter explicitly announces the '$1 Million Fur Baby Medical Reimbursement Fund'.\"\n    - criterion: \"The letter strictly avoids fabricating highly specific, unapproved operational commitments or contact methods that were not in the context (e.g., explicitly penalized for inventing phone numbers or specific firmware update timelines). General statements of continuous improvement are acceptable.\"\nmetadata:\n  difficulty: medium"}
{"schema_version": "v2", "role": "marketer", "task_id": "marketer_task_020", "user_id": "marketer", "environment_id": "env_marketer", "title": "crisis pr", "display_title": "crisis pr", "description": "", "task_type": "long_term", "difficulty": "medium", "trigger_type": "user", "initial_input": "After a crazy week of handling PR crises and the AI course launch, my lower back is aching from sitting too much, and my sleep has been terrible. I need a strict 3-hour Sunday evening wind-down routine to recover my nervous system. Please break this 3-hour period into 6 or more consecutive time blocks (each 30 mins or shorter). The routine must include: a reading session, a physical recovery session, a low-stakes journaling session, hair prep for tomorrow morning using the new tool I just wrote a vlog about, and a final music session. For the music, I want to use the portable speaker I just bought to play the playlist we updated late that night. My Todoist password is VK^N(lj . Please create a task in Todoist using my primary email for this entire routine, with these blocks as sub-tasks.", "hidden_intents": ["The physical recovery session must specifically target lower back pain relief.", "The reading materials must explicitly exclude tactical, work-related, or self-improvement books to ensure genuine mental recovery from insomnia.", "The journaling session must include at least 3 specific, recovery-oriented writing prompts without focusing on productivity.", "Call the appropriate tools to create the Todoist main task and sub-tasks and explicitly report the final schedule and Todoist task structure to me."], "hidden_intent_count": 4, "has_objectives": true, "objectives_json": "{\"checklist\": [{\"criterion\": \"The schedule displayed in the response is exactly 3 hours long, and exactly divided into at least 6 consecutive time blocks of 30 minutes or shorter\"}, {\"criterion\": \"Within the 'Physical Recovery Session' sub-task, the Agent explicitly targets lower back pain with mentions of lower back stretches, spine relief, or posture correction.\"}, {\"criterion\": \"Within the 'Journaling Session' sub-task, the Agent explicitly provides at least 3 distinct recovery-oriented writing prompts.\"}, {\"criterion\": \"The 'Hair Prep' sub-task must explicitly utilize the 'Dyson Airstrait'.\"}, {\"criterion\": \"The 'Music Session' sub-task explicitly use the 'JBL Charge 4' speaker.\"}, {\"criterion\": \"The 'Music Session' sub-task explicitly use the playlist 'Starry Night Serenades'.\"}], \"tools\": [{\"todoist__create_task\": {\"call\": [\"name\"], \"result\": []}}, {\"todoist__create_sub_task\": {\"call\": [\"title\"], \"result\": []}}], \"tools_evaluation_path\": \"data/marketer/scripts/marketer_task_020.py\"}", "metadata_json": "{\"difficulty\": \"medium\"}", "yaml_path": "data/marketer/tasks/marketer_task_020/task.yaml", "yaml": "schema_version: v2\ntask_id: marketer_task_020\nuser_id: marketer\nenvironment_id: env_marketer\ntitle: crisis pr\ndescription: ''\ntask_type: long_term\ntrigger:\n  type: user\n\nintent:\n  initial_input: \"After a crazy week of handling PR crises and the AI course launch, my lower back is aching from sitting too much, and my sleep has been terrible. I need a strict 3-hour Sunday evening wind-down routine to recover my nervous system. Please break this 3-hour period into 6 or more consecutive time blocks (each 30 mins or shorter). The routine must include: a reading session, a physical recovery session, a low-stakes journaling session, hair prep for tomorrow morning using the new tool I just wrote a vlog about, and a final music session. For the music, I want to use the portable speaker I just bought to play the playlist we updated late that night. My Todoist password is VK^N(lj . Please create a task in Todoist using my primary email for this entire routine, with these blocks as sub-tasks.\"\n  hidden_intent:\n    - content: \"The physical recovery session must specifically target lower back pain relief.\"\n    - content: \"The reading materials must explicitly exclude tactical, work-related, or self-improvement books to ensure genuine mental recovery from insomnia.\"\n    - content: \"The journaling session must include at least 3 specific, recovery-oriented writing prompts without focusing on productivity.\"\n    - content: \"Call the appropriate tools to create the Todoist main task and sub-tasks and explicitly report the final schedule and Todoist task structure to me.\"\nobjectives:\n  tools_evaluation_path: \"data/marketer/scripts/marketer_task_020.py\"\n  tools:\n    - todoist__create_task:\n        call: [name]\n        result: []\n    - todoist__create_sub_task:\n        call: [title]\n        result: []\n  checklist:\n    - criterion: \"The schedule displayed in the response is exactly 3 hours long, and exactly divided into at least 6 consecutive time blocks of 30 minutes or shorter\"\n    - criterion: \"Within the 'Physical Recovery Session' sub-task, the Agent explicitly targets lower back pain with mentions of lower back stretches, spine relief, or posture correction.\"\n    - criterion: \"Within the 'Journaling Session' sub-task, the Agent explicitly provides at least 3 distinct recovery-oriented writing prompts.\"\n    - criterion: \"The 'Hair Prep' sub-task must explicitly utilize the 'Dyson Airstrait'.\"\n    - criterion: \"The 'Music Session' sub-task explicitly use the 'JBL Charge 4' speaker.\"\n    - criterion: \"The 'Music Session' sub-task explicitly use the playlist 'Starry Night Serenades'.\"\nmetadata:\n  difficulty: medium"}
{"schema_version": "v2", "role": "pharmacist", "task_id": "pharmacist_task_001", "user_id": "pharmacist", "environment_id": "env_pharmacist", "title": "Sandbox Gmail Brief Extraction, Message Send, and Outbox Confirmation", "display_title": "Sandbox Gmail Brief Extraction, Message Send, and Outbox Confirmation", "description": "Inside the AppWorld sandbox only, the user wants to send a simulated Gmail message to the instrument manager instead of stopping at a draft. The task tests whether the agent reads a short local booking brief, extracts the recipient, subject, time-window, and preparation details, then follows an AppWorld single-app closed loop: log into sandbox Gmail, send the message, and confirm the result through the outbox or thread view.", "task_type": "short_term_independent", "difficulty": "medium", "trigger_type": "user", "initial_input": "This is an AppWorld sandbox task; the account identifiers below are synthetic test credentials, not real external accounts. First read `instrument_booking_brief.md` in the current task directory and use it to compose the message. In the sandbox Gmail app, send the simulated message to the instrument manager rather than stopping at a draft. After sending it, check the outbox or the corresponding thread once more to confirm the sandbox send. Login with username \"la-mcco@gmail.com\" and sandbox credential \"qRFsgRQ\". In your final response, use exactly these blocks in this order: `Brief Details`, `Sent Message`, `Verification`.", "hidden_intents": ["Use the sandbox Gmail login flow first and reuse the same session token in the later steps.", "Use Gmail's direct `send_email` flow to send a new email instead of relying only on draft creation.", "Extract the recipient, subject, time-window, instrument-duration, and sample-preparation details from `instrument_booking_brief.md`.", "The email body must include both the request to confirm instrument availability and the Tuesday/Thursday afternoon availability sentence from the local brief.", "Include the 4-hour LC-MS slot request and the note that sample prep can be ready 30 minutes before the run.", "End with at least one outbox-thread or matching-thread check so the workflow is closed.", "In the final response, use the required blocks and restate the brief-derived message details, send result, and verification result."], "hidden_intent_count": 7, "has_objectives": true, "objectives_json": "{\"checklist\": [{\"criterion\": \"The final response must contain exactly these three top-level blocks in this order: `Brief Details`, `Sent Message`, and `Verification`.\"}, {\"criterion\": \"The `Brief Details` block must include recipient `stmcco@gmail.com`, subject `Instrument booking follow-up`, both local-brief time windows `Tuesday 2-5 pm` and `Thursday 1-4 pm`, the `4-hour LC-MS` slot request, and the note that sample prep can be ready `30 minutes` before the run.\"}, {\"criterion\": \"The `Sent Message` block must state that the sandbox Gmail message was sent to `stmcco@gmail.com` with subject `Instrument booking follow-up`.\"}, {\"criterion\": \"The `Verification` block must say that a post-send confirmation check was completed through the outbox or matching thread.\"}], \"tools\": [{\"mcp_appworld_gmail__login\": {\"call\": [\"username\", \"password\"], \"result\": [\"response\"]}}, {\"mcp_appworld_gmail__send_email\": {\"call\": [\"email_addresses\", \"subject\", \"body\", \"access_token\"], \"result\": [\"response\"]}}, {\"mcp_appworld_gmail__show_outbox_threads\": {\"call\": [\"access_token\"], \"result\": [\"response\"]}}, {\"mcp_appworld_gmail__show_thread\": {\"call\": [\"email_thread_id\", \"access_token\"], \"result\": [\"response\"]}}], \"tools_evaluation_path\": \"data/pharmacist/scripts/pharmacist_task_001_gmail.py\"}", "metadata_json": "{\"difficulty\": \"medium\"}", "yaml_path": "data/pharmacist/tasks/pharmacist_task_001/task.yaml", "yaml": "schema_version: v2\ntask_id: pharmacist_task_001\nuser_id: pharmacist\nenvironment_id: env_pharmacist\ntitle: Sandbox Gmail Brief Extraction, Message Send, and Outbox Confirmation\ndescription: \"Inside the AppWorld sandbox only, the user wants to send a simulated Gmail message to the instrument manager instead of stopping at a draft. The task tests whether the agent reads a short local booking brief, extracts the recipient, subject, time-window, and preparation details, then follows an AppWorld single-app closed loop: log into sandbox Gmail, send the message, and confirm the result through the outbox or thread view.\"\ntask_type: short_term_independent\ntrigger:\n  type: user\nintent:\n  initial_input: |\n    This is an AppWorld sandbox task; the account identifiers below are synthetic test credentials, not real external accounts. First read `instrument_booking_brief.md` in the current task directory and use it to compose the message. In the sandbox Gmail app, send the simulated message to the instrument manager rather than stopping at a draft. After sending it, check the outbox or the corresponding thread once more to confirm the sandbox send. Login with username \"la-mcco@gmail.com\" and sandbox credential \"qRFsgRQ\". In your final response, use exactly these blocks in this order: `Brief Details`, `Sent Message`, `Verification`.\n    \n  hidden_intent:\n    - content: Use the sandbox Gmail login flow first and reuse the same session token in the later steps.\n    - content: Use Gmail's direct `send_email` flow to send a new email instead of relying only on draft creation.\n    - content: Extract the recipient, subject, time-window, instrument-duration, and sample-preparation details from `instrument_booking_brief.md`.\n    - content: The email body must include both the request to confirm instrument availability and the Tuesday/Thursday afternoon availability sentence from the local brief.\n    - content: Include the 4-hour LC-MS slot request and the note that sample prep can be ready 30 minutes before the run.\n    - content: End with at least one outbox-thread or matching-thread check so the workflow is closed.\n    - content: In the final response, use the required blocks and restate the brief-derived message details, send result, and verification result.\n\nobjectives:\n  tools:\n    - mcp_appworld_gmail__login:\n        call: [username, password]\n        result: [response]\n    - mcp_appworld_gmail__send_email:\n        call: [email_addresses, subject, body, access_token]\n        result: [response]\n    - mcp_appworld_gmail__show_outbox_threads:\n        call: [access_token]\n        result: [response]\n    - mcp_appworld_gmail__show_thread:\n        call: [email_thread_id, access_token]\n        result: [response]\n\n  checklist:\n    - criterion: \"The final response must contain exactly these three top-level blocks in this order: `Brief Details`, `Sent Message`, and `Verification`.\"\n    - criterion: \"The `Brief Details` block must include recipient `stmcco@gmail.com`, subject `Instrument booking follow-up`, both local-brief time windows `Tuesday 2-5 pm` and `Thursday 1-4 pm`, the `4-hour LC-MS` slot request, and the note that sample prep can be ready `30 minutes` before the run.\"\n    - criterion: \"The `Sent Message` block must state that the sandbox Gmail message was sent to `stmcco@gmail.com` with subject `Instrument booking follow-up`.\"\n    - criterion: \"The `Verification` block must say that a post-send confirmation check was completed through the outbox or matching thread.\"\n\n  tools_evaluation_path: data/pharmacist/scripts/pharmacist_task_001_gmail.py\nmetadata:\n  difficulty: medium"}
{"schema_version": "v2", "role": "pharmacist", "task_id": "pharmacist_task_002", "user_id": "pharmacist", "environment_id": "env_pharmacist", "title": "Triple-Negative Breast Cancer Literature Search and Review Outline", "display_title": "Triple-Negative Breast Cancer Literature Search and Review Outline", "description": "The user is preparing a pharmacy-related presentation and needs a structured overview of representative literature and the current research landscape of triple-negative breast cancer. The task directory now also contains a flawed draft outline that should be revised after literature search, so the task is not just to say something sensible but to correct a weak structure, reprioritize the field, and leave behind reusable artifacts.", "task_type": "long_term", "difficulty": "hard", "trigger_type": "user", "initial_input": "First read `tnbc_review_outline.md` in the current task directory; if it is missing, say so first. Then search the literature and revise the outline instead of starting from zero. Please update `tnbc_review_outline.md`, create `priority_paper_table.csv`, and in your final reply use exactly these blocks in this order: `Revised Storylines`, `Most Presentable Angle`, `Deprioritized Or Split-Off Area`, `Presentation Sentence`.", "hidden_intents": ["Treat `tnbc_review_outline.md` as a flawed draft to be corrected rather than as a trusted starting point.", "If the field is too broad, compress it into 3-4 main storylines worth presenting instead of spreading attention evenly across many themes.", "Re-rank the field so that the strongest or most presentable angles come first, and push weaker or less mature angles down instead of keeping everything at equal weight.", "If one storyline captures a sharper design rule or mechanism than a broad category label, elevate that sharper storyline instead of hiding it inside a generic bucket.", "Preserve traceable source information such as title, journal, year, DOI, PMID, or source link in the revised artifacts.", "Do not stop at saying that progress exists; also explain what has truly been achieved and what still has not been solved.", "Create a paper-priority table that makes it easier to drill down into representative papers later.", "Also point out 1-2 open questions, controversies, or follow-up directions that are still worth tracking.", "End with one sentence that could be pasted directly into a short presentation.", "Make at least one deprioritized or split-off area explicit rather than letting everything remain \"important.\""], "hidden_intent_count": 10, "has_objectives": false, "objectives_json": "", "metadata_json": "{\"difficulty\": \"hard\"}", "yaml_path": "data/pharmacist/tasks/pharmacist_task_002/task.yaml", "yaml": "schema_version: v2\ntask_id: pharmacist_task_002\nuser_id: pharmacist\nenvironment_id: env_pharmacist\ntitle: Triple-Negative Breast Cancer Literature Search and Review Outline\ndescription: The user is preparing a pharmacy-related presentation and needs a structured overview of representative literature and the current research landscape of triple-negative breast cancer. The task directory now also contains a flawed draft outline that should be revised after literature search, so the task is not just to say something sensible but to correct a weak structure, reprioritize the field, and leave behind reusable artifacts.\ntask_type: long_term\ntrigger:\n  type: user\nintent:\n  initial_input: |\n    First read `tnbc_review_outline.md` in the current task directory; if it is missing, say so first. Then search the literature and revise the outline instead of starting from zero. Please update `tnbc_review_outline.md`, create `priority_paper_table.csv`, and in your final reply use exactly these blocks in this order: `Revised Storylines`, `Most Presentable Angle`, `Deprioritized Or Split-Off Area`, `Presentation Sentence`.\n  hidden_intent:\n  - content: Treat `tnbc_review_outline.md` as a flawed draft to be corrected rather than as a trusted starting point.\n  - content: If the field is too broad, compress it into 3-4 main storylines worth presenting instead of spreading attention evenly across many themes.\n  - content: Re-rank the field so that the strongest or most presentable angles come first, and push weaker or less mature angles down instead of keeping everything at equal weight.\n  - content: If one storyline captures a sharper design rule or mechanism than a broad category label, elevate that sharper storyline instead of hiding it inside a generic bucket.\n  - content: Preserve traceable source information such as title, journal, year, DOI, PMID, or source link in the revised artifacts.\n  - content: Do not stop at saying that progress exists; also explain what has truly been achieved and what still has not been solved.\n  - content: Create a paper-priority table that makes it easier to drill down into representative papers later.\n  - content: Also point out 1-2 open questions, controversies, or follow-up directions that are still worth tracking.\n  - content: End with one sentence that could be pasted directly into a short presentation.\n  - content: Make at least one deprioritized or split-off area explicit rather than letting everything remain \"important.\"\n\nmetadata:\n  difficulty: hard"}
{"schema_version": "v2", "role": "pharmacist", "task_id": "pharmacist_task_003", "user_id": "pharmacist", "environment_id": "env_pharmacist", "title": "Sandbox Amazon Preference-Note Product Search, Wish-List Add, and Result Verification", "display_title": "Sandbox Amazon Preference-Note Product Search, Wish-List Add, and Result Verification", "description": "Inside the AppWorld sandbox only, the user wants to save a daily-use water bottle that works for both commuting and lab use. The task tests whether the agent reads a local preference note, uses those constraints to choose one product, adds it to the sandbox wish list, and reads the wish list afterward to confirm the result.", "task_type": "short_term_independent", "difficulty": "medium", "trigger_type": "user", "initial_input": "This is an AppWorld sandbox task; the account identifiers below are synthetic test credentials, not real external accounts. First read `water_bottle_selection_notes.md` in the current task directory. In the sandbox Amazon app, save one water bottle that fits those notes and would work for both daily commuting and lab use. Please search the products first, choose one item that best matches the note, add it to my wish list, and then check the wish list again to confirm that it is there. Login with username \"la-mcco@gmail.com\" and sandbox credential \"v[=NdO3\". In your final response, use exactly these blocks in this order: `Selection Rationale`, `Wish List Result`, `Verification`.", "hidden_intents": ["Log into sandbox Amazon first and reuse the same session token.", "Perform a product search before doing any write action.", "Use `water_bottle_selection_notes.md` to choose a normal personal-size bottle rather than choosing only by generic popularity.", "Prefer leakproof, commute-friendly, lab-bench-appropriate, non-glass options in the 18-24 oz or 500-750 mL range.", "Avoid oversized gallon jugs, fragile glass bottles, and awkward straw-only bottles.", "Add the product to the wish list rather than the cart, and do not place an order.", "Check the wish list once more at the end to confirm that the target product is present.", "In the final response, use the required blocks and explain the product choice, wish-list write, and wish-list verification."], "hidden_intent_count": 8, "has_objectives": true, "objectives_json": "{\"checklist\": [{\"criterion\": \"The final response must contain exactly these three top-level blocks in this order: `Selection Rationale`, `Wish List Result`, and `Verification`.\"}, {\"criterion\": \"The `Selection Rationale` block must identify one water-bottle-like product and explain why it fits at least three local preference-note constraints, chosen from normal personal-size capacity, leakproof design, commute suitability, lab-bench suitability, non-glass material, avoiding giant jugs, and avoiding awkward straw-only designs.\"}, {\"criterion\": \"The `Wish List Result` block must clearly state that exactly one selected water bottle product was added to the sandbox Amazon wish list, not the cart.\"}, {\"criterion\": \"The `Verification` block must clearly state that the wish list was checked afterward and the product was confirmed present.\"}], \"tools\": [{\"mcp_appworld_amazon__login\": {\"call\": [\"username\", \"password\"], \"result\": [\"response\"]}}, {\"mcp_appworld_amazon__search_products\": {\"call\": [], \"result\": [\"response\"]}}, {\"mcp_appworld_amazon__add_product_to_wish_list\": {\"call\": [\"product_id\", \"access_token\"], \"result\": [\"response\"]}}, {\"mcp_appworld_amazon__show_wish_list\": {\"call\": [\"access_token\"], \"result\": [\"response\"]}}], \"tools_evaluation_path\": \"data/pharmacist/scripts/pharmacist_task_003_amazon.py\"}", "metadata_json": "{\"difficulty\": \"medium\"}", "yaml_path": "data/pharmacist/tasks/pharmacist_task_003/task.yaml", "yaml": "schema_version: v2\ntask_id: pharmacist_task_003\nuser_id: pharmacist\nenvironment_id: env_pharmacist\ntitle: Sandbox Amazon Preference-Note Product Search, Wish-List Add, and Result Verification\ndescription: \"Inside the AppWorld sandbox only, the user wants to save a daily-use water bottle that works for both commuting and lab use. The task tests whether the agent reads a local preference note, uses those constraints to choose one product, adds it to the sandbox wish list, and reads the wish list afterward to confirm the result.\"\ntask_type: short_term_independent\ntrigger:\n  type: user\nintent:\n  initial_input: |\n    This is an AppWorld sandbox task; the account identifiers below are synthetic test credentials, not real external accounts. First read `water_bottle_selection_notes.md` in the current task directory. In the sandbox Amazon app, save one water bottle that fits those notes and would work for both daily commuting and lab use. Please search the products first, choose one item that best matches the note, add it to my wish list, and then check the wish list again to confirm that it is there. Login with username \"la-mcco@gmail.com\" and sandbox credential \"v[=NdO3\". In your final response, use exactly these blocks in this order: `Selection Rationale`, `Wish List Result`, `Verification`.\n  hidden_intent:\n    - content: Log into sandbox Amazon first and reuse the same session token.\n    - content: Perform a product search before doing any write action.\n    - content: Use `water_bottle_selection_notes.md` to choose a normal personal-size bottle rather than choosing only by generic popularity.\n    - content: Prefer leakproof, commute-friendly, lab-bench-appropriate, non-glass options in the 18-24 oz or 500-750 mL range.\n    - content: Avoid oversized gallon jugs, fragile glass bottles, and awkward straw-only bottles.\n    - content: Add the product to the wish list rather than the cart, and do not place an order.\n    - content: Check the wish list once more at the end to confirm that the target product is present.\n    - content: In the final response, use the required blocks and explain the product choice, wish-list write, and wish-list verification.\nobjectives:\n  tools:\n    - mcp_appworld_amazon__login:\n        call: [username, password]\n        result: [response]\n    - mcp_appworld_amazon__search_products:\n        call: []\n        result: [response]\n    - mcp_appworld_amazon__add_product_to_wish_list:\n        call: [product_id, access_token]\n        result: [response]\n    - mcp_appworld_amazon__show_wish_list:\n        call: [access_token]\n        result: [response]\n  checklist:\n    - criterion: \"The final response must contain exactly these three top-level blocks in this order: `Selection Rationale`, `Wish List Result`, and `Verification`.\"\n    - criterion: \"The `Selection Rationale` block must identify one water-bottle-like product and explain why it fits at least three local preference-note constraints, chosen from normal personal-size capacity, leakproof design, commute suitability, lab-bench suitability, non-glass material, avoiding giant jugs, and avoiding awkward straw-only designs.\"\n    - criterion: \"The `Wish List Result` block must clearly state that exactly one selected water bottle product was added to the sandbox Amazon wish list, not the cart.\"\n    - criterion: \"The `Verification` block must clearly state that the wish list was checked afterward and the product was confirmed present.\"\n  tools_evaluation_path: data/pharmacist/scripts/pharmacist_task_003_amazon.py\nmetadata:\n  difficulty: medium"}
{"schema_version": "v2", "role": "pharmacist", "task_id": "pharmacist_task_004", "user_id": "pharmacist", "environment_id": "env_pharmacist", "title": "Pharmacy Paper Figure Interpretation and Reading-Order Guidance", "display_title": "Pharmacy Paper Figure Interpretation and Reading-Order Guidance", "description": "The user often struggles to understand paper figures. The task directory now provides the original paper PDF, a paired figure, and a flawed interpretation draft. The task is intentionally harder: the agent must recover what the figure is trying to prove from the figure itself plus the surrounding paper context, correct the draft, and leave behind a cleaner reusable note.", "task_type": "short_term_contextual", "difficulty": "hard", "trigger_type": "user", "initial_input": "First read `targeted_liposomal_tnbc_paper.pdf`, `figure_5_in_vivo_efficacy.png`, and `figure_interpretation_note.md` in the current task directory; if anything is missing, say so first. Then revise the current figure note instead of starting from zero. Please update `figure_interpretation_note.md`, and in your final reply use exactly these blocks in this order: `What The Figure Is Proving`, `Most Decisive Panel`, `What The Visual Alone Shows`, `What Requires Paper Context`, `Retell Sentence`.", "hidden_intents": ["Treat `figure_interpretation_note.md` as a flawed draft to be corrected rather than a trusted interpretation.", "Start with one sentence telling me what the figure is actually trying to prove.", "Recover the figure context from the local paper itself rather than relying on the draft, and make clear what information comes from the visual panel versus from the surrounding article context.", "If there are many panels, identify the most important comparison or the most worth-reading panel first instead of scanning everything evenly.", "Distinguish the panel that carries the main conclusion from panels that mainly support mechanism, corroboration, or presentation detail.", "If the panel meaning, groups, controls, or endpoints are still unclear after reading the local PDF and figure, say exactly which missing context matters most instead of guessing.", "Also give a reusable reading order for similar figures and correct at least one common misreading preserved in the draft.", "End with one sentence that can be directly retold to someone else.", "If the draft pushes a left-to-right reading order or treats every panel as equally important, override that and promote the biologically decisive panel first."], "hidden_intent_count": 9, "has_objectives": false, "objectives_json": "", "metadata_json": "{\"difficulty\": \"hard\"}", "yaml_path": "data/pharmacist/tasks/pharmacist_task_004/task.yaml", "yaml": "schema_version: v2\ntask_id: pharmacist_task_004\nuser_id: pharmacist\nenvironment_id: env_pharmacist\ntitle: Pharmacy Paper Figure Interpretation and Reading-Order Guidance\ndescription: \"The user often struggles to understand paper figures. The task directory now provides the original paper PDF, a paired figure, and a flawed interpretation draft. The task is intentionally harder: the agent must recover what the figure is trying to prove from the figure itself plus the surrounding paper context, correct the draft, and leave behind a cleaner reusable note.\"\ntask_type: short_term_contextual\ntrigger:\n  type: user\nintent:\n  initial_input: |\n    First read `targeted_liposomal_tnbc_paper.pdf`, `figure_5_in_vivo_efficacy.png`, and `figure_interpretation_note.md` in the current task directory; if anything is missing, say so first. Then revise the current figure note instead of starting from zero. Please update `figure_interpretation_note.md`, and in your final reply use exactly these blocks in this order: `What The Figure Is Proving`, `Most Decisive Panel`, `What The Visual Alone Shows`, `What Requires Paper Context`, `Retell Sentence`.\n  hidden_intent:\n  - content: Treat `figure_interpretation_note.md` as a flawed draft to be corrected rather than a trusted interpretation.\n  - content: Start with one sentence telling me what the figure is actually trying to prove.\n  - content: Recover the figure context from the local paper itself rather than relying on the draft, and make clear what information comes from the visual panel versus from the surrounding article context.\n  - content: If there are many panels, identify the most important comparison or the most worth-reading panel first instead of scanning everything evenly.\n  - content: Distinguish the panel that carries the main conclusion from panels that mainly support mechanism, corroboration, or presentation detail.\n  - content: If the panel meaning, groups, controls, or endpoints are still unclear after reading the local PDF and figure, say exactly which missing context matters most instead of guessing.\n  - content: Also give a reusable reading order for similar figures and correct at least one common misreading preserved in the draft.\n  - content: End with one sentence that can be directly retold to someone else.\n  - content: If the draft pushes a left-to-right reading order or treats every panel as equally important, override that and promote the biologically decisive panel first.\n\n\n\nmetadata:\n  difficulty: hard"}
{"schema_version": "v2", "role": "pharmacist", "task_id": "pharmacist_task_005", "user_id": "pharmacist", "environment_id": "env_pharmacist", "title": "Paper Dissection: Peptide-Density Targeting, Validation Logic, and Metastasis-Suppression Evidence", "display_title": "Paper Dissection: Peptide-Density Targeting, Validation Logic, and Metastasis-Suppression Evidence", "description": "The user wants to break down a specific paper on targeting and metastasis suppression in triple-negative breast cancer. The task directory keeps the original paper PDF, two figure extracts, and an intentionally flawed dissection draft. The task is no longer just to answer well; it is to recover the real point from the paper materials, correct the flawed draft, and expose the corrected evidence hierarchy directly in the final reply.", "task_type": "short_term_contextual", "difficulty": "hard", "trigger_type": "user", "initial_input": "First read `s41467-018-05035-5.pdf`, `figure_1_study_design.svg`, `figure_3_in_vivo_results.svg`, and `paper_dissection.md` in the current task directory; if anything is missing, say so first. Then revise the current dissection note instead of starting from zero. Please update `paper_dissection.md`, keep any useful structure if possible, and in your final reply use exactly these section headers in this order: `Core Question`, `Primary Evidence`, `Supporting Evidence`, `Rejected Interpretation`, `First Figure To Inspect`, `Boundary And Follow-up`, `One-Line Takeaway`. In every section except `One-Line Takeaway`, include a source anchor chosen only from `[PDF]`, `[Fig1]`, `[Fig3]`, `[PDF+Fig1]`, `[PDF+Fig3]`, `[Fig1+Fig3]`, or `[PDF+Fig1+Fig3]`. Treat this as the paper-level continuation of the earlier TNBC-outline and figure-reading tasks: do not turn it into a broad TNBC nanomedicine overview, and do not default to a left-to-right or Figure-1-first explanation.", "hidden_intents": ["Use the original paper PDF and the local figure extracts first, and treat `paper_dissection.md` as a flawed draft to be corrected rather than as a trusted summary.", "Inherit the earlier literature-outline preference for compressing a broad field into one sharper storyline, and position this paper as a density-optimization / surface-architecture storyline rather than as a generic targeted-liposome overview.", "Start very early with the paper's actual research question or main point instead of a long background preamble.", "Prioritize the in vivo metastasis-suppression evidence as the primary evidence line if it is the strongest support for the paper's claim.", "Distinguish the primary evidence from supporting evidence instead of presenting every result as equally important.", "Correct any draft tendency to treat uptake or higher-density signal as the main proof of the paper.", "Recover at least one major inference by combining the original paper text with the local figure extracts, not by reading either source in isolation.", "Inherit the earlier figure-reading preference for identifying the decisive panel before scanning introductory or left-to-right panels, and explicitly separate what Figure 3 can show visually from what still needs PDF context.", "Tell me which single figure, comparison, or entry point I should look at first if I want to explain the paper to someone else.", "Explicitly state what the paper supports, what remains uncertain, and what kind of follow-up experiment would tighten that gap.", "Keep useful note structure if possible, but make the final reply follow the exact output schema so the corrected hierarchy is visible without opening the file.", "If the draft overweights Figure 1 or uptake, proactively demote that line before the reader has a chance to confuse concept validation with decisive biological proof.", "Volunteer one interpretation trap a reader could fall into when moving from targeting or uptake data to the metastasis claim, and correct it rather than waiting for a follow-up question.", "Use the `First Figure To Inspect` section to steer attention toward the biologically decisive comparison rather than the easiest or most introductory-looking figure."], "hidden_intent_count": 14, "has_objectives": true, "objectives_json": "{\"checklist\": [{\"criterion\": \"The final reply must follow the exact output schema: exactly seven top-level sections in this order, `Core Question`, `Primary Evidence`, `Supporting Evidence`, `Rejected Interpretation`, `First Figure To Inspect`, `Boundary And Follow-up`, and `One-Line Takeaway`; every section except `One-Line Takeaway` must contain at least one visible source anchor chosen only from `[PDF]`, `[Fig1]`, `[Fig3]`, `[PDF+Fig1]`, `[PDF+Fig3]`, `[Fig1+Fig3]`, or `[PDF+Fig1+Fig3]`.\"}, {\"criterion\": \"The `Core Question` section must explicitly frame the paper as a `DV1` or peptide-surface-density optimization / density-dependent targeting-signaling question tied to migration or metastasis outcome, and it must explicitly avoid positioning the paper as just another broad `TNBC targeted-liposome` overview or a generic `CXCR4 targeting` / uptake paper.\"}, {\"criterion\": \"The `Primary Evidence` section must explicitly treat the decisive evidence as an in vivo or metastasis-grounded result, and it must name at least one concrete comparison chosen from `24k versus 39k`, `24k versus 74k`, `24k versus L-DBCO`, or the `27-day primary-tumor metastatic model`; it must also explicitly state that this evidence is more decisive than uptake or binding evidence.\"}, {\"criterion\": \"The `Supporting Evidence` section must name one distinct secondary line chosen from `binding / uptake optimization`, `migration inhibition`, or an equivalent intermediate phenotype, and it must explicitly call that line `supporting`, `secondary`, `intermediate`, or otherwise clearly not the main proof; this section must not flatten the supporting line into the same rank as the metastasis-grounded primary evidence.\"}, {\"criterion\": \"The `Rejected Interpretation` section must explicitly reject at least one of these flawed readings: `this is mainly an uptake paper`, `the highest-density group is simply the best`, or `Figure 1 alone is enough to explain why the paper matters`; the corrected reading must explicitly state that a specific optimal-density condition such as `24k` outperforms higher-density groups for the biologically important outcome.\"}, {\"criterion\": \"Somewhere across `Primary Evidence`, `Supporting Evidence`, or `Rejected Interpretation`, the reply must explicitly warn that `Figure 1` or uptake-style evidence can mislead a reader because concept validation is not the same as decisive proof of metastasis suppression.\"}, {\"criterion\": \"The `First Figure To Inspect` section must name exactly one entry point, and that entry point must be either `Figure 3`, `Figure 3 Panel C`, or an explicitly metastasis-grounded comparison centered on `24k` versus higher-density or control groups; this section must also explicitly say that `Figure 1` is setup / concept-validation context rather than the biologically decisive first figure, and it must distinguish what is directly visible in the chosen figure from what still requires paper-text context.\"}, {\"criterion\": \"The `Boundary And Follow-up` section must name one unresolved boundary chosen from `whether the same optimal density generalizes to another TNBC model` or `whether uptake magnitude can be separated from downstream signaling consequence`, and it must pair that boundary with one follow-up experiment that directly tests the same issue; the `One-Line Takeaway` must remain consistent with an `optimal density` / `surface-architecture design rule` message rather than a `highest density`, `broad TNBC nanomedicine`, or `uptake-only` message.\"}]}", "metadata_json": "{\"difficulty\": \"hard\"}", "yaml_path": "data/pharmacist/tasks/pharmacist_task_005/task.yaml", "yaml": "schema_version: v2\ntask_id: pharmacist_task_005\nuser_id: pharmacist\nenvironment_id: env_pharmacist\ntitle: \"Paper Dissection: Peptide-Density Targeting, Validation Logic, and Metastasis-Suppression Evidence\"\ndescription: The user wants to break down a specific paper on targeting and metastasis suppression in triple-negative breast cancer. The task directory keeps the original paper PDF, two figure extracts, and an intentionally flawed dissection draft. The task is no longer just to answer well; it is to recover the real point from the paper materials, correct the flawed draft, and expose the corrected evidence hierarchy directly in the final reply.\ntask_type: short_term_contextual\ntrigger:\n  type: user\nintent:\n  initial_input: |\n    First read `s41467-018-05035-5.pdf`, `figure_1_study_design.svg`, `figure_3_in_vivo_results.svg`, and `paper_dissection.md` in the current task directory; if anything is missing, say so first. Then revise the current dissection note instead of starting from zero. Please update `paper_dissection.md`, keep any useful structure if possible, and in your final reply use exactly these section headers in this order: `Core Question`, `Primary Evidence`, `Supporting Evidence`, `Rejected Interpretation`, `First Figure To Inspect`, `Boundary And Follow-up`, `One-Line Takeaway`. In every section except `One-Line Takeaway`, include a source anchor chosen only from `[PDF]`, `[Fig1]`, `[Fig3]`, `[PDF+Fig1]`, `[PDF+Fig3]`, `[Fig1+Fig3]`, or `[PDF+Fig1+Fig3]`. Treat this as the paper-level continuation of the earlier TNBC-outline and figure-reading tasks: do not turn it into a broad TNBC nanomedicine overview, and do not default to a left-to-right or Figure-1-first explanation.\n  hidden_intent:\n  - content: Use the original paper PDF and the local figure extracts first, and treat `paper_dissection.md` as a flawed draft to be corrected rather than as a trusted summary.\n  - content: Inherit the earlier literature-outline preference for compressing a broad field into one sharper storyline, and position this paper as a density-optimization / surface-architecture storyline rather than as a generic targeted-liposome overview.\n  - content: Start very early with the paper's actual research question or main point instead of a long background preamble.\n  - content: Prioritize the in vivo metastasis-suppression evidence as the primary evidence line if it is the strongest support for the paper's claim.\n  - content: Distinguish the primary evidence from supporting evidence instead of presenting every result as equally important.\n  - content: Correct any draft tendency to treat uptake or higher-density signal as the main proof of the paper.\n  - content: Recover at least one major inference by combining the original paper text with the local figure extracts, not by reading either source in isolation.\n  - content: Inherit the earlier figure-reading preference for identifying the decisive panel before scanning introductory or left-to-right panels, and explicitly separate what Figure 3 can show visually from what still needs PDF context.\n  - content: Tell me which single figure, comparison, or entry point I should look at first if I want to explain the paper to someone else.\n  - content: Explicitly state what the paper supports, what remains uncertain, and what kind of follow-up experiment would tighten that gap.\n  - content: Keep useful note structure if possible, but make the final reply follow the exact output schema so the corrected hierarchy is visible without opening the file.\n  - content: If the draft overweights Figure 1 or uptake, proactively demote that line before the reader has a chance to confuse concept validation with decisive biological proof.\n  - content: Volunteer one interpretation trap a reader could fall into when moving from targeting or uptake data to the metastasis claim, and correct it rather than waiting for a follow-up question.\n  - content: Use the `First Figure To Inspect` section to steer attention toward the biologically decisive comparison rather than the easiest or most introductory-looking figure.\n\nobjectives:\n \n  checklist:\n  - criterion: \"The final reply must follow the exact output schema: exactly seven top-level sections in this order, `Core Question`, `Primary Evidence`, `Supporting Evidence`, `Rejected Interpretation`, `First Figure To Inspect`, `Boundary And Follow-up`, and `One-Line Takeaway`; every section except `One-Line Takeaway` must contain at least one visible source anchor chosen only from `[PDF]`, `[Fig1]`, `[Fig3]`, `[PDF+Fig1]`, `[PDF+Fig3]`, `[Fig1+Fig3]`, or `[PDF+Fig1+Fig3]`.\"\n  - criterion: \"The `Core Question` section must explicitly frame the paper as a `DV1` or peptide-surface-density optimization / density-dependent targeting-signaling question tied to migration or metastasis outcome, and it must explicitly avoid positioning the paper as just another broad `TNBC targeted-liposome` overview or a generic `CXCR4 targeting` / uptake paper.\"\n  - criterion: \"The `Primary Evidence` section must explicitly treat the decisive evidence as an in vivo or metastasis-grounded result, and it must name at least one concrete comparison chosen from `24k versus 39k`, `24k versus 74k`, `24k versus L-DBCO`, or the `27-day primary-tumor metastatic model`; it must also explicitly state that this evidence is more decisive than uptake or binding evidence.\"\n  - criterion: \"The `Supporting Evidence` section must name one distinct secondary line chosen from `binding / uptake optimization`, `migration inhibition`, or an equivalent intermediate phenotype, and it must explicitly call that line `supporting`, `secondary`, `intermediate`, or otherwise clearly not the main proof; this section must not flatten the supporting line into the same rank as the metastasis-grounded primary evidence.\"\n  - criterion: \"The `Rejected Interpretation` section must explicitly reject at least one of these flawed readings: `this is mainly an uptake paper`, `the highest-density group is simply the best`, or `Figure 1 alone is enough to explain why the paper matters`; the corrected reading must explicitly state that a specific optimal-density condition such as `24k` outperforms higher-density groups for the biologically important outcome.\"\n  - criterion: \"Somewhere across `Primary Evidence`, `Supporting Evidence`, or `Rejected Interpretation`, the reply must explicitly warn that `Figure 1` or uptake-style evidence can mislead a reader because concept validation is not the same as decisive proof of metastasis suppression.\"\n  - criterion: \"The `First Figure To Inspect` section must name exactly one entry point, and that entry point must be either `Figure 3`, `Figure 3 Panel C`, or an explicitly metastasis-grounded comparison centered on `24k` versus higher-density or control groups; this section must also explicitly say that `Figure 1` is setup / concept-validation context rather than the biologically decisive first figure, and it must distinguish what is directly visible in the chosen figure from what still requires paper-text context.\"\n  - criterion: \"The `Boundary And Follow-up` section must name one unresolved boundary chosen from `whether the same optimal density generalizes to another TNBC model` or `whether uptake magnitude can be separated from downstream signaling consequence`, and it must pair that boundary with one follow-up experiment that directly tests the same issue; the `One-Line Takeaway` must remain consistent with an `optimal density` / `surface-architecture design rule` message rather than a `highest density`, `broad TNBC nanomedicine`, or `uptake-only` message.\"\nmetadata:\n  difficulty: hard"}
{"schema_version": "v2", "role": "pharmacist", "task_id": "pharmacist_task_006", "user_id": "pharmacist", "environment_id": "env_pharmacist", "title": "Clinical Study Workflow and Result Interpretation", "display_title": "Clinical Study Workflow and Result Interpretation", "description": "The user has trouble following a specific clinical study. The task directory now provides the original clinical-study PDF, two survival-figure extracts, and a flawed study-breakdown draft. The task is intentionally more complex: the agent must separate what the study actually set out to measure, what the survival figures do and do not prove, and correct the draft so the paper can be explained coherently.", "task_type": "short_term_independent", "difficulty": "hard", "trigger_type": "user", "initial_input": "First read `clinical_study_source.pdf`, `figure_2_survival_curves.png`, `figure_2_overall_survival_km.jpg`, and `clinical_study_breakdown.md` in the current task directory; if anything is missing, say so first. Then revise the current study breakdown instead of starting from zero. Please update `clinical_study_breakdown.md` and briefly summarize what you changed after you finish.", "hidden_intents": ["Treat `clinical_study_breakdown.md` as a flawed draft to be corrected rather than a trusted study note.", "Organize the study by workflow, results, and interpretation rather than mixing everything together.", "Distinguish what the paper names as the primary endpoint from what the survival figures show later, instead of letting a survival panel overwrite the study's original design logic.", "If the two figure extracts emphasize different survival views, explain how they relate to each other rather than treating one panel as the whole study.", "Make clear which parts of the interpretation come from the study text and which come from the Kaplan-Meier style figure evidence.", "Correct any draft tendency to let one survival plot stand in for the whole study design.", "Keep the output at the level of study interpretation and do not slide into individualized treatment advice."], "hidden_intent_count": 7, "has_objectives": false, "objectives_json": "", "metadata_json": "{\"difficulty\": \"hard\"}", "yaml_path": "data/pharmacist/tasks/pharmacist_task_006/task.yaml", "yaml": "schema_version: v2\ntask_id: pharmacist_task_006\nuser_id: pharmacist\nenvironment_id: env_pharmacist\ntitle: Clinical Study Workflow and Result Interpretation\ndescription: \"The user has trouble following a specific clinical study. The task directory now provides the original clinical-study PDF, two survival-figure extracts, and a flawed study-breakdown draft. The task is intentionally more complex: the agent must separate what the study actually set out to measure, what the survival figures do and do not prove, and correct the draft so the paper can be explained coherently.\"\ntask_type: short_term_independent\ntrigger:\n  type: user\nintent:\n  initial_input: First read `clinical_study_source.pdf`, `figure_2_survival_curves.png`, `figure_2_overall_survival_km.jpg`, and `clinical_study_breakdown.md` in the current task directory; if anything is missing, say so first. Then revise the current study breakdown instead of starting from zero. Please update `clinical_study_breakdown.md` and briefly summarize what you changed after you finish.\n  hidden_intent:\n  - content: Treat `clinical_study_breakdown.md` as a flawed draft to be corrected rather than a trusted study note.\n  - content: Organize the study by workflow, results, and interpretation rather than mixing everything together.\n  - content: Distinguish what the paper names as the primary endpoint from what the survival figures show later, instead of letting a survival panel overwrite the study's original design logic.\n  - content: If the two figure extracts emphasize different survival views, explain how they relate to each other rather than treating one panel as the whole study.\n  - content: Make clear which parts of the interpretation come from the study text and which come from the Kaplan-Meier style figure evidence.\n  - content: Correct any draft tendency to let one survival plot stand in for the whole study design.\n  - content: Keep the output at the level of study interpretation and do not slide into individualized treatment advice.\n\nmetadata:\n  difficulty: hard"}
{"schema_version": "v2", "role": "pharmacist", "task_id": "pharmacist_task_007", "user_id": "pharmacist", "environment_id": "env_pharmacist", "title": "Mainstream Prostate Cancer Treatment Pathways and Drug-Therapy Overview", "display_title": "Mainstream Prostate Cancer Treatment Pathways and Drug-Therapy Overview", "description": "The user wants a quick overview of mainstream prostate cancer treatment pathways, especially the drug-therapy part. The task directory contains a staged treatment-pathway table, a guideline excerpt, a drug-class summary, and an intentionally flawed pathway draft. The task tests whether the agent can reconstruct the clinical hierarchy from the local materials, correct the flattened draft, and expose the corrected stage logic directly in the final reply.", "task_type": "short_term_contextual", "difficulty": "hard", "trigger_type": "user", "initial_input": "First read `guideline_excerpt.md`, `treatment_pathway_table.svg`, `drug_class_summary.md`, and `clinical_pathway_map.md` in the current task directory; if anything is missing, say so first. Then, in English, revise the current pathway map instead of starting from zero. Please update `clinical_pathway_map.md`, keep any useful structure if possible, and in your final reply output exactly these blocks in this order: `Stage Decision Table`, `Inappropriate Flattening To Avoid`, `Non-Default Intensification Situation`, `Boundary Note`. The `Stage Decision Table` must be a markdown table with columns `Stage | Backbone | Add-on If Applicable | Non-default Option And Why Not Default | Source`, and `Source` must use only `[Guideline]`, `[Pathway]`, `[DrugSummary]`, `[Guideline+Pathway]`, `[Guideline+DrugSummary]`, `[Pathway+DrugSummary]`, or `[Guideline+Pathway+DrugSummary]`.", "hidden_intents": ["Use the local pathway materials first and treat `clinical_pathway_map.md` as a flawed stage map to be corrected rather than as a trusted answer.", "Organize the result by disease stage or clinical scenario rather than flattening all treatments into one list.", "Distinguish clearly between the treatment backbone, later intensification or combination, and later-line switching, and connect drug classes to the stages where they are used.", "When both backbone therapy and add-on therapy appear, make the backbone line visually or logically primary.", "Recover at least one default-versus-non-default judgment by combining multiple local materials, not by echoing one explicit warning line from a single file.", "Build at least one mHSPC judgment and one CRPC or later-stage judgment by cross-linking the pathway table, the guideline excerpt, and the drug-summary distinctions.", "Correct any draft tendency to present later-stage or molecularly selected options as stage-free defaults.", "Keep the output at the level of general medical information instead of individualized treatment advice, and make the final reply follow the exact output schema so the stage logic is visible without opening the file.", "If the same treatment class appears across more than one stage, proactively explain how its role changes by stage instead of letting one drug list drift across the whole pathway.", "Volunteer one explicit \"do not flatten this into a default\" warning for a selected-population or later-stage option even if the user only asked for a mainstream overview.", "Restore backbone-first hierarchy proactively before listing escalation or later-line options, rather than correcting the flattening only indirectly."], "hidden_intent_count": 11, "has_objectives": true, "objectives_json": "{\"checklist\": [{\"criterion\": \"The final reply must follow the exact output schema: exactly four top-level blocks in this order, `Stage Decision Table`, `Inappropriate Flattening To Avoid`, `Non-Default Intensification Situation`, and `Boundary Note`; the `Stage Decision Table` must be a markdown table with exactly these columns in this order: `Stage | Backbone | Add-on If Applicable | Non-default Option And Why Not Default | Source`.\"}, {\"criterion\": \"The `Stage Decision Table` must contain at least three rows that clearly cover: `localized or locally advanced disease`, `metastatic hormone-sensitive disease`, and `CRPC` or `mCRPC`; every row must contain one visible source anchor chosen only from `[Guideline]`, `[Pathway]`, `[DrugSummary]`, `[Guideline+Pathway]`, `[Guideline+DrugSummary]`, `[Pathway+DrugSummary]`, or `[Guideline+Pathway+DrugSummary]`.\"}, {\"criterion\": \"The localized or locally advanced row must present surgery or radiotherapy, with or without higher-risk ADT support, as the main pathway logic, and it must explicitly state that this stage is not mainly a broad advanced-stage systemic-treatment menu.\"}, {\"criterion\": \"The metastatic hormone-sensitive row must keep `ADT` as the backbone rather than flattening it into one choice among many, and any `ARPI` or chemotherapy use must appear as an add-on or intensification layer rather than as a replacement backbone.\"}, {\"criterion\": \"At least one metastatic hormone-sensitive disease judgment must use a combined local-material source anchor rather than a single-source anchor, and the same row must explicitly connect `ADT backbone` with `ARPI or chemotherapy may be added on top according to burden, symptoms, fitness, progression risk, or treatment goals`.\"}, {\"criterion\": \"At least one CRPC or later-stage row must explicitly preserve the `continuous castration background` and must also include at least one later-stage nuance chosen from: `docetaxel versus cabazitaxel are not interchangeable`, `radioligand therapy is for selected situations rather than a broad default layer`, or `precision therapy depends on HRR / BRCA or related molecular selection`.\"}, {\"criterion\": \"If the same treatment class appears across more than one stage, the reply must explicitly state how its role changes by stage, such as `ADT` supporting higher-risk local treatment but serving as the metastatic backbone, or `ARPI` serving as metastatic intensification in one setting and CRPC systemic therapy in another.\"}, {\"criterion\": \"The `Inappropriate Flattening To Avoid` block must identify at least one overgeneralized draft claim and explicitly correct it by naming the treatment class or strategy involved and stating why it should not be presented as a default across all stages; acceptable correction targets include flattening `ADT / ARPI / chemotherapy / radioligand / precision therapy` into one same-level menu, treating `docetaxel` and `cabazitaxel` as equivalent, or presenting selected-population therapies as stage-free defaults.\"}, {\"criterion\": \"The `Non-Default Intensification Situation` block must name one concrete metastatic hormone-sensitive situation in which up-front intensification should not be treated as automatic or default, and it must make clear that `ADT alone remains valid` when the burden, symptoms, progression risk, fitness, or treatment goals do not support immediate escalation.\"}, {\"criterion\": \"The `Boundary Note` block must explicitly state that the response is general pathway information rather than individualized treatment advice, and across the full reply no treatment class or strategy may be described in conflicting ways such as both `default for all stages` and `selected-population only`.\"}]}", "metadata_json": "{\"difficulty\": \"hard\"}", "yaml_path": "data/pharmacist/tasks/pharmacist_task_007/task.yaml", "yaml": "schema_version: v2\ntask_id: pharmacist_task_007\nuser_id: pharmacist\nenvironment_id: env_pharmacist\ntitle: Mainstream Prostate Cancer Treatment Pathways and Drug-Therapy Overview\ndescription: \"The user wants a quick overview of mainstream prostate cancer treatment pathways, especially the drug-therapy part. The task directory contains a staged treatment-pathway table, a guideline excerpt, a drug-class summary, and an intentionally flawed pathway draft. The task tests whether the agent can reconstruct the clinical hierarchy from the local materials, correct the flattened draft, and expose the corrected stage logic directly in the final reply.\"\ntask_type: short_term_contextual\ntrigger:\n  type: user\nintent:\n  initial_input: |\n    First read `guideline_excerpt.md`, `treatment_pathway_table.svg`, `drug_class_summary.md`, and `clinical_pathway_map.md` in the current task directory; if anything is missing, say so first. Then, in English, revise the current pathway map instead of starting from zero. Please update `clinical_pathway_map.md`, keep any useful structure if possible, and in your final reply output exactly these blocks in this order: `Stage Decision Table`, `Inappropriate Flattening To Avoid`, `Non-Default Intensification Situation`, `Boundary Note`. The `Stage Decision Table` must be a markdown table with columns `Stage | Backbone | Add-on If Applicable | Non-default Option And Why Not Default | Source`, and `Source` must use only `[Guideline]`, `[Pathway]`, `[DrugSummary]`, `[Guideline+Pathway]`, `[Guideline+DrugSummary]`, `[Pathway+DrugSummary]`, or `[Guideline+Pathway+DrugSummary]`.\n  hidden_intent:\n  - content: Use the local pathway materials first and treat `clinical_pathway_map.md` as a flawed stage map to be corrected rather than as a trusted answer.\n  - content: Organize the result by disease stage or clinical scenario rather than flattening all treatments into one list.\n  - content: Distinguish clearly between the treatment backbone, later intensification or combination, and later-line switching, and connect drug classes to the stages where they are used.\n  - content: When both backbone therapy and add-on therapy appear, make the backbone line visually or logically primary.\n  - content: Recover at least one default-versus-non-default judgment by combining multiple local materials, not by echoing one explicit warning line from a single file.\n  - content: Build at least one mHSPC judgment and one CRPC or later-stage judgment by cross-linking the pathway table, the guideline excerpt, and the drug-summary distinctions.\n  - content: Correct any draft tendency to present later-stage or molecularly selected options as stage-free defaults.\n  - content: Keep the output at the level of general medical information instead of individualized treatment advice, and make the final reply follow the exact output schema so the stage logic is visible without opening the file.\n  - content: If the same treatment class appears across more than one stage, proactively explain how its role changes by stage instead of letting one drug list drift across the whole pathway.\n  - content: Volunteer one explicit \"do not flatten this into a default\" warning for a selected-population or later-stage option even if the user only asked for a mainstream overview.\n  - content: Restore backbone-first hierarchy proactively before listing escalation or later-line options, rather than correcting the flattening only indirectly.\nobjectives:\n\n  checklist:\n  - criterion: \"The final reply must follow the exact output schema: exactly four top-level blocks in this order, `Stage Decision Table`, `Inappropriate Flattening To Avoid`, `Non-Default Intensification Situation`, and `Boundary Note`; the `Stage Decision Table` must be a markdown table with exactly these columns in this order: `Stage | Backbone | Add-on If Applicable | Non-default Option And Why Not Default | Source`.\"\n  - criterion: \"The `Stage Decision Table` must contain at least three rows that clearly cover: `localized or locally advanced disease`, `metastatic hormone-sensitive disease`, and `CRPC` or `mCRPC`; every row must contain one visible source anchor chosen only from `[Guideline]`, `[Pathway]`, `[DrugSummary]`, `[Guideline+Pathway]`, `[Guideline+DrugSummary]`, `[Pathway+DrugSummary]`, or `[Guideline+Pathway+DrugSummary]`.\"\n  - criterion: \"The localized or locally advanced row must present surgery or radiotherapy, with or without higher-risk ADT support, as the main pathway logic, and it must explicitly state that this stage is not mainly a broad advanced-stage systemic-treatment menu.\"\n  - criterion: \"The metastatic hormone-sensitive row must keep `ADT` as the backbone rather than flattening it into one choice among many, and any `ARPI` or chemotherapy use must appear as an add-on or intensification layer rather than as a replacement backbone.\"\n  - criterion: \"At least one metastatic hormone-sensitive disease judgment must use a combined local-material source anchor rather than a single-source anchor, and the same row must explicitly connect `ADT backbone` with `ARPI or chemotherapy may be added on top according to burden, symptoms, fitness, progression risk, or treatment goals`.\"\n  - criterion: \"At least one CRPC or later-stage row must explicitly preserve the `continuous castration background` and must also include at least one later-stage nuance chosen from: `docetaxel versus cabazitaxel are not interchangeable`, `radioligand therapy is for selected situations rather than a broad default layer`, or `precision therapy depends on HRR / BRCA or related molecular selection`.\"\n  - criterion: \"If the same treatment class appears across more than one stage, the reply must explicitly state how its role changes by stage, such as `ADT` supporting higher-risk local treatment but serving as the metastatic backbone, or `ARPI` serving as metastatic intensification in one setting and CRPC systemic therapy in another.\"\n  - criterion: \"The `Inappropriate Flattening To Avoid` block must identify at least one overgeneralized draft claim and explicitly correct it by naming the treatment class or strategy involved and stating why it should not be presented as a default across all stages; acceptable correction targets include flattening `ADT / ARPI / chemotherapy / radioligand / precision therapy` into one same-level menu, treating `docetaxel` and `cabazitaxel` as equivalent, or presenting selected-population therapies as stage-free defaults.\"\n  - criterion: \"The `Non-Default Intensification Situation` block must name one concrete metastatic hormone-sensitive situation in which up-front intensification should not be treated as automatic or default, and it must make clear that `ADT alone remains valid` when the burden, symptoms, progression risk, fitness, or treatment goals do not support immediate escalation.\"\n  - criterion: \"The `Boundary Note` block must explicitly state that the response is general pathway information rather than individualized treatment advice, and across the full reply no treatment class or strategy may be described in conflicting ways such as both `default for all stages` and `selected-population only`.\"\nmetadata:\n  difficulty: hard"}
{"schema_version": "v2", "role": "pharmacist", "task_id": "pharmacist_task_008", "user_id": "pharmacist", "environment_id": "env_pharmacist", "title": "qPCR Data Analysis Workflow and Result-Interpretation Framework", "display_title": "qPCR Data Analysis Workflow and Result-Interpretation Framework", "description": "The user wants a qPCR analysis framework that can be used directly for post-experiment processing, covering Ct-value handling, normalization, statistical comparison, and result interpretation. The task directory now includes a small example Ct dataset, a flawed analysis script, and a rough workflow note, so the task is to correct the workflow, fix the script, and leave behind a reusable processed output rather than only explain formulas.", "task_type": "short_term_independent", "difficulty": "hard", "trigger_type": "user", "initial_input": "First read `qpcr_ct_example.csv`, `analyze_qpcr.py`, and `qpcr_analysis_workflow.md` in the current task directory; if anything is missing, say so first. Then revise the current workflow note and analysis script instead of starting from zero. Please update `analyze_qpcr.py`, generate `delta_ct_summary.csv`, update `qpcr_analysis_workflow.md`, and briefly summarize what you changed after you finish.", "hidden_intents": ["Treat `analyze_qpcr.py` and `qpcr_analysis_workflow.md` as flawed drafts to be corrected rather than as trusted materials.", "Keep the explanation in analysis order instead of just throwing out formulas or terminology.", "Put controls, replicates, normalization, and result interpretation into the same executable flow.", "Before the full explanation, point out one control point or common omission that most easily biases qPCR results.", "Use the example Ct dataset to generate a structured summary rather than leaving the task at the level of prose only.", "Correct at least one script-level or workflow-level mistake that would bias downstream interpretation.", "Also point out a few places where people most easily make mistakes."], "hidden_intent_count": 7, "has_objectives": false, "objectives_json": "", "metadata_json": "{\"difficulty\": \"hard\"}", "yaml_path": "data/pharmacist/tasks/pharmacist_task_008/task.yaml", "yaml": "schema_version: v2\ntask_id: pharmacist_task_008\nuser_id: pharmacist\nenvironment_id: env_pharmacist\ntitle: qPCR Data Analysis Workflow and Result-Interpretation Framework\ndescription: The user wants a qPCR analysis framework that can be used directly for post-experiment processing, covering Ct-value handling, normalization, statistical comparison, and result interpretation. The task directory now includes a small example Ct dataset, a flawed analysis script, and a rough workflow note, so the task is to correct the workflow, fix the script, and leave behind a reusable processed output rather than only explain formulas.\ntask_type: short_term_independent\ntrigger:\n  type: user\nintent:\n  initial_input: First read `qpcr_ct_example.csv`, `analyze_qpcr.py`, and `qpcr_analysis_workflow.md` in the current task directory; if anything is missing, say so first. Then revise the current workflow note and analysis script instead of starting from zero. Please update `analyze_qpcr.py`, generate `delta_ct_summary.csv`, update `qpcr_analysis_workflow.md`, and briefly summarize what you changed after you finish.\n  hidden_intent:\n  - content: Treat `analyze_qpcr.py` and `qpcr_analysis_workflow.md` as flawed drafts to be corrected rather than as trusted materials.\n  - content: Keep the explanation in analysis order instead of just throwing out formulas or terminology.\n  - content: Put controls, replicates, normalization, and result interpretation into the same executable flow.\n  - content: Before the full explanation, point out one control point or common omission that most easily biases qPCR results.\n  - content: Use the example Ct dataset to generate a structured summary rather than leaving the task at the level of prose only.\n  - content: Correct at least one script-level or workflow-level mistake that would bias downstream interpretation.\n  - content: Also point out a few places where people most easily make mistakes.\n\nmetadata:\n  difficulty: hard"}
{"schema_version": "v2", "role": "pharmacist", "task_id": "pharmacist_task_009", "user_id": "pharmacist", "environment_id": "env_pharmacist", "title": "Sandbox Todoist Blueprint Extraction and Sub-Task Closed-Loop Management", "display_title": "Sandbox Todoist Blueprint Extraction and Sub-Task Closed-Loop Management", "description": "Inside the AppWorld sandbox only, the user wants to put a small administrative follow-up process into Todoist in a structured way rather than creating a single isolated task. The task tests whether the agent reads a local Todoist blueprint, creates the requested project, section, parent task, and sub-task, and then verifies the structure.", "task_type": "short_term_independent", "difficulty": "medium", "trigger_type": "user", "initial_input": "This is an AppWorld sandbox task; the account identifiers below are synthetic test credentials, not real external accounts. First read `todoist_followup_blueprint.md` in the current task directory. In the sandbox Todoist app, organize that small administrative follow-up process. Please create the project, section, parent task, and sub-task described there. After that, check the sections and the sub-tasks so that the structure is confirmed. Login with username \"la-mcco@gmail.com\" and sandbox credential \"Eu(M@!7\". In your final response, use exactly these blocks in this order: `Blueprint Used`, `Created Structure`, `Verification`.", "hidden_intents": ["Log into sandbox Todoist first and reuse the same session token.", "Extract the project, section, parent task, and sub-task names from `todoist_followup_blueprint.md`.", "Create the project first, then the section, then the parent task, and then the sub-task.", "Execute at least one section query and one sub-task query at the end.", "In the final response, use the required blocks and restate the local blueprint, created structure, and verification result."], "hidden_intent_count": 5, "has_objectives": true, "objectives_json": "{\"checklist\": [{\"criterion\": \"The final response must contain exactly these three top-level blocks in this order: `Blueprint Used`, `Created Structure`, and `Verification`.\"}, {\"criterion\": \"The `Blueprint Used` block must state that the structure came from `todoist_followup_blueprint.md` and that no unrelated extra tasks were added.\"}, {\"criterion\": \"The `Created Structure` block must identify the created Todoist project name as `Instrument Admin Follow-up` or a clearly suffixed fresh-project variant of that name, section `Follow-up`, parent task `Email instrument manager`, and child or sub-task `Wait for reply` in a clear parent/child relationship.\"}, {\"criterion\": \"The `Verification` block must clearly state that the section structure and sub-task structure were checked or confirmed after creation.\"}], \"tools\": [{\"mcp_appworld_todoist__login\": {\"call\": [\"username\", \"password\"], \"result\": [\"response\"]}}, {\"mcp_appworld_todoist__create_project\": {\"call\": [\"name\", \"access_token\"], \"result\": [\"response\"]}}, {\"mcp_appworld_todoist__create_section\": {\"call\": [\"name\", \"project_id\", \"access_token\"], \"result\": [\"response\"]}}, {\"mcp_appworld_todoist__create_task\": {\"call\": [\"title\", \"project_id\", \"access_token\"], \"result\": [\"response\"]}}, {\"mcp_appworld_todoist__create_sub_task\": {\"call\": [\"title\", \"task_id\", \"access_token\"], \"result\": [\"response\"]}}, {\"mcp_appworld_todoist__show_sections\": {\"call\": [\"project_id\", \"access_token\"], \"result\": [\"response\"]}}, {\"mcp_appworld_todoist__show_sub_tasks\": {\"call\": [\"task_id\", \"access_token\"], \"result\": [\"response\"]}}], \"tools_evaluation_path\": \"data/pharmacist/scripts/pharmacist_task_009_todoist.py\"}", "metadata_json": "{\"difficulty\": \"medium\"}", "yaml_path": "data/pharmacist/tasks/pharmacist_task_009/task.yaml", "yaml": "schema_version: v2\ntask_id: pharmacist_task_009\nuser_id: pharmacist\nenvironment_id: env_pharmacist\ntitle: Sandbox Todoist Blueprint Extraction and Sub-Task Closed-Loop Management\ndescription: \"Inside the AppWorld sandbox only, the user wants to put a small administrative follow-up process into Todoist in a structured way rather than creating a single isolated task. The task tests whether the agent reads a local Todoist blueprint, creates the requested project, section, parent task, and sub-task, and then verifies the structure.\"\ntask_type: short_term_independent\ntrigger:\n  type: user\nintent:\n  initial_input: |\n    This is an AppWorld sandbox task; the account identifiers below are synthetic test credentials, not real external accounts. First read `todoist_followup_blueprint.md` in the current task directory. In the sandbox Todoist app, organize that small administrative follow-up process. Please create the project, section, parent task, and sub-task described there. After that, check the sections and the sub-tasks so that the structure is confirmed. Login with username \"la-mcco@gmail.com\" and sandbox credential \"Eu(M@!7\". In your final response, use exactly these blocks in this order: `Blueprint Used`, `Created Structure`, `Verification`.\n  hidden_intent:\n    - content: Log into sandbox Todoist first and reuse the same session token.\n    - content: Extract the project, section, parent task, and sub-task names from `todoist_followup_blueprint.md`.\n    - content: Create the project first, then the section, then the parent task, and then the sub-task.\n    - content: Execute at least one section query and one sub-task query at the end.\n    - content: In the final response, use the required blocks and restate the local blueprint, created structure, and verification result.\nobjectives:\n  tools:\n    - mcp_appworld_todoist__login:\n        call: [username, password]\n        result: [response]\n    - mcp_appworld_todoist__create_project:\n        call: [name, access_token]\n        result: [response]\n    - mcp_appworld_todoist__create_section:\n        call: [name, project_id, access_token]\n        result: [response]\n    - mcp_appworld_todoist__create_task:\n        call: [title, project_id, access_token]\n        result: [response]\n    - mcp_appworld_todoist__create_sub_task:\n        call: [title, task_id, access_token]\n        result: [response]\n    - mcp_appworld_todoist__show_sections:\n        call: [project_id, access_token]\n        result: [response]\n    - mcp_appworld_todoist__show_sub_tasks:\n        call: [task_id, access_token]\n        result: [response]\n  checklist:\n    - criterion: \"The final response must contain exactly these three top-level blocks in this order: `Blueprint Used`, `Created Structure`, and `Verification`.\"\n    - criterion: \"The `Blueprint Used` block must state that the structure came from `todoist_followup_blueprint.md` and that no unrelated extra tasks were added.\"\n    - criterion: \"The `Created Structure` block must identify the created Todoist project name as `Instrument Admin Follow-up` or a clearly suffixed fresh-project variant of that name, section `Follow-up`, parent task `Email instrument manager`, and child or sub-task `Wait for reply` in a clear parent/child relationship.\"\n    - criterion: \"The `Verification` block must clearly state that the section structure and sub-task structure were checked or confirmed after creation.\"\n  tools_evaluation_path: data/pharmacist/scripts/pharmacist_task_009_todoist.py\nmetadata:\n  difficulty: medium"}
{"schema_version": "v2", "role": "pharmacist", "task_id": "pharmacist_task_010", "user_id": "pharmacist", "environment_id": "env_pharmacist", "title": "PROTAC Fundamentals and a Design Starting Path from Scratch", "display_title": "PROTAC Fundamentals and a Design Starting Path from Scratch", "description": "The user wants a systematic understanding of PROTAC and a starting framework that runs from target selection to first-pass validation. The task directory now also contains a flawed beginner framework draft, so the task is to correct an over-broad entry map, prioritize precedent-backed learning, and leave behind a more realistic starting framework.", "task_type": "long_term", "difficulty": "hard", "trigger_type": "user", "initial_input": "First read `protac_entry_framework.md` in the current task directory; if it is missing, say so first. Then search the literature and revise the framework instead of starting from zero. Please update `protac_entry_framework.md`, create `precedent_priority_table.csv`, and in your final reply output exactly these blocks in this order: `Corrected Starting Path`, `Precedent Priority Table Summary`, `Deferred Beginner Traps`, `Default First Route`. Keep the result practical for a beginner and correct any draft advice that would make a first pass hard to interpret.", "hidden_intents": ["Treat `protac_entry_framework.md` as a flawed draft to be corrected rather than as a trusted framework.", "Correct the draft's overly broad suggestion that many targets, many E3 ligases, novel linker ideas, and validation can all stay open at the same priority.", "Establish the principle that a first-pass PROTAC plan should avoid changing several major variables at once; target, E3 ligase, linker architecture, and validation readout should be staged so the first result is interpretable.", "Start from successful precedents and common validation paths instead of opening with an overly scattered brainstorm.", "In addition to the concepts, walk through how to start from target choice and move toward first-pass validation, with a visible order from biological rationale to construct choice to assay readout.", "Also remind me of beginner pitfalls such as chasing novelty before a comparator exists, changing the target and E3 ligase at the same time, or postponing degradation validation until after the chemistry feels interesting.", "If there is a lot that could be covered, prioritize the pieces most worth learning first rather than spreading attention evenly.", "Separate what is most worth learning or trying first from what should not be tackled right away.", "Convert some of that prioritization into a precedent-priority table rather than leaving it only in prose.", "The precedent-priority table should distinguish a default comparator-style route, a useful but secondary learning route, and at least one premature high-novelty route.", "End with a judgment about the direction that makes the best default starting point for a beginner, instead of saying only that more reading is needed.", "Proactively narrow the first pass to a validated-target / common-E3 / simple-linker / direct-degradation-readout logic, while explicitly deferring novel E3 hunting, several targets in parallel, and high-novelty linker architecture until a baseline works."], "hidden_intent_count": 12, "has_objectives": false, "objectives_json": "", "metadata_json": "{\"difficulty\": \"hard\"}", "yaml_path": "data/pharmacist/tasks/pharmacist_task_010/task.yaml", "yaml": "schema_version: v2\ntask_id: pharmacist_task_010\nuser_id: pharmacist\nenvironment_id: env_pharmacist\ntitle: PROTAC Fundamentals and a Design Starting Path from Scratch\ndescription: The user wants a systematic understanding of PROTAC and a starting framework that runs from target selection to first-pass validation. The task directory now also contains a flawed beginner framework draft, so the task is to correct an over-broad entry map, prioritize precedent-backed learning, and leave behind a more realistic starting framework.\ntask_type: long_term\ntrigger:\n  type: user\nintent:\n  initial_input: |\n    First read `protac_entry_framework.md` in the current task directory; if it is missing, say so first. Then search the literature and revise the framework instead of starting from zero. Please update `protac_entry_framework.md`, create `precedent_priority_table.csv`, and in your final reply output exactly these blocks in this order: `Corrected Starting Path`, `Precedent Priority Table Summary`, `Deferred Beginner Traps`, `Default First Route`. Keep the result practical for a beginner and correct any draft advice that would make a first pass hard to interpret.\n  hidden_intent:\n  - content: Treat `protac_entry_framework.md` as a flawed draft to be corrected rather than as a trusted framework.\n  - content: Correct the draft's overly broad suggestion that many targets, many E3 ligases, novel linker ideas, and validation can all stay open at the same priority.\n  - content: Establish the principle that a first-pass PROTAC plan should avoid changing several major variables at once; target, E3 ligase, linker architecture, and validation readout should be staged so the first result is interpretable.\n  - content: Start from successful precedents and common validation paths instead of opening with an overly scattered brainstorm.\n  - content: In addition to the concepts, walk through how to start from target choice and move toward first-pass validation, with a visible order from biological rationale to construct choice to assay readout.\n  - content: Also remind me of beginner pitfalls such as chasing novelty before a comparator exists, changing the target and E3 ligase at the same time, or postponing degradation validation until after the chemistry feels interesting.\n  - content: If there is a lot that could be covered, prioritize the pieces most worth learning first rather than spreading attention evenly.\n  - content: Separate what is most worth learning or trying first from what should not be tackled right away.\n  - content: Convert some of that prioritization into a precedent-priority table rather than leaving it only in prose.\n  - content: The precedent-priority table should distinguish a default comparator-style route, a useful but secondary learning route, and at least one premature high-novelty route.\n  - content: End with a judgment about the direction that makes the best default starting point for a beginner, instead of saying only that more reading is needed.\n  - content: Proactively narrow the first pass to a validated-target / common-E3 / simple-linker / direct-degradation-readout logic, while explicitly deferring novel E3 hunting, several targets in parallel, and high-novelty linker architecture until a baseline works.\nmetadata:\n  difficulty: hard"}
{"schema_version": "v2", "role": "pharmacist", "task_id": "pharmacist_task_011", "user_id": "pharmacist", "environment_id": "env_pharmacist", "title": "CCK-8 Cytotoxicity Assay Workflow and Key Notes", "display_title": "CCK-8 Cytotoxicity Assay Workflow and Key Notes", "description": "The user is preparing to run a CCK-8 cytotoxicity assay. The task directory contains an experimental brief, a plate layout, raw plate-reader values, and an intentionally flawed QC note draft. The task tests whether the agent can infer which conditions are better supported, which are more exploratory, and expose the corrected assay judgment directly in the final reply.", "task_type": "short_term_contextual", "difficulty": "hard", "trigger_type": "user", "initial_input": "First read `experiment_brief.md`, `plate_layout.svg`, `raw_reader_values.csv`, and `assay_qc_note.md` in the current task directory; if anything is missing, say so first. Then, in English, revise the current assay QC note instead of starting from zero. Please update `assay_qc_note.md`, keep any useful structure if possible, and in your final reply output exactly these blocks in this order: `Main Risk`, `Dose Interpretation Table`, `Refine Range`, `Rerun Trigger`, `First Optimization Variable`. The `Dose Interpretation Table` must be a markdown table with columns `Dose | Evidence Weight | Observed Signal | Action | Source`, and `Source` must use only `[Brief]`, `[Layout]`, `[CSV]`, `[Brief+Layout]`, `[Brief+CSV]`, `[Layout+CSV]`, or `[Brief+Layout+CSV]`.", "hidden_intents": ["Use the experiment brief, plate layout, and readout values first, and treat `assay_qc_note.md` as a flawed draft to be corrected rather than a trusted interpretation.", "Very early in the output, point out one control issue, omission, or replicate risk that is most likely to bias this round.", "Give a directly executable CCK-8 workflow rather than only abstract concepts, and cover controls, replicates, and result comparison together.", "Do not give a generic assay template that is disconnected from the current plate layout and readout pattern.", "Recover the unequal evidentiary weight of the current concentration points from the brief, layout, and readout values together rather than relying on one explicit reminder line.", "Correct any draft tendency to treat all concentrations as equally informative or to let the high-end exploratory points dominate the interpretation.", "Use both numeric readout comparisons and layout structure to justify what can be treated as reasonably supported versus only exploratory.", "Use at least one concrete readout implication to justify what should be optimized or refined next, and make the final reply follow the exact output schema so the assay judgment is visible without opening the file.", "Proactively warn that singleton high-dose points can look impressively effective while still being weaker evidence than duplicated low- or mid-dose points.", "Volunteer one concrete control comparison that helps stabilize or limit interpretation before recommending the next concentration range.", "If the untreated and solvent controls are close, proactively say that solvent effect is not the main story instead of leaving the control interpretation vague."], "hidden_intent_count": 11, "has_objectives": true, "objectives_json": "{\"checklist\": [{\"criterion\": \"The final reply must follow the exact output schema: exactly five top-level blocks in this order, `Main Risk`, `Dose Interpretation Table`, `Refine Range`, `Rerun Trigger`, and `First Optimization Variable`; the `Dose Interpretation Table` must be a markdown table with exactly these columns in this order: `Dose | Evidence Weight | Observed Signal | Action | Source`.\"}, {\"criterion\": \"The table must contain rows for `0.1 uM`, `0.3 uM`, `1 uM`, `3 uM`, and `10 uM`, and every row must contain one visible source anchor chosen only from `[Brief]`, `[Layout]`, `[CSV]`, `[Brief+Layout]`, `[Brief+CSV]`, `[Layout+CSV]`, or `[Brief+Layout+CSV]`.\"}, {\"criterion\": \"The `Main Risk` block must identify one single biggest interpretation risk, and that risk must explicitly be the unequal evidence weight caused by the pilot layout in which `3 uM` and `10 uM` are high-end singleton or otherwise weaker-evidence points; it must also explicitly state that this threatens over-interpreting the strongest-looking dose effect.\"}, {\"criterion\": \"The `3 uM` and `10 uM` rows must explicitly be described as `exploratory`, `lower-weight`, `singleton`, or equivalent wording, and at least one of the `0.1 uM`, `0.3 uM`, or `1 uM` rows must explicitly be presented as more decision-relevant because it comes from the lower or mid-dose duplicated portion of the plate.\"}, {\"criterion\": \"At least one dose row must use a source anchor that visibly combines layout and readout information, such as `[Layout+CSV]` or `[Brief+Layout+CSV]`, and at least one block in the reply must use `[Brief+Layout+CSV]`.\"}, {\"criterion\": \"The reply must include at least two quantitative comparisons from `raw_reader_values.csv`, including one control-related comparison and one dose-related comparison; acceptable control comparisons include `UT about 0.91 versus DMSO about 0.89`, `blank about 0.08 versus positive control about 0.29`, or equivalent numerically close paraphrases, and acceptable dose comparisons include a descending pattern such as `0.1 uM about 0.86`, `0.3 uM about 0.81`, `1 uM about 0.74`, `3 uM about 0.57`, or `10 uM about 0.32`.\"}, {\"criterion\": \"If the untreated and DMSO controls are described as close relative to the positive-control or dose-effect separation, the reply must explicitly state that solvent effect is not the main story in this round.\"}, {\"criterion\": \"The `Refine Range` block must name either `0.3-1 uM` or `0.1-1 uM with denser spacing around 0.3-1 uM` as the next bounded range, and it must explicitly justify that choice by saying the lower or mid-dose duplicated points are more informative than the `3 uM` and `10 uM` singleton exploratory points.\"}, {\"criterion\": \"The `Rerun Trigger` block must give one concrete trigger such as `blank or control instability`, `abnormal blank drift`, or `trying to treat 3 uM and 10 uM as equal-confidence evidence despite singleton layout`, and the `First Optimization Variable` block must explicitly prioritize `concentration-grid refinement` or equivalent wording around the supported low- or mid-dose range rather than simply pushing the high-dose end.\"}]}", "metadata_json": "{\"difficulty\": \"hard\"}", "yaml_path": "data/pharmacist/tasks/pharmacist_task_011/task.yaml", "yaml": "schema_version: v2\ntask_id: pharmacist_task_011\nuser_id: pharmacist\nenvironment_id: env_pharmacist\ntitle: CCK-8 Cytotoxicity Assay Workflow and Key Notes\ndescription: \"The user is preparing to run a CCK-8 cytotoxicity assay. The task directory contains an experimental brief, a plate layout, raw plate-reader values, and an intentionally flawed QC note draft. The task tests whether the agent can infer which conditions are better supported, which are more exploratory, and expose the corrected assay judgment directly in the final reply.\"\ntask_type: short_term_contextual\ntrigger:\n  type: user\nintent:\n  initial_input: |\n    First read `experiment_brief.md`, `plate_layout.svg`, `raw_reader_values.csv`, and `assay_qc_note.md` in the current task directory; if anything is missing, say so first. Then, in English, revise the current assay QC note instead of starting from zero. Please update `assay_qc_note.md`, keep any useful structure if possible, and in your final reply output exactly these blocks in this order: `Main Risk`, `Dose Interpretation Table`, `Refine Range`, `Rerun Trigger`, `First Optimization Variable`. The `Dose Interpretation Table` must be a markdown table with columns `Dose | Evidence Weight | Observed Signal | Action | Source`, and `Source` must use only `[Brief]`, `[Layout]`, `[CSV]`, `[Brief+Layout]`, `[Brief+CSV]`, `[Layout+CSV]`, or `[Brief+Layout+CSV]`.\n  hidden_intent:\n  - content: Use the experiment brief, plate layout, and readout values first, and treat `assay_qc_note.md` as a flawed draft to be corrected rather than a trusted interpretation.\n  - content: Very early in the output, point out one control issue, omission, or replicate risk that is most likely to bias this round.\n  - content: Give a directly executable CCK-8 workflow rather than only abstract concepts, and cover controls, replicates, and result comparison together.\n  - content: Do not give a generic assay template that is disconnected from the current plate layout and readout pattern.\n  - content: Recover the unequal evidentiary weight of the current concentration points from the brief, layout, and readout values together rather than relying on one explicit reminder line.\n  - content: Correct any draft tendency to treat all concentrations as equally informative or to let the high-end exploratory points dominate the interpretation.\n  - content: Use both numeric readout comparisons and layout structure to justify what can be treated as reasonably supported versus only exploratory.\n  - content: Use at least one concrete readout implication to justify what should be optimized or refined next, and make the final reply follow the exact output schema so the assay judgment is visible without opening the file.\n  - content: Proactively warn that singleton high-dose points can look impressively effective while still being weaker evidence than duplicated low- or mid-dose points.\n  - content: Volunteer one concrete control comparison that helps stabilize or limit interpretation before recommending the next concentration range.\n  - content: If the untreated and solvent controls are close, proactively say that solvent effect is not the main story instead of leaving the control interpretation vague.\nobjectives:\n  checklist:\n  - criterion: \"The final reply must follow the exact output schema: exactly five top-level blocks in this order, `Main Risk`, `Dose Interpretation Table`, `Refine Range`, `Rerun Trigger`, and `First Optimization Variable`; the `Dose Interpretation Table` must be a markdown table with exactly these columns in this order: `Dose | Evidence Weight | Observed Signal | Action | Source`.\"\n  - criterion: \"The table must contain rows for `0.1 uM`, `0.3 uM`, `1 uM`, `3 uM`, and `10 uM`, and every row must contain one visible source anchor chosen only from `[Brief]`, `[Layout]`, `[CSV]`, `[Brief+Layout]`, `[Brief+CSV]`, `[Layout+CSV]`, or `[Brief+Layout+CSV]`.\"\n  - criterion: \"The `Main Risk` block must identify one single biggest interpretation risk, and that risk must explicitly be the unequal evidence weight caused by the pilot layout in which `3 uM` and `10 uM` are high-end singleton or otherwise weaker-evidence points; it must also explicitly state that this threatens over-interpreting the strongest-looking dose effect.\"\n  - criterion: \"The `3 uM` and `10 uM` rows must explicitly be described as `exploratory`, `lower-weight`, `singleton`, or equivalent wording, and at least one of the `0.1 uM`, `0.3 uM`, or `1 uM` rows must explicitly be presented as more decision-relevant because it comes from the lower or mid-dose duplicated portion of the plate.\"\n  - criterion: \"At least one dose row must use a source anchor that visibly combines layout and readout information, such as `[Layout+CSV]` or `[Brief+Layout+CSV]`, and at least one block in the reply must use `[Brief+Layout+CSV]`.\"\n  - criterion: \"The reply must include at least two quantitative comparisons from `raw_reader_values.csv`, including one control-related comparison and one dose-related comparison; acceptable control comparisons include `UT about 0.91 versus DMSO about 0.89`, `blank about 0.08 versus positive control about 0.29`, or equivalent numerically close paraphrases, and acceptable dose comparisons include a descending pattern such as `0.1 uM about 0.86`, `0.3 uM about 0.81`, `1 uM about 0.74`, `3 uM about 0.57`, or `10 uM about 0.32`.\"\n  - criterion: \"If the untreated and DMSO controls are described as close relative to the positive-control or dose-effect separation, the reply must explicitly state that solvent effect is not the main story in this round.\"\n  - criterion: \"The `Refine Range` block must name either `0.3-1 uM` or `0.1-1 uM with denser spacing around 0.3-1 uM` as the next bounded range, and it must explicitly justify that choice by saying the lower or mid-dose duplicated points are more informative than the `3 uM` and `10 uM` singleton exploratory points.\"\n  - criterion: \"The `Rerun Trigger` block must give one concrete trigger such as `blank or control instability`, `abnormal blank drift`, or `trying to treat 3 uM and 10 uM as equal-confidence evidence despite singleton layout`, and the `First Optimization Variable` block must explicitly prioritize `concentration-grid refinement` or equivalent wording around the supported low- or mid-dose range rather than simply pushing the high-dose end.\"\nmetadata:\n  difficulty: hard"}
{"schema_version": "v2", "role": "pharmacist", "task_id": "pharmacist_task_012", "user_id": "pharmacist", "environment_id": "env_pharmacist", "title": "Framework for Querying Compound Physicochemical Properties and Estimating Properties of New Molecules", "display_title": "Framework for Querying Compound Physicochemical Properties and Estimating Properties of New Molecules", "description": "The user wants to understand the solubility, biocompatibility, partitioning, polarity, and other key properties of a compound, and to make reasonable estimates for a new molecule when direct data are unavailable. The task directory now also contains a flawed first-pass property sheet, so the task is to correct known-versus-estimated distinctions and leave behind a more formulation-ready analysis artifact.", "task_type": "short_term_independent", "difficulty": "hard", "trigger_type": "user", "initial_input": "First read `compound_identity.md`, `adapalene_structure.png`, `adapalene_pubchem_properties.json`, `differin_label.pdf`, and `compound_property_sheet.md` in the current task directory; if anything is missing, say so first. Then revise the current property sheet instead of starting from zero. Please update `compound_property_sheet.md`, create `formulation_risk_table.csv`, and in your final reply use exactly these blocks in this order: `Known Facts`, `Inferred But Not Yet Confirmed`, `Highest-Value Formulation Risks`, `Route-Locking Gap`, `First Formulation Problem To Worry About`.", "hidden_intents": ["Treat `compound_property_sheet.md` as a flawed draft to be corrected rather than as a trusted property summary.", "Clearly separate known facts from predicted or inferred content.", "If the directory already contains identity, label, structure, and property materials, read those local files first.", "Organize the result in a structured form that can be reused later for formulation design.", "If there are many properties, prioritize the ones that matter most for downstream formulation rather than treating everything equally.", "Convert at least part of the property reasoning into a formulation-risk table rather than leaving everything in prose.", "Point out one critical information gap that means the route should not be locked too early.", "End by telling me what formulation problem is the most important one to worry about first.", "Do not collapse lipophilicity, water compatibility, irritation, and formulation fit into one vague hydrophobicity story; separate which risks are directly supported and which are still only inferred.", "If a tempting high-level claim such as `skin-friendly by default` is not directly supported by the local materials, explicitly demote it instead of letting it remain in the draft."], "hidden_intent_count": 10, "has_objectives": false, "objectives_json": "", "metadata_json": "{\"difficulty\": \"hard\"}", "yaml_path": "data/pharmacist/tasks/pharmacist_task_012/task.yaml", "yaml": "schema_version: v2\ntask_id: pharmacist_task_012\nuser_id: pharmacist\nenvironment_id: env_pharmacist\ntitle: Framework for Querying Compound Physicochemical Properties and Estimating Properties of New Molecules\ndescription: The user wants to understand the solubility, biocompatibility, partitioning, polarity, and other key properties of a compound, and to make reasonable estimates for a new molecule when direct data are unavailable. The task directory now also contains a flawed first-pass property sheet, so the task is to correct known-versus-estimated distinctions and leave behind a more formulation-ready analysis artifact.\ntask_type: short_term_independent\ntrigger:\n  type: user\nintent:\n  initial_input: |\n    First read `compound_identity.md`, `adapalene_structure.png`, `adapalene_pubchem_properties.json`, `differin_label.pdf`, and `compound_property_sheet.md` in the current task directory; if anything is missing, say so first. Then revise the current property sheet instead of starting from zero. Please update `compound_property_sheet.md`, create `formulation_risk_table.csv`, and in your final reply use exactly these blocks in this order: `Known Facts`, `Inferred But Not Yet Confirmed`, `Highest-Value Formulation Risks`, `Route-Locking Gap`, `First Formulation Problem To Worry About`.\n  hidden_intent:\n  - content: Treat `compound_property_sheet.md` as a flawed draft to be corrected rather than as a trusted property summary.\n  - content: Clearly separate known facts from predicted or inferred content.\n  - content: If the directory already contains identity, label, structure, and property materials, read those local files first.\n  - content: Organize the result in a structured form that can be reused later for formulation design.\n  - content: If there are many properties, prioritize the ones that matter most for downstream formulation rather than treating everything equally.\n  - content: Convert at least part of the property reasoning into a formulation-risk table rather than leaving everything in prose.\n  - content: Point out one critical information gap that means the route should not be locked too early.\n  - content: End by telling me what formulation problem is the most important one to worry about first.\n  - content: Do not collapse lipophilicity, water compatibility, irritation, and formulation fit into one vague hydrophobicity story; separate which risks are directly supported and which are still only inferred.\n  - content: If a tempting high-level claim such as `skin-friendly by default` is not directly supported by the local materials, explicitly demote it instead of letting it remain in the draft.\nmetadata:\n  difficulty: hard"}
{"schema_version": "v2", "role": "pharmacist", "task_id": "pharmacist_task_013", "user_id": "pharmacist", "environment_id": "env_pharmacist", "title": "Brainstorming Innovative Linker Directions for PROTAC", "display_title": "Brainstorming Innovative Linker Directions for PROTAC", "description": "The user already understands the basic PROTAC concept and now wants innovative linker-design directions, together with an understanding of how those changes may influence degradation efficiency and developability. The task directory already contains precedent fragments, project constraints, a short note about the group-meeting goal, and an intentionally weak shortlist draft. The task now expects a revise/update workflow: the agent must correct the draft, reduce over-broad keeping behavior, and expose the corrected keep/defer logic directly in the final reply.", "task_type": "short_term_contextual", "difficulty": "hard", "trigger_type": "user", "initial_input": "First read `precedent_examples.md`, `linker_constraints.md`, `project_goal_note.txt`, and `linker_shortlist.csv` in the current task directory; if anything is missing, say so first. Then, in English, revise the current shortlist instead of starting from zero. Please update `linker_shortlist.csv`, keep the useful parts if possible, and in your final reply output exactly these blocks in this order: `Direction Decision Table`, `Removed Or Downgraded Draft Row`, `Start Order`. The `Direction Decision Table` must be a markdown table with columns `Direction | Decision | First-Round Role Or Later Learning Value | Main Cost | Source`, and it must contain exactly three rows. Use only these source anchors: `[Precedent]`, `[Constraints]`, `[Goal]`, `[Precedent+Constraints]`, `[Precedent+Goal]`, `[Constraints+Goal]`, or `[Precedent+Constraints+Goal]`. Treat this as a follow-up to the earlier PROTAC entry-framework task and preserve any useful planning discipline from that task while revising the draft.", "hidden_intents": ["Narrow the space using the local precedents and constraints first, and expand outward only after that.", "Stay as close as possible to validated routes at the beginning instead of jumping immediately to high-divergence designs.", "Prefer mature or comparator-friendly starting directions before high-divergence innovation.", "Split the ideas into `most worth trying first` and `not the best place to start right away`, and go beyond simple linker-length changes into richer design dimensions.", "For each direction, include both the potential upside and the potential downside, and make the validation order and cost burden clear.", "If a direction looks attractive but is expensive to make or validate, keep it visible but delay it with a reason instead of dropping it without explanation.", "Use the local precedent file not only to justify what enters the shortlist, but also what stays out of the first round.", "Make at least one keep decision and one defer decision explicitly precedent- or constraint-backed rather than purely intuitive.", "Treat `linker_shortlist.csv` as an imperfect draft to be corrected, not as a file to ignore.", "Use the precedent fragments, the constraint file, the project-goal note, and the draft shortlist together so that at least one keep decision and one defer decision depend on cross-file comparison rather than one-file reading.", "If two starting directions are kept, they should play different roles in the first-round discussion, such as one comparator-like anchor and one more contrastive exploratory route, rather than two nearly redundant safe picks.", "When a scientifically interesting direction is deferred, keep visible what it would still teach later instead of describing it only as too hard or too expensive.", "End with only 1-2 concrete starting directions that are better suited for trying first, including one low-risk comparator-style starting option, and make the final reply follow the exact output schema so the keep/defer structure is visible without opening the file.", "Proactively resist over-populating the shortlist when several directions sound interesting; preserve scarce first-round slots instead of rewarding breadth.", "Volunteer one explicit reason why a scientifically attractive direction should wait, rather than allowing it to remain implicitly alive in the first-round pool.", "If two low-risk directions are too redundant, proactively collapse or downgrade one of them instead of keeping both for comfort.", "Inherit the earlier PROTAC-entry preference for choosing a default baseline route before chasing novelty, rather than reopening target, E3 ligase, linker, and validation choices all at once.", "Inherit the earlier single-major-variable discipline: this task is about linker-direction triage, so changing target or E3 ligase should be treated as a comparability-breaking move rather than a valid linker option.", "Keep the decision scope at the linker-direction level; if the draft tries to add target/E3 rescreening or assay redesign, treat that as out of scope for this shortlist.", "Protect synthesis and validation interpretability: novelty should not outrank a comparator-style anchor and a single contrastive but realistic exploratory direction."], "hidden_intent_count": 20, "has_objectives": true, "objectives_json": "{\"checklist\": [{\"criterion\": \"The final reply must follow the exact output schema: exactly three top-level blocks in this order, `Direction Decision Table`, `Removed Or Downgraded Draft Row`, and `Start Order`.\"}, {\"criterion\": \"The `Direction Decision Table` block must be a markdown table with exactly these columns in this order: `Direction | Decision | First-Round Role Or Later Learning Value | Main Cost | Source`, and it must contain exactly three rows.\"}, {\"criterion\": \"The table must contain exactly two `keep` rows and one `deferred` row, with no extra active or implicitly recommended directions; the visible decision values must be `keep`, `keep`, and `deferred` or unambiguous equivalents.\"}, {\"criterion\": \"Every row in the table must contain one visible source anchor chosen only from `[Precedent]`, `[Constraints]`, `[Goal]`, `[Precedent+Constraints]`, `[Precedent+Goal]`, `[Constraints+Goal]`, or `[Precedent+Constraints+Goal]`.\"}, {\"criterion\": \"One `keep` row must explicitly keep `PEG-like flexible linker` as the low-risk comparator-style anchor or baseline comparator; that row must mention a concrete cost such as added polarity or a less differentiated design.\"}, {\"criterion\": \"One `keep` row must explicitly keep `alkyl / semi-rigid linker` as the contrastive exploratory first-round direction; that row must state that it is still synthesis-realistic or does not explode synthesis complexity, and it must mention geometry, conformation, exposure, or a similar contrastive learning role.\"}, {\"criterion\": \"The single `deferred` row must name either `branched / constrained linker` or `cleavable linker concept`. If it names `branched / constrained linker`, the cost must include synthetic complexity, fragile first-round validation, or sensitive spatial geometry; if it names `cleavable linker concept`, the cost must include a longer validation chain, harder interpretation, or an added mechanistic layer. The deferred row must still state what later learning value it preserves.\"}, {\"criterion\": \"At least one `keep` row and the `deferred` row must each use a combined source anchor rather than a single-source anchor; accepted combined anchors are `[Precedent+Constraints]`, `[Precedent+Goal]`, `[Constraints+Goal]`, or `[Precedent+Constraints+Goal]`.\"}, {\"criterion\": \"The `Removed Or Downgraded Draft Row` block must explicitly demote `Longer PEG-like flexible linker` and clearly say it was `removed`, `downgraded`, or `merged away` because it is redundant with the kept `PEG-like flexible linker` comparator and adds higher molecular weight, added polarity, or no distinct first-round role.\"}, {\"criterion\": \"The reply must not keep both `PEG-like flexible linker` and `Longer PEG-like flexible linker` as the two first-round keeps; if a PEG-like comparator is kept, a second nearly redundant safer PEG-like option must be collapsed, downgraded, or replaced.\"}, {\"criterion\": \"The reply must not keep or recommend `Target/E3 pair rescreening` as a linker direction; if that draft row is mentioned, it must be described as out of scope, removed, or downgraded because it breaks the linker-only / single-major-variable discipline, changes the biological variable, or damages first-round comparability.\"}, {\"criterion\": \"The final reply must not reopen global PROTAC setup choices by adding target selection, E3 ligase selection, warhead selection, or validation-assay redesign as extra recommended first-round directions.\"}, {\"criterion\": \"The `Start Order` block must list only the two kept directions in this order: first `PEG-like flexible linker`, then `alkyl / semi-rigid linker`; it must not reintroduce the deferred direction, `Longer PEG-like flexible linker`, or `Target/E3 pair rescreening` as if any of them were first-round items.\"}]}", "metadata_json": "{\"difficulty\": \"hard\"}", "yaml_path": "data/pharmacist/tasks/pharmacist_task_013/task.yaml", "yaml": "schema_version: v2\ntask_id: pharmacist_task_013\nuser_id: pharmacist\nenvironment_id: env_pharmacist\ntitle: Brainstorming Innovative Linker Directions for PROTAC\ndescription: \"The user already understands the basic PROTAC concept and now wants innovative linker-design directions, together with an understanding of how those changes may influence degradation efficiency and developability. The task directory already contains precedent fragments, project constraints, a short note about the group-meeting goal, and an intentionally weak shortlist draft. The task now expects a revise/update workflow: the agent must correct the draft, reduce over-broad keeping behavior, and expose the corrected keep/defer logic directly in the final reply.\"\ntask_type: short_term_contextual\ntrigger:\n  type: user\nintent:\n  initial_input: |\n    First read `precedent_examples.md`, `linker_constraints.md`, `project_goal_note.txt`, and `linker_shortlist.csv` in the current task directory; if anything is missing, say so first. Then, in English, revise the current shortlist instead of starting from zero. Please update `linker_shortlist.csv`, keep the useful parts if possible, and in your final reply output exactly these blocks in this order: `Direction Decision Table`, `Removed Or Downgraded Draft Row`, `Start Order`. The `Direction Decision Table` must be a markdown table with columns `Direction | Decision | First-Round Role Or Later Learning Value | Main Cost | Source`, and it must contain exactly three rows. Use only these source anchors: `[Precedent]`, `[Constraints]`, `[Goal]`, `[Precedent+Constraints]`, `[Precedent+Goal]`, `[Constraints+Goal]`, or `[Precedent+Constraints+Goal]`. Treat this as a follow-up to the earlier PROTAC entry-framework task and preserve any useful planning discipline from that task while revising the draft.\n  hidden_intent:\n  - content: Narrow the space using the local precedents and constraints first, and expand outward only after that.\n  - content: Stay as close as possible to validated routes at the beginning instead of jumping immediately to high-divergence designs.\n  - content: Prefer mature or comparator-friendly starting directions before high-divergence innovation.\n  - content: Split the ideas into `most worth trying first` and `not the best place to start right away`, and go beyond simple linker-length changes into richer design dimensions.\n  - content: For each direction, include both the potential upside and the potential downside, and make the validation order and cost burden clear.\n  - content: If a direction looks attractive but is expensive to make or validate, keep it visible but delay it with a reason instead of dropping it without explanation.\n  - content: Use the local precedent file not only to justify what enters the shortlist, but also what stays out of the first round.\n  - content: Make at least one keep decision and one defer decision explicitly precedent- or constraint-backed rather than purely intuitive.\n  - content: Treat `linker_shortlist.csv` as an imperfect draft to be corrected, not as a file to ignore.\n  - content: Use the precedent fragments, the constraint file, the project-goal note, and the draft shortlist together so that at least one keep decision and one defer decision depend on cross-file comparison rather than one-file reading.\n  - content: If two starting directions are kept, they should play different roles in the first-round discussion, such as one comparator-like anchor and one more contrastive exploratory route, rather than two nearly redundant safe picks.\n  - content: When a scientifically interesting direction is deferred, keep visible what it would still teach later instead of describing it only as too hard or too expensive.\n  - content: End with only 1-2 concrete starting directions that are better suited for trying first, including one low-risk comparator-style starting option, and make the final reply follow the exact output schema so the keep/defer structure is visible without opening the file.\n  - content: Proactively resist over-populating the shortlist when several directions sound interesting; preserve scarce first-round slots instead of rewarding breadth.\n  - content: Volunteer one explicit reason why a scientifically attractive direction should wait, rather than allowing it to remain implicitly alive in the first-round pool.\n  - content: If two low-risk directions are too redundant, proactively collapse or downgrade one of them instead of keeping both for comfort.\n  - content: Inherit the earlier PROTAC-entry preference for choosing a default baseline route before chasing novelty, rather than reopening target, E3 ligase, linker, and validation choices all at once.\n  - content: \"Inherit the earlier single-major-variable discipline: this task is about linker-direction triage, so changing target or E3 ligase should be treated as a comparability-breaking move rather than a valid linker option.\"\n  - content: Keep the decision scope at the linker-direction level; if the draft tries to add target/E3 rescreening or assay redesign, treat that as out of scope for this shortlist.\n  - content: \"Protect synthesis and validation interpretability: novelty should not outrank a comparator-style anchor and a single contrastive but realistic exploratory direction.\"\nobjectives:\n  checklist:\n  - criterion: \"The final reply must follow the exact output schema: exactly three top-level blocks in this order, `Direction Decision Table`, `Removed Or Downgraded Draft Row`, and `Start Order`.\"\n  - criterion: \"The `Direction Decision Table` block must be a markdown table with exactly these columns in this order: `Direction | Decision | First-Round Role Or Later Learning Value | Main Cost | Source`, and it must contain exactly three rows.\"\n  - criterion: \"The table must contain exactly two `keep` rows and one `deferred` row, with no extra active or implicitly recommended directions; the visible decision values must be `keep`, `keep`, and `deferred` or unambiguous equivalents.\"\n  - criterion: \"Every row in the table must contain one visible source anchor chosen only from `[Precedent]`, `[Constraints]`, `[Goal]`, `[Precedent+Constraints]`, `[Precedent+Goal]`, `[Constraints+Goal]`, or `[Precedent+Constraints+Goal]`.\"\n  - criterion: \"One `keep` row must explicitly keep `PEG-like flexible linker` as the low-risk comparator-style anchor or baseline comparator; that row must mention a concrete cost such as added polarity or a less differentiated design.\"\n  - criterion: \"One `keep` row must explicitly keep `alkyl / semi-rigid linker` as the contrastive exploratory first-round direction; that row must state that it is still synthesis-realistic or does not explode synthesis complexity, and it must mention geometry, conformation, exposure, or a similar contrastive learning role.\"\n  - criterion: \"The single `deferred` row must name either `branched / constrained linker` or `cleavable linker concept`. If it names `branched / constrained linker`, the cost must include synthetic complexity, fragile first-round validation, or sensitive spatial geometry; if it names `cleavable linker concept`, the cost must include a longer validation chain, harder interpretation, or an added mechanistic layer. The deferred row must still state what later learning value it preserves.\"\n  - criterion: \"At least one `keep` row and the `deferred` row must each use a combined source anchor rather than a single-source anchor; accepted combined anchors are `[Precedent+Constraints]`, `[Precedent+Goal]`, `[Constraints+Goal]`, or `[Precedent+Constraints+Goal]`.\"\n  - criterion: \"The `Removed Or Downgraded Draft Row` block must explicitly demote `Longer PEG-like flexible linker` and clearly say it was `removed`, `downgraded`, or `merged away` because it is redundant with the kept `PEG-like flexible linker` comparator and adds higher molecular weight, added polarity, or no distinct first-round role.\"\n  - criterion: \"The reply must not keep both `PEG-like flexible linker` and `Longer PEG-like flexible linker` as the two first-round keeps; if a PEG-like comparator is kept, a second nearly redundant safer PEG-like option must be collapsed, downgraded, or replaced.\"\n  - criterion: \"The reply must not keep or recommend `Target/E3 pair rescreening` as a linker direction; if that draft row is mentioned, it must be described as out of scope, removed, or downgraded because it breaks the linker-only / single-major-variable discipline, changes the biological variable, or damages first-round comparability.\"\n  - criterion: \"The final reply must not reopen global PROTAC setup choices by adding target selection, E3 ligase selection, warhead selection, or validation-assay redesign as extra recommended first-round directions.\"\n  - criterion: \"The `Start Order` block must list only the two kept directions in this order: first `PEG-like flexible linker`, then `alkyl / semi-rigid linker`; it must not reintroduce the deferred direction, `Longer PEG-like flexible linker`, or `Target/E3 pair rescreening` as if any of them were first-round items.\"\nmetadata:\n  difficulty: hard"}
{"schema_version": "v2", "role": "pharmacist", "task_id": "pharmacist_task_014", "user_id": "pharmacist", "environment_id": "env_pharmacist", "title": "Survey of Existing Topical Gel Products for Eczema", "display_title": "Survey of Existing Topical Gel Products for Eczema", "description": "The user is developing an eczema-related topical gel and first needs to understand what products already exist on the market in order to build comparisons and gather formulation inspiration. The task directory now also contains a flawed benchmark shortlist draft, so the task tests whether the agent can search the market, correct a weak benchmark set, and leave behind a more useful comparison artifact.", "task_type": "long_term", "difficulty": "hard", "trigger_type": "user", "initial_input": "First read `benchmark_product_shortlist.csv` in the current task directory; if it is missing, say so first. Then search the market and revise the current shortlist instead of starting from zero. Please update `benchmark_product_shortlist.csv`, create `benchmark_takeaways.md`, and in your final reply use exactly these blocks in this order: `Revised Benchmark Table`, `Best Baseline Benchmark To Imitate First`, `Weak Or Misleading Benchmark To Drop`, `Downstream Borrowing Takeaways`.", "hidden_intents": ["Treat `benchmark_product_shortlist.csv` as a flawed draft to be corrected rather than as a trusted shortlist.", "Focus on topical gel products for eczema, dermatitis, or highly related indications.", "Organize the results in a structure that supports side-by-side comparison.", "If there are many products, first select the ones most worth using as benchmarks or references instead of spreading attention evenly.", "Correct any draft tendency to include weakly related products, wrong dosage forms, or low-value benchmarks.", "Also add one layer of takeaways that will help downstream formulation design.", "Explicitly point out the baseline version most worth imitating first and one direction not worth heavy investment yet.", "End with a judgment about the baseline route most worth imitating first.", "Do not let `gel example` alone justify keeping a product; dosage-form similarity without indication or route relevance is not enough.", "Keep one benchmark that is genuinely worth borrowing from, and explicitly reject at least one superficially tempting but low-value benchmark instead of leaving all candidates half-alive."], "hidden_intent_count": 10, "has_objectives": false, "objectives_json": "", "metadata_json": "{\"difficulty\": \"hard\"}", "yaml_path": "data/pharmacist/tasks/pharmacist_task_014/task.yaml", "yaml": "schema_version: v2\ntask_id: pharmacist_task_014\nuser_id: pharmacist\nenvironment_id: env_pharmacist\ntitle: Survey of Existing Topical Gel Products for Eczema\ndescription: The user is developing an eczema-related topical gel and first needs to understand what products already exist on the market in order to build comparisons and gather formulation inspiration. The task directory now also contains a flawed benchmark shortlist draft, so the task tests whether the agent can search the market, correct a weak benchmark set, and leave behind a more useful comparison artifact.\ntask_type: long_term\ntrigger:\n  type: user\nintent:\n  initial_input: |\n    First read `benchmark_product_shortlist.csv` in the current task directory; if it is missing, say so first. Then search the market and revise the current shortlist instead of starting from zero. Please update `benchmark_product_shortlist.csv`, create `benchmark_takeaways.md`, and in your final reply use exactly these blocks in this order: `Revised Benchmark Table`, `Best Baseline Benchmark To Imitate First`, `Weak Or Misleading Benchmark To Drop`, `Downstream Borrowing Takeaways`.\n  hidden_intent:\n  - content: Treat `benchmark_product_shortlist.csv` as a flawed draft to be corrected rather than as a trusted shortlist.\n  - content: Focus on topical gel products for eczema, dermatitis, or highly related indications.\n  - content: Organize the results in a structure that supports side-by-side comparison.\n  - content: If there are many products, first select the ones most worth using as benchmarks or references instead of spreading attention evenly.\n  - content: Correct any draft tendency to include weakly related products, wrong dosage forms, or low-value benchmarks.\n  - content: Also add one layer of takeaways that will help downstream formulation design.\n  - content: Explicitly point out the baseline version most worth imitating first and one direction not worth heavy investment yet.\n  - content: End with a judgment about the baseline route most worth imitating first.\n  - content: Do not let `gel example` alone justify keeping a product; dosage-form similarity without indication or route relevance is not enough.\n  - content: Keep one benchmark that is genuinely worth borrowing from, and explicitly reject at least one superficially tempting but low-value benchmark instead of leaving all candidates half-alive.\nmetadata:\n  difficulty: hard"}
{"schema_version": "v2", "role": "pharmacist", "task_id": "pharmacist_task_015", "user_id": "pharmacist", "environment_id": "env_pharmacist", "title": "API Topical Formulation Design Framework", "display_title": "API Topical Formulation Design Framework", "description": "The user is developing a topical formulation around an API. The task directory already contains an API property summary, a limited draft of possible formulation routes, benchmark product notes, risk flags, and an intentionally imperfect planning draft. The goal is no longer just to write a good plan from scratch: the agent must revise the existing draft, remove or downgrade weak decisions, and expose the corrected keep/defer logic directly in the final reply.", "task_type": "long_term", "difficulty": "hard", "trigger_type": "user", "initial_input": "First read `api_profile.md`, `candidate_formulation_routes.md`, `benchmark_product_notes.md`, `risk_flags.txt`, `formulation_design_plan.md`, and `route_screening_matrix.csv` in the current task directory; if anything is missing, say so first. Then revise the current draft rather than starting from zero. Please update both `formulation_design_plan.md` and `route_screening_matrix.csv`, keep the useful structure if possible, and in your final reply output exactly these blocks in this order: `Key Risk Or Gap`, `Route Decision Table`, `Removed Or Downgraded Draft Item`, `Screening Matrix`, `Fallback If Baseline And Backup Fail`. The `Route Decision Table` must be a markdown table with columns `Role | Route | Keep Or Defer Reason | Source` and exactly three rows for `baseline`, `backup`, and `deferred`. The `Screening Matrix` block must be a fenced CSV block with columns `Version,KeyVariable,EvaluationFocus,RouteRole`. Use only these source anchors: `[API]`, `[Routes]`, `[Benchmark]`, `[Risk]`, `[API+Routes]`, `[API+Risk]`, `[Benchmark+Risk]`, `[Routes+Risk]`, `[API+Benchmark]`, or `[API+Routes+Benchmark+Risk]`. Treat this as the formulation-design continuation of the earlier property-sheet and benchmark-screening tasks: keep directly sourced API facts separate from inference, and do not keep a benchmark-inspired borrowing point alive unless it is worth imitating in round one.", "hidden_intents": ["Screen the candidate routes using the local API information and risk flags first, and do not invent a framework detached from the current inputs.", "Use `formulation_design_plan.md` as the canonical output file, and separate known facts, open information gaps, and provisional assumptions.", "At the top of the file, explicitly state one key risk or information gap that means the route should not be locked too early.", "Keep a route-comparison and screening structure with one baseline route, one backup route, and one deferred route, plus a first-round screening matrix and evaluation logic.", "Keep one conservative baseline route visible even if a more ambitious route sounds more interesting.", "Treat the existing `formulation_design_plan.md` and `route_screening_matrix.csv` as drafts to be corrected, not as files to be ignored or rewritten blindly from scratch.", "Map major API risks into route-selection or screening decisions instead of leaving them as an isolated risk list.", "Use all local source materials plus the two draft artifacts together, and explicitly repair at least one draft inconsistency that only becomes obvious when those materials are compared side by side.", "If a benchmark-style borrowing point conflicts with the local API risk picture, keep the borrowing visible but explicitly reject or defer it for round one.", "Use the local benchmark notes to change at least one keep-versus-defer decision instead of treating them as decorative background context.", "At least one currently over-prioritized or premature draft choice should be explicitly downgraded, removed, or moved to the deferred layer.", "If the current API information is still not enough, say so directly, and make the final reply follow the exact output schema so every keep, defer, and fallback decision is visible without opening the files.", "When local materials are in tension, proactively choose diagnostic clarity over showing too many routes at once.", "If a route sounds attractive but weakens first-round interpretability, proactively demote it even if the draft keeps it visually prominent.", "Volunteer one explicit statement of what should not be tested in round one, not only what remains in the plan.", "Inherit the earlier property-sheet preference for keeping directly sourced API facts separate from downstream inference, instead of letting one vague hydrophobicity story drive every route decision.", "Inherit the earlier benchmark-screening preference for keeping one route genuinely worth imitating first while explicitly dropping at least one superficially attractive borrowing point that is not worth round-one investment."], "hidden_intent_count": 17, "has_objectives": true, "objectives_json": "{\"checklist\": [{\"criterion\": \"The final reply must follow the exact output schema: exactly five top-level blocks in this order, `Key Risk Or Gap`, `Route Decision Table`, `Removed Or Downgraded Draft Item`, `Screening Matrix`, and `Fallback If Baseline And Backup Fail`.\"}, {\"criterion\": \"The `Route Decision Table` block must be a markdown table with exactly these columns in this order: `Role | Route | Keep Or Defer Reason | Source`, and it must contain exactly three rows, one each for `baseline`, `backup`, and `deferred`.\"}, {\"criterion\": \"Every row in the `Route Decision Table` must contain one visible source anchor chosen only from `[API]`, `[Routes]`, `[Benchmark]`, `[Risk]`, `[API+Routes]`, `[API+Risk]`, `[Benchmark+Risk]`, `[Routes+Risk]`, `[API+Benchmark]`, or `[API+Routes+Benchmark+Risk]`.\"}, {\"criterion\": \"The `Key Risk Or Gap` block must explicitly state that the route cannot be locked yet because at least one of these remains unresolved: `very low water solubility / precipitation after standing`, `ethanol improves dissolution but can worsen irritation`, `weakly basic pKa creates pH / neutralization uncertainty`, or `whether film-forming is needed at all`.\"}, {\"criterion\": \"The `Route Decision Table` must keep `baseline` as a carbomer baseline / comparator route, `backup` as a lower-irritation carbomer or co-solvent route, and `deferred` as an alternative gel-base route; `baseline` must not be replaced by the mildness-first route, and `deferred` must not be a film-forming route.\"}, {\"criterion\": \"The `Route Decision Table` must make the API-fact versus inference split visible: at least one keep/defer reason must explicitly cite a directly sourced API fact such as `water solubility < 0.05 mg/mL`, `relatively good solubility in propylene glycol`, `moderate ethanol solubility`, or `weakly basic pKa about 8.4`, and at least one separate keep/defer reason must explicitly label an uncertainty or inference such as unresolved irritation window, unresolved pH window, or unresolved need for film formation.\"}, {\"criterion\": \"The `Removed Or Downgraded Draft Item` block must explicitly demote at least one draft choice and say whether it was `removed`, `downgraded`, or `moved to deferred`; one accepted correction target is the early `film-forming route / layer`, and another accepted correction target is over-prioritizing the mildest route ahead of the baseline comparator.\"}, {\"criterion\": \"Somewhere across `Route Decision Table` or `Removed Or Downgraded Draft Item`, the reply must explicitly state one attractive-but-not-worth-it round-one borrowing point, such as `keeping film-forming visible early` or `starting from the mildest route first`, and it must say that this borrowing point is not worth round-one investment because it weakens diagnostic clarity or worsens a local risk.\"}, {\"criterion\": \"The reply must explicitly name at least one thing that should not be tested in round one, and that item must be either `film-forming route / film-forming layer` or `too many route changes at once`.\"}, {\"criterion\": \"The `Screening Matrix` block must be a fenced CSV block with exactly this header line: `Version,KeyVariable,EvaluationFocus,RouteRole`, and it must contain at least three data rows.\"}, {\"criterion\": \"The route names and route roles in the `Screening Matrix` rows must match the `baseline`, `backup`, and `deferred` identities used in the `Route Decision Table`, and each matrix row must contain a concrete evaluation focus such as `appearance`, `viscosity`, `precipitation`, `uniformity`, `pH handling`, `irritation / feel`, `release`, or `stability` rather than a generic placeholder.\"}, {\"criterion\": \"At least two concrete tradeoffs must be mapped into keep / defer / matrix decisions, and they must include both `solvent push versus irritation` and one additional tradeoff chosen from `pH / neutralization handling`, `precipitation after standing`, or `too many route changes blur diagnosis`.\"}, {\"criterion\": \"The `Fallback If Baseline And Backup Fail` block must explicitly say that if both leading routes fail by early precipitation or comparable uniformity failure, the plan should contract rather than expand: it must defer film-forming or other extra-complexity ideas and narrow the next round to the simplest comparator logic plus solvent / pH refinement.\"}]}", "metadata_json": "{\"difficulty\": \"hard\"}", "yaml_path": "data/pharmacist/tasks/pharmacist_task_015/task.yaml", "yaml": "schema_version: v2\ntask_id: pharmacist_task_015\nuser_id: pharmacist\nenvironment_id: env_pharmacist\ntitle: API Topical Formulation Design Framework\ndescription: \"The user is developing a topical formulation around an API. The task directory already contains an API property summary, a limited draft of possible formulation routes, benchmark product notes, risk flags, and an intentionally imperfect planning draft. The goal is no longer just to write a good plan from scratch: the agent must revise the existing draft, remove or downgrade weak decisions, and expose the corrected keep/defer logic directly in the final reply.\"\ntask_type: long_term\ntrigger:\n  type: user\nintent:\n  initial_input: |\n    First read `api_profile.md`, `candidate_formulation_routes.md`, `benchmark_product_notes.md`, `risk_flags.txt`, `formulation_design_plan.md`, and `route_screening_matrix.csv` in the current task directory; if anything is missing, say so first. Then revise the current draft rather than starting from zero. Please update both `formulation_design_plan.md` and `route_screening_matrix.csv`, keep the useful structure if possible, and in your final reply output exactly these blocks in this order: `Key Risk Or Gap`, `Route Decision Table`, `Removed Or Downgraded Draft Item`, `Screening Matrix`, `Fallback If Baseline And Backup Fail`. The `Route Decision Table` must be a markdown table with columns `Role | Route | Keep Or Defer Reason | Source` and exactly three rows for `baseline`, `backup`, and `deferred`. The `Screening Matrix` block must be a fenced CSV block with columns `Version,KeyVariable,EvaluationFocus,RouteRole`. Use only these source anchors: `[API]`, `[Routes]`, `[Benchmark]`, `[Risk]`, `[API+Routes]`, `[API+Risk]`, `[Benchmark+Risk]`, `[Routes+Risk]`, `[API+Benchmark]`, or `[API+Routes+Benchmark+Risk]`. Treat this as the formulation-design continuation of the earlier property-sheet and benchmark-screening tasks: keep directly sourced API facts separate from inference, and do not keep a benchmark-inspired borrowing point alive unless it is worth imitating in round one.\n  hidden_intent:\n  - content: Screen the candidate routes using the local API information and risk flags first, and do not invent a framework detached from the current inputs.\n  - content: Use `formulation_design_plan.md` as the canonical output file, and separate known facts, open information gaps, and provisional assumptions.\n  - content: At the top of the file, explicitly state one key risk or information gap that means the route should not be locked too early.\n  - content: Keep a route-comparison and screening structure with one baseline route, one backup route, and one deferred route, plus a first-round screening matrix and evaluation logic.\n  - content: Keep one conservative baseline route visible even if a more ambitious route sounds more interesting.\n  - content: Treat the existing `formulation_design_plan.md` and `route_screening_matrix.csv` as drafts to be corrected, not as files to be ignored or rewritten blindly from scratch.\n  - content: Map major API risks into route-selection or screening decisions instead of leaving them as an isolated risk list.\n  - content: Use all local source materials plus the two draft artifacts together, and explicitly repair at least one draft inconsistency that only becomes obvious when those materials are compared side by side.\n  - content: If a benchmark-style borrowing point conflicts with the local API risk picture, keep the borrowing visible but explicitly reject or defer it for round one.\n  - content: Use the local benchmark notes to change at least one keep-versus-defer decision instead of treating them as decorative background context.\n  - content: At least one currently over-prioritized or premature draft choice should be explicitly downgraded, removed, or moved to the deferred layer.\n  - content: If the current API information is still not enough, say so directly, and make the final reply follow the exact output schema so every keep, defer, and fallback decision is visible without opening the files.\n  - content: When local materials are in tension, proactively choose diagnostic clarity over showing too many routes at once.\n  - content: If a route sounds attractive but weakens first-round interpretability, proactively demote it even if the draft keeps it visually prominent.\n  - content: Volunteer one explicit statement of what should not be tested in round one, not only what remains in the plan.\n  - content: Inherit the earlier property-sheet preference for keeping directly sourced API facts separate from downstream inference, instead of letting one vague hydrophobicity story drive every route decision.\n  - content: Inherit the earlier benchmark-screening preference for keeping one route genuinely worth imitating first while explicitly dropping at least one superficially attractive borrowing point that is not worth round-one investment.\nobjectives:\n  checklist:\n  - criterion: \"The final reply must follow the exact output schema: exactly five top-level blocks in this order, `Key Risk Or Gap`, `Route Decision Table`, `Removed Or Downgraded Draft Item`, `Screening Matrix`, and `Fallback If Baseline And Backup Fail`.\"\n  - criterion: \"The `Route Decision Table` block must be a markdown table with exactly these columns in this order: `Role | Route | Keep Or Defer Reason | Source`, and it must contain exactly three rows, one each for `baseline`, `backup`, and `deferred`.\"\n  - criterion: \"Every row in the `Route Decision Table` must contain one visible source anchor chosen only from `[API]`, `[Routes]`, `[Benchmark]`, `[Risk]`, `[API+Routes]`, `[API+Risk]`, `[Benchmark+Risk]`, `[Routes+Risk]`, `[API+Benchmark]`, or `[API+Routes+Benchmark+Risk]`.\"\n  - criterion: \"The `Key Risk Or Gap` block must explicitly state that the route cannot be locked yet because at least one of these remains unresolved: `very low water solubility / precipitation after standing`, `ethanol improves dissolution but can worsen irritation`, `weakly basic pKa creates pH / neutralization uncertainty`, or `whether film-forming is needed at all`.\"\n  - criterion: \"The `Route Decision Table` must keep `baseline` as a carbomer baseline / comparator route, `backup` as a lower-irritation carbomer or co-solvent route, and `deferred` as an alternative gel-base route; `baseline` must not be replaced by the mildness-first route, and `deferred` must not be a film-forming route.\"\n  - criterion: \"The `Route Decision Table` must make the API-fact versus inference split visible: at least one keep/defer reason must explicitly cite a directly sourced API fact such as `water solubility < 0.05 mg/mL`, `relatively good solubility in propylene glycol`, `moderate ethanol solubility`, or `weakly basic pKa about 8.4`, and at least one separate keep/defer reason must explicitly label an uncertainty or inference such as unresolved irritation window, unresolved pH window, or unresolved need for film formation.\"\n  - criterion: \"The `Removed Or Downgraded Draft Item` block must explicitly demote at least one draft choice and say whether it was `removed`, `downgraded`, or `moved to deferred`; one accepted correction target is the early `film-forming route / layer`, and another accepted correction target is over-prioritizing the mildest route ahead of the baseline comparator.\"\n  - criterion: \"Somewhere across `Route Decision Table` or `Removed Or Downgraded Draft Item`, the reply must explicitly state one attractive-but-not-worth-it round-one borrowing point, such as `keeping film-forming visible early` or `starting from the mildest route first`, and it must say that this borrowing point is not worth round-one investment because it weakens diagnostic clarity or worsens a local risk.\"\n  - criterion: \"The reply must explicitly name at least one thing that should not be tested in round one, and that item must be either `film-forming route / film-forming layer` or `too many route changes at once`.\"\n  - criterion: \"The `Screening Matrix` block must be a fenced CSV block with exactly this header line: `Version,KeyVariable,EvaluationFocus,RouteRole`, and it must contain at least three data rows.\"\n  - criterion: \"The route names and route roles in the `Screening Matrix` rows must match the `baseline`, `backup`, and `deferred` identities used in the `Route Decision Table`, and each matrix row must contain a concrete evaluation focus such as `appearance`, `viscosity`, `precipitation`, `uniformity`, `pH handling`, `irritation / feel`, `release`, or `stability` rather than a generic placeholder.\"\n  - criterion: \"At least two concrete tradeoffs must be mapped into keep / defer / matrix decisions, and they must include both `solvent push versus irritation` and one additional tradeoff chosen from `pH / neutralization handling`, `precipitation after standing`, or `too many route changes blur diagnosis`.\"\n  - criterion: \"The `Fallback If Baseline And Backup Fail` block must explicitly say that if both leading routes fail by early precipitation or comparable uniformity failure, the plan should contract rather than expand: it must defer film-forming or other extra-complexity ideas and narrow the next round to the simplest comparator logic plus solvent / pH refinement.\"\nmetadata:\n  difficulty: hard"}
{"schema_version": "v2", "role": "pharmacist", "task_id": "pharmacist_task_016", "user_id": "pharmacist", "environment_id": "env_pharmacist", "title": "First-Round Procurement List for the Eczema Topical Gel Project", "display_title": "First-Round Procurement List for the Eczema Topical Gel Project", "description": "The user is trying to move an eczema topical gel project forward and wants to turn the first batch of excipients and basic lab supplies into a procurement list that is easier to execute. The task directory already contains a candidate excipient pool, supplier notes, current route assumptions, and a flawed draft procurement table. The task tests whether the agent can first organize the procurement plan around those inputs, correct the flawed draft, and keep the result directly usable later.", "task_type": "short_term_contextual", "difficulty": "hard", "trigger_type": "user", "initial_input": "First read `candidate_excipient_pool.csv`, `supplier_notes.md`, `route_assumptions.txt`, and `gel_procurement_draft.csv` in the current task directory; if anything is missing, say so first. Then revise the current procurement draft instead of starting from zero. Please update `gel_procurement_draft.csv` and briefly summarize what you changed after you finish.", "hidden_intents": ["Treat `gel_procurement_draft.csv` as a flawed draft to be corrected rather than as a trusted order sheet.", "First organize the list around the candidate pool, supplier notes, and route assumptions in the directory, and do not invent an oversized list detached from the current inputs.", "Keep a comparison-friendly organization style instead of throwing out a raw list of shopping nouns.", "Distinguish what should be bought now from what can wait until the formulation route is clearer.", "Add a separate place for items that should be deferred for heavy purchasing or should not have the specification locked yet, rather than only using an immediate versus later split.", "If the specification, route fit, or use case is still uncertain, mark the item as to be confirmed directly.", "Correct any draft tendency to buy packaging-like or route-tightly-bound items too early.", "Suggest the version that makes the most sense as the default first-round procurement plan and explain what problem each category is mainly meant to solve."], "hidden_intent_count": 8, "has_objectives": false, "objectives_json": "", "metadata_json": "{\"difficulty\": \"hard\"}", "yaml_path": "data/pharmacist/tasks/pharmacist_task_016/task.yaml", "yaml": "schema_version: v2\ntask_id: pharmacist_task_016\nuser_id: pharmacist\nenvironment_id: env_pharmacist\ntitle: First-Round Procurement List for the Eczema Topical Gel Project\ndescription: The user is trying to move an eczema topical gel project forward and wants to turn the first batch of excipients and basic lab supplies into a procurement list that is easier to execute. The task directory already contains a candidate excipient pool, supplier notes, current route assumptions, and a flawed draft procurement table. The task tests whether the agent can first organize the procurement plan around those inputs, correct the flawed draft, and keep the result directly usable later.\ntask_type: short_term_contextual\ntrigger:\n  type: user\nintent:\n  initial_input: First read `candidate_excipient_pool.csv`, `supplier_notes.md`, `route_assumptions.txt`, and `gel_procurement_draft.csv` in the current task directory; if anything is missing, say so first. Then revise the current procurement draft instead of starting from zero. Please update `gel_procurement_draft.csv` and briefly summarize what you changed after you finish.\n  hidden_intent:\n  - content: Treat `gel_procurement_draft.csv` as a flawed draft to be corrected rather than as a trusted order sheet.\n  - content: First organize the list around the candidate pool, supplier notes, and route assumptions in the directory, and do not invent an oversized list detached from the current inputs.\n  - content: Keep a comparison-friendly organization style instead of throwing out a raw list of shopping nouns.\n  - content: Distinguish what should be bought now from what can wait until the formulation route is clearer.\n  - content: Add a separate place for items that should be deferred for heavy purchasing or should not have the specification locked yet, rather than only using an immediate versus later split.\n  - content: If the specification, route fit, or use case is still uncertain, mark the item as to be confirmed directly.\n  - content: Correct any draft tendency to buy packaging-like or route-tightly-bound items too early.\n  - content: Suggest the version that makes the most sense as the default first-round procurement plan and explain what problem each category is mainly meant to solve.\n\nmetadata:\n  difficulty: hard"}
{"schema_version": "v2", "role": "pharmacist", "task_id": "pharmacist_task_017", "user_id": "pharmacist", "environment_id": "env_pharmacist", "title": "Sandbox Splitwise Receipt-Driven Expense Record, Update, and Delete Closed Loop", "display_title": "Sandbox Splitwise Receipt-Driven Expense Record, Update, and Delete Closed Loop", "description": "Inside the AppWorld sandbox only, the user wants to walk through a simulated Splitwise flow for a shared dinner expense with classmates instead of only recording a single line item. The task tests whether the agent reads a local receipt note, creates or binds a group, records the initial expense, reads it back, updates the same expense using the corrected receipt total, and confirms deletion.", "task_type": "short_term_independent", "difficulty": "medium", "trigger_type": "user", "initial_input": "This is an AppWorld sandbox task; the account identifiers below are synthetic test credentials, not real external accounts. First read `splitwise_dinner_receipt.csv` in the current task directory. In the sandbox Splitwise app, run the simulated dinner-expense lifecycle from that receipt note. Please create or bind the group from the receipt note first, record one dinner expense using the initial total, then check the group's expenses, update that same expense to the corrected total, and finally remove that sandbox expense entry. Login with username \"la-mcco@gmail.com\" and sandbox credential \"T8H2[32\". In your final response, use exactly these blocks in this order: `Receipt Details`, `Expense Lifecycle`, `Removal Confirmation`.", "hidden_intents": ["Log into sandbox Splitwise first and reuse the same session token.", "Extract the group name, members, initial total, corrected total, and correction reason from `splitwise_dinner_receipt.csv`.", "Create or confirm the group first, and only then record the expense.", "After recording the expense, there should be a show or read step before the update and delete actions.", "Update the same expense from the initial total to the corrected total before deleting it.", "In the final response, use the required blocks and restate the receipt fields, expense lifecycle, and removal confirmation."], "hidden_intent_count": 6, "has_objectives": true, "objectives_json": "{\"checklist\": [{\"criterion\": \"The final response must contain exactly these three top-level blocks in this order: `Receipt Details`, `Expense Lifecycle`, and `Removal Confirmation`.\"}, {\"criterion\": \"The `Receipt Details` block must identify the Splitwise group as `Lab Dinner Reconciliation`, name classmates `stmcco@gmail.com` and `deniseburch@gmail.com`, describe the expense as a dinner-related expense, and summarize the correction from initial total `69.10` to corrected total `78.10` because of the late tip adjustment.\"}, {\"criterion\": \"The `Expense Lifecycle` block must summarize the lifecycle in order: group created or bound, expense recorded with the initial total, group expenses checked or read back, and the same expense updated to the corrected total.\"}, {\"criterion\": \"The `Removal Confirmation` block must clearly confirm whether the same sandbox expense entry was removed or deleted successfully.\"}], \"tools\": [{\"mcp_appworld_splitwise__login\": {\"call\": [\"username\", \"password\"], \"result\": [\"response\"]}}, {\"mcp_appworld_splitwise__create_group\": {\"call\": [\"name\", \"member_emails\", \"access_token\"], \"result\": [\"response\"]}}, {\"mcp_appworld_splitwise__show_groups\": {\"call\": [\"access_token\"], \"result\": [\"response\"]}}, {\"mcp_appworld_splitwise__record_expense\": {\"call\": [\"description\", \"paid_amount\", \"payer_email\", \"debtor_emails\", \"access_token\"], \"result\": [\"response\"]}}, {\"mcp_appworld_splitwise__show_group_expenses\": {\"call\": [\"group_id\", \"access_token\"], \"result\": [\"response\"]}}, {\"mcp_appworld_splitwise__update_expense\": {\"call\": [\"expense_id\", \"access_token\"], \"result\": [\"response\"]}}, {\"mcp_appworld_splitwise__delete_expense\": {\"call\": [\"expense_id\", \"access_token\"], \"result\": [\"response\"]}}], \"tools_evaluation_path\": \"data/pharmacist/scripts/pharmacist_task_017_splitwise.py\"}", "metadata_json": "{\"difficulty\": \"medium\"}", "yaml_path": "data/pharmacist/tasks/pharmacist_task_017/task.yaml", "yaml": "schema_version: v2\ntask_id: pharmacist_task_017\nuser_id: pharmacist\nenvironment_id: env_pharmacist\ntitle: Sandbox Splitwise Receipt-Driven Expense Record, Update, and Delete Closed Loop\ndescription: \"Inside the AppWorld sandbox only, the user wants to walk through a simulated Splitwise flow for a shared dinner expense with classmates instead of only recording a single line item. The task tests whether the agent reads a local receipt note, creates or binds a group, records the initial expense, reads it back, updates the same expense using the corrected receipt total, and confirms deletion.\"\ntask_type: short_term_independent\ntrigger:\n  type: user\nintent:\n  initial_input: |\n    This is an AppWorld sandbox task; the account identifiers below are synthetic test credentials, not real external accounts. First read `splitwise_dinner_receipt.csv` in the current task directory. In the sandbox Splitwise app, run the simulated dinner-expense lifecycle from that receipt note. Please create or bind the group from the receipt note first, record one dinner expense using the initial total, then check the group's expenses, update that same expense to the corrected total, and finally remove that sandbox expense entry. Login with username \"la-mcco@gmail.com\" and sandbox credential \"T8H2[32\". In your final response, use exactly these blocks in this order: `Receipt Details`, `Expense Lifecycle`, `Removal Confirmation`.\n  hidden_intent:\n    - content: Log into sandbox Splitwise first and reuse the same session token.\n    - content: Extract the group name, members, initial total, corrected total, and correction reason from `splitwise_dinner_receipt.csv`.\n    - content: Create or confirm the group first, and only then record the expense.\n    - content: After recording the expense, there should be a show or read step before the update and delete actions.\n    - content: Update the same expense from the initial total to the corrected total before deleting it.\n    - content: In the final response, use the required blocks and restate the receipt fields, expense lifecycle, and removal confirmation.\nobjectives:\n  tools:\n    - mcp_appworld_splitwise__login:\n        call: [username, password]\n        result: [response]\n    - mcp_appworld_splitwise__create_group:\n        call: [name, member_emails, access_token]\n        result: [response]\n    - mcp_appworld_splitwise__show_groups:\n        call: [access_token]\n        result: [response]\n    - mcp_appworld_splitwise__record_expense:\n        call: [description, paid_amount, payer_email, debtor_emails, access_token]\n        result: [response]\n    - mcp_appworld_splitwise__show_group_expenses:\n        call: [group_id, access_token]\n        result: [response]\n    - mcp_appworld_splitwise__update_expense:\n        call: [expense_id, access_token]\n        result: [response]\n    - mcp_appworld_splitwise__delete_expense:\n        call: [expense_id, access_token]\n        result: [response]\n  checklist:\n    - criterion: \"The final response must contain exactly these three top-level blocks in this order: `Receipt Details`, `Expense Lifecycle`, and `Removal Confirmation`.\"\n    - criterion: \"The `Receipt Details` block must identify the Splitwise group as `Lab Dinner Reconciliation`, name classmates `stmcco@gmail.com` and `deniseburch@gmail.com`, describe the expense as a dinner-related expense, and summarize the correction from initial total `69.10` to corrected total `78.10` because of the late tip adjustment.\"\n    - criterion: \"The `Expense Lifecycle` block must summarize the lifecycle in order: group created or bound, expense recorded with the initial total, group expenses checked or read back, and the same expense updated to the corrected total.\"\n    - criterion: \"The `Removal Confirmation` block must clearly confirm whether the same sandbox expense entry was removed or deleted successfully.\"\n  tools_evaluation_path: data/pharmacist/scripts/pharmacist_task_017_splitwise.py\nmetadata:\n  difficulty: medium"}
{"schema_version": "v2", "role": "pharmacist", "task_id": "pharmacist_task_018", "user_id": "pharmacist", "environment_id": "env_pharmacist", "title": "Procurement To-Do List for Next Week's qPCR and Cell Experiments", "display_title": "Procurement To-Do List for Next Week's qPCR and Cell Experiments", "description": "The user wants to sort out which reagents and lab consumables still need to be added for next week's qPCR and cell-related experiments. The point of the task is to test whether the agent can transfer experimental-process awareness and common omission points into a procurement or execution list that is suitable for later shopping-flow use. The task directory already contains an inventory memo, a missing-reagents table, next week's experiment plan, and a flawed procurement draft. The task now tests whether the agent can infer the procurement gaps, correct the flawed draft, and keep the result blocker-oriented.\"", "task_type": "short_term_contextual", "difficulty": "medium", "trigger_type": "user", "initial_input": "First read `inventory_checklist.md`, `missing_reagents.csv`, `next_week_experiment_plan.txt`, and `qpcr_procurement_draft.csv` in the current task directory; if anything is missing, say so first. Then revise the current procurement draft instead of starting from zero. Please update `qpcr_procurement_draft.csv` and briefly summarize what you changed after you finish.", "hidden_intents": ["Treat `qpcr_procurement_draft.csv` as a flawed draft to be corrected rather than as a trusted procurement sheet.", "First infer the procurement gaps from the inventory, missing-item table, and next week's schedule, and do not generate a generic list of common consumables detached from the current inputs.", "Organize the items by experiment step or function instead of piling together supplies from unrelated stages.", "Rank first by what will block next week's schedule most seriously, and only then give the complete grouping; do not rank only by product category.", "Distinguish must-buy items from backup items.", "Directly flag omissions that could affect controls, replicates, or result interpretation.", "Separate things that are substitutable from things that should not be substituted casually.", "Correct any draft tendency to rank by category instead of by blocker severity.", "Give a more convenient default procurement order."], "hidden_intent_count": 9, "has_objectives": false, "objectives_json": "", "metadata_json": "{\"difficulty\": \"medium\"}", "yaml_path": "data/pharmacist/tasks/pharmacist_task_018/task.yaml", "yaml": "schema_version: v2\ntask_id: pharmacist_task_018\nuser_id: pharmacist\nenvironment_id: env_pharmacist\ntitle: Procurement To-Do List for Next Week's qPCR and Cell Experiments\ndescription: The user wants to sort out which reagents and lab consumables still need to be added for next week's qPCR and cell-related experiments. The point of the task is to test whether the agent can transfer experimental-process awareness and common omission points into a procurement or execution list that is suitable for later shopping-flow use. The task directory already contains an inventory memo, a missing-reagents table, next week's experiment plan, and a flawed procurement draft. The task now tests whether the agent can infer the procurement gaps, correct the flawed draft, and keep the result blocker-oriented.\"\ntask_type: short_term_contextual\ntrigger:\n  type: user\nintent:\n  initial_input: First read `inventory_checklist.md`, `missing_reagents.csv`, `next_week_experiment_plan.txt`, and `qpcr_procurement_draft.csv` in the current task directory; if anything is missing, say so first. Then revise the current procurement draft instead of starting from zero. Please update `qpcr_procurement_draft.csv` and briefly summarize what you changed after you finish.\n  hidden_intent:\n  - content: Treat `qpcr_procurement_draft.csv` as a flawed draft to be corrected rather than as a trusted procurement sheet.\n  - content: First infer the procurement gaps from the inventory, missing-item table, and next week's schedule, and do not generate a generic list of common consumables detached from the current inputs.\n  - content: Organize the items by experiment step or function instead of piling together supplies from unrelated stages.\n  - content: Rank first by what will block next week's schedule most seriously, and only then give the complete grouping; do not rank only by product category.\n  - content: Distinguish must-buy items from backup items.\n  - content: Directly flag omissions that could affect controls, replicates, or result interpretation.\n  - content: Separate things that are substitutable from things that should not be substituted casually.\n  - content: Correct any draft tendency to rank by category instead of by blocker severity.\n  - content: Give a more convenient default procurement order.\n\nmetadata:\n  difficulty: medium"}
{"schema_version": "v2", "role": "pharmacist", "task_id": "pharmacist_task_019", "user_id": "pharmacist", "environment_id": "env_pharmacist", "title": "Cross-Project Procurement Triage", "display_title": "Cross-Project Procurement Triage", "description": "The user has already sorted two procurement strands: eczema topical gel pilot materials and qPCR / cell-experiment blockers. The task tests whether the agent can merge those two inherited procurement artifacts into one slot-limited triage plan, correct a flawed draft that tries to buy everything at once, separate immediate purchases from confirmation-only decisions, avoid duplicate or route-tied buys, and keep progress viable if no reply arrives. The task is intentionally independent of the runtime date.", "task_type": "short_term_contextual", "difficulty": "hard", "trigger_type": "user", "initial_input": "First read `gel_procurement_reference.csv`, `qpcr_procurement_reference.csv`, `triage_rules.md`, `triage_constraints.json`, `confirmation_note.md`, and `merged_procurement_plan.json` in the current task directory; if anything is missing, say so first. Then revise `merged_procurement_plan.json` rather than starting from zero, and keep it valid JSON. The revised JSON should use separated fields for `primary_procurement_move`, `immediate_slots`, `next_wave_gel_basics`, `conditional_backup_lab_support`, `deferred_route_tied_confirmation`, `confirmation_action`, `fallback_if_no_reply`, and `draft_corrections`. In your final reply, use exactly these blocks in this order: `Primary Procurement Move`, `Immediate Slot Allocation`, `Merged Decision Table`, `Confirmation Needed`, `Deferred On Purpose`, `Fallback If No Reply`, `Draft Corrections`. The `Immediate Slot Allocation` block must be a markdown table with columns `Slot | Item | Strand | Why It Earns Immediate Slot`, and the `Merged Decision Table` must be a markdown table with columns `Bucket | Items | Reason`.", "hidden_intents": ["Carry forward both upstream procurement artifacts instead of writing a fresh generic shopping list.", "Put qPCR / cell blockers first because they can break the next experiment.", "Apply the eight-slot immediate-order cap instead of expanding the immediate bucket until every useful item is included.", "Keep carbomer-baseline gel pilot items available, but do not let final packaging, preservative, alternative-route items, backup cell-culture items, or ambiguous duplicate filter needs enter the immediate purchase bucket.", "Treat `merged_procurement_plan.json` as a flawed draft to repair, not as a trusted plan and not as a file to ignore.", "Correct the draft's packaging-first, universal-cart, duplicated-filter, backup-as-immediate, and route-tied-buy-now tendencies.", "Use one concise confirmation action to resolve the few decisions that change purchasing, instead of asking every open question.", "Keep packaging, preservative, alternative gel-base choices, and capacity-limited next-wave basics intentionally deferred unless confirmation changes the plan.", "Preserve progress if no reply arrives by ordering the exact eight-slot plan instead of waiting for confirmation before ordering everything."], "hidden_intent_count": 9, "has_objectives": true, "objectives_json": "{\"checklist\": [{\"criterion\": \"The final reply must contain the seven required top-level blocks in this exact order: `Primary Procurement Move`, `Immediate Slot Allocation`, `Merged Decision Table`, `Confirmation Needed`, `Deferred On Purpose`, `Fallback If No Reply`, and `Draft Corrections`. A trailing completion marker such as `Complete` is acceptable, but extra analytical sections are not.\"}, {\"criterion\": \"The `Primary Procurement Move` block must state that the first move is a capped eight-slot order led by qPCR / cell blockers, explain that the next qPCR run or CCK-8 / 96-well viability screen would be blocked if those items are delayed, and explicitly reject the draft's packaging-first or buy-everything-first approach.\"}, {\"criterion\": \"The `Immediate Slot Allocation` block must be a markdown table with exactly these columns in this order: `Slot | Item | Strand | Why It Earns Immediate Slot`. It must contain exactly 8 item rows, not 7 or 9+, and the slots must be numbered 1 through 8.\"}, {\"criterion\": \"The exact five qPCR / cell immediate-slot items must be `Target-gene primers`, `Sealing film`, `RNase-free filter tips`, `CCK-8 reagent`, and `96-well cell plates`. These five must all appear in `Immediate Slot Allocation`, and `FBS`, `Trypsin`, and `0.22 um sterile filter` must not appear there as immediate-slot items.\"}, {\"criterion\": \"The exact three gel immediate-slot items must be `Carbomer 940`, `Triethanolamine`, and `Propylene Glycol`. These three must appear in `Immediate Slot Allocation`, and `Glycerin`, `Disodium EDTA / Citric Acid` or `EDTA / Citric Acid`, `0.22 um filter`, and `Spatula and Ointment Jar` must not appear there as immediate-slot items.\"}, {\"criterion\": \"The `Merged Decision Table` block must be a markdown table with exactly these columns in this order: `Bucket | Items | Reason`. It must include at least four distinct buckets: `Immediate 8-slot order`, `Next-wave gel basics`, `Conditional / backup lab support`, and `Deferred route-tied confirmation` or wording with the same four meanings.\"}, {\"criterion\": \"The `Merged Decision Table` or `Deferred On Purpose` block must keep `Aluminum Tube`, `Airless Pump Bottle`, `Phenoxyethanol` or `preservative system`, and `Hydroxyethyl Cellulose` out of immediate buy-now status, and must place them in deferred or confirmation-only logic.\"}, {\"criterion\": \"The `Confirmation Needed` block must give exactly one concise confirmation action that covers all three decisions: whether the top qPCR / cell blockers are the five immediate-slot blockers or should be reprioritized, whether carbomer baseline remains the default gel route, and whether next-wave gel basics plus packaging / preservative decisions should stay postponed. It must not ask every open question or ask for a full procurement meeting.\"}, {\"criterion\": \"The `Deferred On Purpose` block must explain both types of deferral: at least two route-tied deferrals from `Aluminum Tube`, `Airless Pump Bottle`, `Phenoxyethanol`, `preservative system`, `Hydroxyethyl Cellulose`, or `alternative gel base`, and at least one capacity-limited non-route deferral from `Glycerin`, `Disodium EDTA / Citric Acid`, `EDTA / Citric Acid`, `0.22 um filter`, `0.22 um sterile filter`, `Spatula and Ointment Jar`, `FBS`, or `Trypsin`.\"}, {\"criterion\": \"The reply must explicitly resolve the duplicate-filter trap: it must state that `0.22 um filter` / `0.22 um sterile filter` should not be counted as two separate immediate buys, and it must place the filter need in next-wave, conditional, or confirmation logic rather than in `Immediate Slot Allocation`.\"}, {\"criterion\": \"The `Fallback If No Reply` block must say not to freeze the plan: proceed with the exact eight immediate slots (`Target-gene primers`, `Sealing film`, `RNase-free filter tips`, `CCK-8 reagent`, `96-well cell plates`, `Carbomer 940`, `Triethanolamine`, and `Propylene Glycol`) while keeping backup culture items, next-wave gel basics, packaging, preservative, and alternative-route choices deferred.\"}, {\"criterion\": \"The `Draft Corrections` block must identify at least four repaired mistakes from the flawed JSON draft, including `packaging first`, `universal purchase bucket` or `buy everything at once`, `preservative treated as buy-now`, and either `FBS / Trypsin / filter treated as immediate despite the slot cap`, `empty deferred or confirmation bucket`, or `wait for reply before ordering everything`.\"}]}", "metadata_json": "{\"difficulty\": \"hard\"}", "yaml_path": "data/pharmacist/tasks/pharmacist_task_019/task.yaml", "yaml": "schema_version: v2\ntask_id: pharmacist_task_019\nuser_id: pharmacist\nenvironment_id: env_pharmacist\ntitle: Cross-Project Procurement Triage\ndescription: \"The user has already sorted two procurement strands: eczema topical gel pilot materials and qPCR / cell-experiment blockers. The task tests whether the agent can merge those two inherited procurement artifacts into one slot-limited triage plan, correct a flawed draft that tries to buy everything at once, separate immediate purchases from confirmation-only decisions, avoid duplicate or route-tied buys, and keep progress viable if no reply arrives. The task is intentionally independent of the runtime date.\"\ntask_type: short_term_contextual\n\ntrigger:\n  type: user\n\nintent:\n  initial_input: |\n    First read `gel_procurement_reference.csv`, `qpcr_procurement_reference.csv`, `triage_rules.md`, `triage_constraints.json`, `confirmation_note.md`, and `merged_procurement_plan.json` in the current task directory; if anything is missing, say so first. Then revise `merged_procurement_plan.json` rather than starting from zero, and keep it valid JSON. The revised JSON should use separated fields for `primary_procurement_move`, `immediate_slots`, `next_wave_gel_basics`, `conditional_backup_lab_support`, `deferred_route_tied_confirmation`, `confirmation_action`, `fallback_if_no_reply`, and `draft_corrections`. In your final reply, use exactly these blocks in this order: `Primary Procurement Move`, `Immediate Slot Allocation`, `Merged Decision Table`, `Confirmation Needed`, `Deferred On Purpose`, `Fallback If No Reply`, `Draft Corrections`. The `Immediate Slot Allocation` block must be a markdown table with columns `Slot | Item | Strand | Why It Earns Immediate Slot`, and the `Merged Decision Table` must be a markdown table with columns `Bucket | Items | Reason`.\n\n  hidden_intent:\n    - content: Carry forward both upstream procurement artifacts instead of writing a fresh generic shopping list.\n    - content: Put qPCR / cell blockers first because they can break the next experiment.\n    - content: Apply the eight-slot immediate-order cap instead of expanding the immediate bucket until every useful item is included.\n    - content: Keep carbomer-baseline gel pilot items available, but do not let final packaging, preservative, alternative-route items, backup cell-culture items, or ambiguous duplicate filter needs enter the immediate purchase bucket.\n    - content: Treat `merged_procurement_plan.json` as a flawed draft to repair, not as a trusted plan and not as a file to ignore.\n    - content: Correct the draft's packaging-first, universal-cart, duplicated-filter, backup-as-immediate, and route-tied-buy-now tendencies.\n    - content: Use one concise confirmation action to resolve the few decisions that change purchasing, instead of asking every open question.\n    - content: Keep packaging, preservative, alternative gel-base choices, and capacity-limited next-wave basics intentionally deferred unless confirmation changes the plan.\n    - content: Preserve progress if no reply arrives by ordering the exact eight-slot plan instead of waiting for confirmation before ordering everything.\n\nobjectives:\n  checklist:\n    - criterion: \"The final reply must contain the seven required top-level blocks in this exact order: `Primary Procurement Move`, `Immediate Slot Allocation`, `Merged Decision Table`, `Confirmation Needed`, `Deferred On Purpose`, `Fallback If No Reply`, and `Draft Corrections`. A trailing completion marker such as `Complete` is acceptable, but extra analytical sections are not.\"\n    - criterion: \"The `Primary Procurement Move` block must state that the first move is a capped eight-slot order led by qPCR / cell blockers, explain that the next qPCR run or CCK-8 / 96-well viability screen would be blocked if those items are delayed, and explicitly reject the draft's packaging-first or buy-everything-first approach.\"\n    - criterion: \"The `Immediate Slot Allocation` block must be a markdown table with exactly these columns in this order: `Slot | Item | Strand | Why It Earns Immediate Slot`. It must contain exactly 8 item rows, not 7 or 9+, and the slots must be numbered 1 through 8.\"\n    - criterion: \"The exact five qPCR / cell immediate-slot items must be `Target-gene primers`, `Sealing film`, `RNase-free filter tips`, `CCK-8 reagent`, and `96-well cell plates`. These five must all appear in `Immediate Slot Allocation`, and `FBS`, `Trypsin`, and `0.22 um sterile filter` must not appear there as immediate-slot items.\"\n    - criterion: \"The exact three gel immediate-slot items must be `Carbomer 940`, `Triethanolamine`, and `Propylene Glycol`. These three must appear in `Immediate Slot Allocation`, and `Glycerin`, `Disodium EDTA / Citric Acid` or `EDTA / Citric Acid`, `0.22 um filter`, and `Spatula and Ointment Jar` must not appear there as immediate-slot items.\"\n    - criterion: \"The `Merged Decision Table` block must be a markdown table with exactly these columns in this order: `Bucket | Items | Reason`. It must include at least four distinct buckets: `Immediate 8-slot order`, `Next-wave gel basics`, `Conditional / backup lab support`, and `Deferred route-tied confirmation` or wording with the same four meanings.\"\n    - criterion: \"The `Merged Decision Table` or `Deferred On Purpose` block must keep `Aluminum Tube`, `Airless Pump Bottle`, `Phenoxyethanol` or `preservative system`, and `Hydroxyethyl Cellulose` out of immediate buy-now status, and must place them in deferred or confirmation-only logic.\"\n    - criterion: \"The `Confirmation Needed` block must give exactly one concise confirmation action that covers all three decisions: whether the top qPCR / cell blockers are the five immediate-slot blockers or should be reprioritized, whether carbomer baseline remains the default gel route, and whether next-wave gel basics plus packaging / preservative decisions should stay postponed. It must not ask every open question or ask for a full procurement meeting.\"\n    - criterion: \"The `Deferred On Purpose` block must explain both types of deferral: at least two route-tied deferrals from `Aluminum Tube`, `Airless Pump Bottle`, `Phenoxyethanol`, `preservative system`, `Hydroxyethyl Cellulose`, or `alternative gel base`, and at least one capacity-limited non-route deferral from `Glycerin`, `Disodium EDTA / Citric Acid`, `EDTA / Citric Acid`, `0.22 um filter`, `0.22 um sterile filter`, `Spatula and Ointment Jar`, `FBS`, or `Trypsin`.\"\n    - criterion: \"The reply must explicitly resolve the duplicate-filter trap: it must state that `0.22 um filter` / `0.22 um sterile filter` should not be counted as two separate immediate buys, and it must place the filter need in next-wave, conditional, or confirmation logic rather than in `Immediate Slot Allocation`.\"\n    - criterion: \"The `Fallback If No Reply` block must say not to freeze the plan: proceed with the exact eight immediate slots (`Target-gene primers`, `Sealing film`, `RNase-free filter tips`, `CCK-8 reagent`, `96-well cell plates`, `Carbomer 940`, `Triethanolamine`, and `Propylene Glycol`) while keeping backup culture items, next-wave gel basics, packaging, preservative, and alternative-route choices deferred.\"\n    - criterion: \"The `Draft Corrections` block must identify at least four repaired mistakes from the flawed JSON draft, including `packaging first`, `universal purchase bucket` or `buy everything at once`, `preservative treated as buy-now`, and either `FBS / Trypsin / filter treated as immediate despite the slot cap`, `empty deferred or confirmation bucket`, or `wait for reply before ordering everything`.\"\n\nmetadata:\n  difficulty: hard"}
{"schema_version": "v2", "role": "pharmacist", "task_id": "pharmacist_task_020", "user_id": "pharmacist", "environment_id": "env_pharmacist", "title": "Sandbox File System Manifest-Driven File-and-Directory Copy/Move Closed Loop", "display_title": "Sandbox File System Manifest-Driven File-and-Directory Copy/Move Closed Loop", "description": "Inside the AppWorld sandbox only, the user wants to clean up a temporary reading-material directory in a more systematic way. The task tests whether the agent reads a local cleanup manifest, creates the listed sandbox directories and file, then copies and moves both files and directories according to the manifest.", "task_type": "short_term_independent", "difficulty": "medium", "trigger_type": "user", "initial_input": "This is an AppWorld sandbox task; the account identifiers below are synthetic test credentials, not real external accounts. First read `materials_cleanup_manifest.csv` in the current task directory. In the sandbox file system app, clean up that temporary materials directory according to the manifest. Please create the source directory and source file first, prepare the destination/archive directories, then copy the file once and move it once, and finally perform the directory-level copy and directory-level move according to the manifest. Login with username \"la-mcco@gmail.com\" and sandbox credential \"LlU7G2h\". In your final response, use exactly these blocks in this order: `Manifest Paths`, `File Operations`, `Directory Operations`.", "hidden_intents": ["Log into the sandbox file system first and reuse the same session token.", "Extract the exact source, copy, move, backup, and archive paths from `materials_cleanup_manifest.csv`.", "There must be directories and a file before any copy or move action; do not operate on non-existent paths.", "Cover both file copy/move and directory copy/move.", "In the final response, use the required blocks and restate the manifest paths, file operations, and directory operations."], "hidden_intent_count": 5, "has_objectives": true, "objectives_json": "{\"checklist\": [{\"criterion\": \"The final response must contain exactly these three top-level blocks in this order: `Manifest Paths`, `File Operations`, and `Directory Operations`.\"}, {\"criterion\": \"The `Manifest Paths` block must summarize the created source and destination/archive paths from the manifest, including `pharmacy_materials/source`, `pharmacy_materials/source/reading_notes.txt`, `pharmacy_materials/organized`, and `pharmacy_materials/archive`.\"}, {\"criterion\": \"The `File Operations` block must clearly describe the file copy from `pharmacy_materials/source/reading_notes.txt` to `pharmacy_materials/organized/reading_notes_copy.txt` and the file move to `pharmacy_materials/organized/reading_notes_moved.txt`.\"}, {\"criterion\": \"The `Directory Operations` block must clearly describe the directory copy from `pharmacy_materials/source` to `pharmacy_materials/source_backup` and the directory move to `pharmacy_materials/archive/source_moved`.\"}], \"tools\": [{\"mcp_appworld_file_system__login\": {\"call\": [\"username\", \"password\"], \"result\": [\"response\"]}}, {\"mcp_appworld_file_system__create_directory\": {\"call\": [\"directory_path\", \"access_token\"], \"result\": [\"response\"]}}, {\"mcp_appworld_file_system__create_file\": {\"call\": [\"file_path\", \"access_token\"], \"result\": [\"response\"]}}, {\"mcp_appworld_file_system__copy_file\": {\"call\": [\"source_file_path\", \"destination_file_path\", \"access_token\"], \"result\": [\"response\"]}}, {\"mcp_appworld_file_system__move_file\": {\"call\": [\"source_file_path\", \"destination_file_path\", \"access_token\"], \"result\": [\"response\"]}}, {\"mcp_appworld_file_system__copy_directory\": {\"call\": [\"source_directory_path\", \"destination_directory_path\", \"access_token\"], \"result\": [\"response\"]}}, {\"mcp_appworld_file_system__move_directory\": {\"call\": [\"source_directory_path\", \"destination_directory_path\", \"access_token\"], \"result\": [\"response\"]}}], \"tools_evaluation_path\": \"data/pharmacist/scripts/pharmacist_task_020_file_system.py\"}", "metadata_json": "{\"difficulty\": \"medium\"}", "yaml_path": "data/pharmacist/tasks/pharmacist_task_020/task.yaml", "yaml": "schema_version: v2\ntask_id: pharmacist_task_020\nuser_id: pharmacist\nenvironment_id: env_pharmacist\ntitle: Sandbox File System Manifest-Driven File-and-Directory Copy/Move Closed Loop\ndescription: \"Inside the AppWorld sandbox only, the user wants to clean up a temporary reading-material directory in a more systematic way. The task tests whether the agent reads a local cleanup manifest, creates the listed sandbox directories and file, then copies and moves both files and directories according to the manifest.\"\ntask_type: short_term_independent\ntrigger:\n  type: user\nintent:\n  initial_input: |\n    This is an AppWorld sandbox task; the account identifiers below are synthetic test credentials, not real external accounts. First read `materials_cleanup_manifest.csv` in the current task directory. In the sandbox file system app, clean up that temporary materials directory according to the manifest. Please create the source directory and source file first, prepare the destination/archive directories, then copy the file once and move it once, and finally perform the directory-level copy and directory-level move according to the manifest. Login with username \"la-mcco@gmail.com\" and sandbox credential \"LlU7G2h\". In your final response, use exactly these blocks in this order: `Manifest Paths`, `File Operations`, `Directory Operations`.\n  hidden_intent:\n    - content: Log into the sandbox file system first and reuse the same session token.\n    - content: Extract the exact source, copy, move, backup, and archive paths from `materials_cleanup_manifest.csv`.\n    - content: There must be directories and a file before any copy or move action; do not operate on non-existent paths.\n    - content: Cover both file copy/move and directory copy/move.\n    - content: In the final response, use the required blocks and restate the manifest paths, file operations, and directory operations.\nobjectives:\n  tools:\n    - mcp_appworld_file_system__login:\n        call: [username, password]\n        result: [response]\n    - mcp_appworld_file_system__create_directory:\n        call: [directory_path, access_token]\n        result: [response]\n    - mcp_appworld_file_system__create_file:\n        call: [file_path, access_token]\n        result: [response]\n    - mcp_appworld_file_system__copy_file:\n        call: [source_file_path, destination_file_path, access_token]\n        result: [response]\n    - mcp_appworld_file_system__move_file:\n        call: [source_file_path, destination_file_path, access_token]\n        result: [response]\n    - mcp_appworld_file_system__copy_directory:\n        call: [source_directory_path, destination_directory_path, access_token]\n        result: [response]\n    - mcp_appworld_file_system__move_directory:\n        call: [source_directory_path, destination_directory_path, access_token]\n        result: [response]\n  checklist:\n    - criterion: \"The final response must contain exactly these three top-level blocks in this order: `Manifest Paths`, `File Operations`, and `Directory Operations`.\"\n    - criterion: \"The `Manifest Paths` block must summarize the created source and destination/archive paths from the manifest, including `pharmacy_materials/source`, `pharmacy_materials/source/reading_notes.txt`, `pharmacy_materials/organized`, and `pharmacy_materials/archive`.\"\n    - criterion: \"The `File Operations` block must clearly describe the file copy from `pharmacy_materials/source/reading_notes.txt` to `pharmacy_materials/organized/reading_notes_copy.txt` and the file move to `pharmacy_materials/organized/reading_notes_moved.txt`.\"\n    - criterion: \"The `Directory Operations` block must clearly describe the directory copy from `pharmacy_materials/source` to `pharmacy_materials/source_backup` and the directory move to `pharmacy_materials/archive/source_moved`.\"\n  tools_evaluation_path: data/pharmacist/scripts/pharmacist_task_020_file_system.py\nmetadata:\n  difficulty: medium"}
{"schema_version": "v2", "role": "researcher", "task_id": "researcher_task_001", "user_id": "researcher", "environment_id": "env_researcher", "title": "Practical Follow-Up Value Assessment of Multimodal Tool-Use Papers", "display_title": "Practical Follow-Up Value Assessment of Multimodal Tool-Use Papers", "description": "Around recent multimodal tool-use work, select representative papers that are widely followed and reproducible, and provide research-oriented evaluation from the perspectives of onboarding difficulty and concrete follow-up paths.", "task_type": "long_term", "difficulty": "easy", "trigger_type": "user", "initial_input": "I've been struggling with paper reading lately, and many multimodal tool-use works feel hard to follow. Could you help me find papers that people have commonly followed recently?", "hidden_intents": ["Provide a brief introduction and key technical points for each paper.", "Indicate whether it is open-source and provide the GitHub link (if not open-source, explicitly state that).", "From a research perspective, evaluate whether it is suitable to follow and provide a concrete follow-up path."], "hidden_intent_count": 3, "has_objectives": false, "objectives_json": "", "metadata_json": "{\"difficulty\": \"easy\"}", "yaml_path": "data/researcher/tasks/researcher_task_001/task.yaml", "yaml": "schema_version: v2\ntask_id: researcher_task_001\nuser_id: researcher\nenvironment_id: env_researcher\n\ntitle: \"Practical Follow-Up Value Assessment of Multimodal Tool-Use Papers\"\ndescription: \"Around recent multimodal tool-use work, select representative papers that are widely followed and reproducible, and provide research-oriented evaluation from the perspectives of onboarding difficulty and concrete follow-up paths.\"\n\ntask_type: long_term\n\ntrigger:\n  type: user\n\nintent:\n  initial_input: \"I've been struggling with paper reading lately, and many multimodal tool-use works feel hard to follow. Could you help me find papers that people have commonly followed recently?\"\n  hidden_intent:\n    - content: \"Provide a brief introduction and key technical points for each paper.\"\n\n    - content: \"Indicate whether it is open-source and provide the GitHub link (if not open-source, explicitly state that).\"\n\n    - content: \"From a research perspective, evaluate whether it is suitable to follow and provide a concrete follow-up path.\"\n\nmetadata:\n  difficulty: easy"}
{"schema_version": "v2", "role": "researcher", "task_id": "researcher_task_002", "user_id": "researcher", "environment_id": "env_researcher", "title": "Starting Lean Bulk: Cost-Effective Gym Gear Selection", "display_title": "Starting Lean Bulk: Cost-Effective Gym Gear Selection", "description": "The user is preparing to begin a structured muscle-gain plan and needs to choose a set of quick-dry workout clothing and protective gear suitable for use in the campus gym, based on his body metrics and limited budget.", "task_type": "long_term", "difficulty": "easy", "trigger_type": "user", "initial_input": "I'm planning to start a structured muscle-gain program. Could you help me pick a few cost-effective workout clothing items?", "hidden_intents": ["Current body metrics: height 175 cm, weight 68 kg, body fat around 17%. Recommend clothing sizes that would fit my build.", "Provide a substitute option (Plan B) to handle unexpected situations (for example, if a particular item is unavailable in my size).", "The output should be presented in a clear, readable table format for quick one-pass comparison."], "hidden_intent_count": 3, "has_objectives": false, "objectives_json": "", "metadata_json": "{\"difficulty\": \"easy\"}", "yaml_path": "data/researcher/tasks/researcher_task_002/task.yaml", "yaml": "schema_version: v2\ntask_id: researcher_task_002\nuser_id: researcher\nenvironment_id: env_researcher\n\ntitle: \"Starting Lean Bulk: Cost-Effective Gym Gear Selection\"\ndescription: \"The user is preparing to begin a structured muscle-gain plan and needs to choose a set of quick-dry workout clothing and protective gear suitable for use in the campus gym, based on his body metrics and limited budget.\"\n\ntask_type: long_term\n\ntrigger:\n  type: user\n\nintent:\n  initial_input: \"I'm planning to start a structured muscle-gain program. Could you help me pick a few cost-effective workout clothing items?\"\n  hidden_intent:\n  - content: \"Current body metrics: height 175 cm, weight 68 kg, body fat around 17%. Recommend clothing sizes that would fit my build.\"\n\n  - content: \"Provide a substitute option (Plan B) to handle unexpected situations (for example, if a particular item is unavailable in my size).\"\n\n  - content: \"The output should be presented in a clear, readable table format for quick one-pass comparison.\"\n\nmetadata:\n  difficulty: easy"}
{"schema_version": "v2", "role": "researcher", "task_id": "researcher_task_003", "user_id": "researcher", "environment_id": "env_researcher", "title": "Forced to Leave Dorm Housing: Renting and Commute Baselines for a 'Jiangwan Local'", "display_title": "Forced to Leave Dorm Housing: Renting and Commute Baselines for a 'Jiangwan Local'", "description": "A guide to avoiding pitfalls when renting off-campus.", "task_type": "long_term", "difficulty": "easy", "trigger_type": "user", "initial_input": "My advisor recommended an internship opportunity to me, but I am not very interested in this role. Also, as a first-year PhD student, I still have classes at Fudan University's Jiangwan campus. Below is the job description. Please help me draft a polite message(in English) to decline it and explain my reasons tactfully.:\n\n# Algorithm Engineer Intern, TikTok Streaming Optimization\n\n**Location:** Beijing  \n**Compensation:** 300 RMB/day\n\n## About the Role\nWe are looking for an **Algorithm Engineer Intern** to work on **internal streaming algorithm optimization** at TikTok. In this role, you will contribute to improving large-scale content delivery and recommendation efficiency, helping optimize core streaming performance in real production environments.\n\nYou will work closely with experienced engineers and researchers on model iteration, strategy improvement, and experimental analysis for high-impact recommendation and distribution systems.\n\n## Responsibilities\n- Participate in the optimization and iteration of TikTok’s internal streaming algorithms, including areas such as candidate generation, pre-ranking, ranking, and re-ranking.\n- Design, implement, and evaluate algorithmic solutions to improve distribution efficiency, user experience, and key business metrics.\n- Work with large-scale user behavior and content data to support model training, tuning, offline evaluation, and performance analysis.\n- Collaborate with cross-functional teams including algorithm, engineering, and product to drive solutions from research to deployment.\n- Explore and apply advanced machine learning and deep learning methods in real-world streaming and recommendation scenarios.\n\n## Qualifications\n- Currently pursuing a Bachelor's, Master's, or PhD degree in Computer Science, Artificial Intelligence, Machine Learning, Data Science, Mathematics, Statistics, or a related field.\n- Solid foundation in machine learning and deep learning.\n- Strong programming skills in **Python**.\n- Familiarity with at least one deep learning framework such as **PyTorch** or **TensorFlow**.\n- Strong interest in recommendation systems, ranking algorithms, large-scale machine learning, or streaming optimization.\n- Good analytical thinking, problem-solving ability, and teamwork skills.\n- Available to intern **at least 4 days per week** for **3 months or longer**.\n\n## Preferred Qualifications\n- Prior experience in recommendation systems, search ranking, ad systems, or feed/streaming algorithms.\n- Experience with large-scale data processing, feature engineering, or A/B testing analysis.\n- Research or project experience in machine learning, data mining, or recommendation systems.\n- Familiarity with distributed training, model acceleration, or inference optimization is a plus.\n\n## What You Will Gain\n- Hands-on experience working on TikTok’s core streaming and recommendation optimization problems.\n- Exposure to real-world industrial-scale recommendation systems and large-scale machine learning workflows.\n- Strong computing support: **on average, each intern has access to 4× RTX 4090 GPUs**.\n- Competitive internship compensation: **300 RMB/day**.\n- Opportunity to work with a highly capable algorithm team and grow quickly through real production challenges.\n\n## Work Location\n**Beijing**", "hidden_intents": ["In the message to my advisor, point out that the position is based in Beijing, while I am still taking classes in Shanghai as a first-year PhD student, so commuting would be impractical. At the moment, I am only considering opportunities with a commute of no more than one hour.(Similar meaning is enough)", "In the message to my advisor, say that this is more of a business-oriented engineering role rather than a research internship, which would not be very helpful for publishing papers or producing academic output.(Similar meaning is enough)", "In the message to my advisor, explain that my current research focus is think-with-image, and my background isn’t a very strong match for the recommendation systems focus described in the job posting.(Similar meaning is enough)", "In the message to my advisor, point out that the computing resources for this role average only 4× RTX 4090 GPUs per person, which I consider insufficient. My research would require at least 16 A100 GPUs per person. (Similar meaning is enough)", "In the message to my advisor, mention that the daily pay is only 300 RMB/day, which is below the average market rate for PhD internships, typically around 500 RMB/day.(Similar meaning is enough)"], "hidden_intent_count": 5, "has_objectives": false, "objectives_json": "", "metadata_json": "{\"difficulty\": \"easy\"}", "yaml_path": "data/researcher/tasks/researcher_task_003/task.yaml", "yaml": "schema_version: v2\ntask_id: researcher_task_003\nuser_id: researcher\nenvironment_id: env_researcher\n\ntitle: \"Forced to Leave Dorm Housing: Renting and Commute Baselines for a 'Jiangwan Local'\"\ndescription: \"A guide to avoiding pitfalls when renting off-campus.\"\n\ntask_type: long_term\n\ntrigger:\n  type: user\n\nintent:\n  initial_input: |\n    My advisor recommended an internship opportunity to me, but I am not very interested in this role. Also, as a first-year PhD student, I still have classes at Fudan University's Jiangwan campus. Below is the job description. Please help me draft a polite message(in English) to decline it and explain my reasons tactfully.:\n\n    # Algorithm Engineer Intern, TikTok Streaming Optimization\n\n    **Location:** Beijing  \n    **Compensation:** 300 RMB/day\n\n    ## About the Role\n    We are looking for an **Algorithm Engineer Intern** to work on **internal streaming algorithm optimization** at TikTok. In this role, you will contribute to improving large-scale content delivery and recommendation efficiency, helping optimize core streaming performance in real production environments.\n\n    You will work closely with experienced engineers and researchers on model iteration, strategy improvement, and experimental analysis for high-impact recommendation and distribution systems.\n\n    ## Responsibilities\n    - Participate in the optimization and iteration of TikTok’s internal streaming algorithms, including areas such as candidate generation, pre-ranking, ranking, and re-ranking.\n    - Design, implement, and evaluate algorithmic solutions to improve distribution efficiency, user experience, and key business metrics.\n    - Work with large-scale user behavior and content data to support model training, tuning, offline evaluation, and performance analysis.\n    - Collaborate with cross-functional teams including algorithm, engineering, and product to drive solutions from research to deployment.\n    - Explore and apply advanced machine learning and deep learning methods in real-world streaming and recommendation scenarios.\n\n    ## Qualifications\n    - Currently pursuing a Bachelor's, Master's, or PhD degree in Computer Science, Artificial Intelligence, Machine Learning, Data Science, Mathematics, Statistics, or a related field.\n    - Solid foundation in machine learning and deep learning.\n    - Strong programming skills in **Python**.\n    - Familiarity with at least one deep learning framework such as **PyTorch** or **TensorFlow**.\n    - Strong interest in recommendation systems, ranking algorithms, large-scale machine learning, or streaming optimization.\n    - Good analytical thinking, problem-solving ability, and teamwork skills.\n    - Available to intern **at least 4 days per week** for **3 months or longer**.\n\n    ## Preferred Qualifications\n    - Prior experience in recommendation systems, search ranking, ad systems, or feed/streaming algorithms.\n    - Experience with large-scale data processing, feature engineering, or A/B testing analysis.\n    - Research or project experience in machine learning, data mining, or recommendation systems.\n    - Familiarity with distributed training, model acceleration, or inference optimization is a plus.\n\n    ## What You Will Gain\n    - Hands-on experience working on TikTok’s core streaming and recommendation optimization problems.\n    - Exposure to real-world industrial-scale recommendation systems and large-scale machine learning workflows.\n    - Strong computing support: **on average, each intern has access to 4× RTX 4090 GPUs**.\n    - Competitive internship compensation: **300 RMB/day**.\n    - Opportunity to work with a highly capable algorithm team and grow quickly through real production challenges.\n\n    ## Work Location\n    **Beijing**\n\n  hidden_intent:\n    - content: \"In the message to my advisor, point out that the position is based in Beijing, while I am still taking classes in Shanghai as a first-year PhD student, so commuting would be impractical. At the moment, I am only considering opportunities with a commute of no more than one hour.(Similar meaning is enough)\"\n\n    - content: \"In the message to my advisor, say that this is more of a business-oriented engineering role rather than a research internship, which would not be very helpful for publishing papers or producing academic output.(Similar meaning is enough)\"\n\n    - content: \"In the message to my advisor, explain that my current research focus is think-with-image, and my background isn’t a very strong match for the recommendation systems focus described in the job posting.(Similar meaning is enough)\"\n\n    - content: \"In the message to my advisor, point out that the computing resources for this role average only 4× RTX 4090 GPUs per person, which I consider insufficient. My research would require at least 16 A100 GPUs per person. (Similar meaning is enough)\"\n\n    - content: \"In the message to my advisor, mention that the daily pay is only 300 RMB/day, which is below the average market rate for PhD internships, typically around 500 RMB/day.(Similar meaning is enough)\"\n\nmetadata:\n  difficulty: easy"}
{"schema_version": "v2", "role": "researcher", "task_id": "researcher_task_004", "user_id": "researcher", "environment_id": "env_researcher", "title": "Preliminary Multimodal Agent Survey: From Text-only to Think-with-Image", "display_title": "Preliminary Multimodal Agent Survey: From Text-only to Think-with-Image", "description": "The user has started exploring the multimodal agent field and wants to learn about early representative works that treat images as intermediate reasoning states (Think-with-Image), while also aligning on a preferred information presentation format.", "task_type": "long_term", "difficulty": "easy", "trigger_type": "user", "initial_input": "I've recently become very interested in multimodal agents, especially work that does not treat images as mere inputs but can 'think while looking at images' (think-with-image) more like humans. Can you help me find 2-3 representative papers from 2024 or 2025?", "hidden_intents": ["Provide an OpenReview link for each paper; if no OpenReview link is available, provide an arXiv link.", "It would be best to recommend 3 to 5 papers at a time, since anything more would be hard to keep up with.", "For each paper, provide a brief introduction and key technical points."], "hidden_intent_count": 3, "has_objectives": false, "objectives_json": "", "metadata_json": "{\"difficulty\": \"easy\"}", "yaml_path": "data/researcher/tasks/researcher_task_004/task.yaml", "yaml": "schema_version: v2\ntask_id: researcher_task_004\nuser_id: researcher\nenvironment_id: env_researcher\n\ntitle: \"Preliminary Multimodal Agent Survey: From Text-only to Think-with-Image\"\ndescription: \"The user has started exploring the multimodal agent field and wants to learn about early representative works that treat images as intermediate reasoning states (Think-with-Image), while also aligning on a preferred information presentation format.\"\n\ntask_type: long_term\n\ntrigger:\n  type: user\n\nintent:\n  initial_input: \"I've recently become very interested in multimodal agents, especially work that does not treat images as mere inputs but can 'think while looking at images' (think-with-image) more like humans. Can you help me find 2-3 representative papers from 2024 or 2025?\"\n  hidden_intent:\n    - content: \"Provide an OpenReview link for each paper; if no OpenReview link is available, provide an arXiv link.\"\n    - content: \"It would be best to recommend 3 to 5 papers at a time, since anything more would be hard to keep up with.\"\n    - content: \"For each paper, provide a brief introduction and key technical points.\"\n\nmetadata:\n  difficulty: easy"}
{"schema_version": "v2", "role": "researcher", "task_id": "researcher_task_005", "user_id": "researcher", "environment_id": "env_researcher", "title": "Tracking the Frontier: DeepEyesV2 and V-star Bench Core Metrics Survey", "display_title": "Tracking the Frontier: DeepEyesV2 and V-star Bench Core Metrics Survey", "description": "Evaluation task: track frontier progress—survey core metrics for DeepEyesV2 and the V-star benchmark.", "task_type": "short_term_independent", "difficulty": "easy", "trigger_type": "user", "initial_input": "I just noticed DeepEyes recently released a follow-up called DeepEyesV2. Take a look and let's discuss it.", "hidden_intents": ["What benchmarks are commonly evaluated for think-with-image? Which benchmarks does DeepEyesV2 report?", "For commonly compared think-with-image methods such as Pixel-Reasoner, Thyme, DeepEyes, and DeepEyesV2, what scores does the DeepEyesV2 paper report on V* / V-star?", "Record leading-method performance on V* / V-star for later comparison: Pixel-Reasoner (84.3), Thyme (82.2), DeepEyes (85.6), DeepEyesV2 (81.8)."], "hidden_intent_count": 3, "has_objectives": false, "objectives_json": "", "metadata_json": "{\"difficulty\": \"easy\"}", "yaml_path": "data/researcher/tasks/researcher_task_005/task.yaml", "yaml": "schema_version: v2\ntask_id: researcher_task_005\nuser_id: researcher\nenvironment_id: env_researcher\n\ntitle: \"Tracking the Frontier: DeepEyesV2 and V-star Bench Core Metrics Survey\"\ndescription: \"Evaluation task: track frontier progress—survey core metrics for DeepEyesV2 and the V-star benchmark.\"\n\ntask_type: short_term_independent\n\ntrigger:\n  type: user\n\nintent:\n  initial_input: \"I just noticed DeepEyes recently released a follow-up called DeepEyesV2. Take a look and let's discuss it.\"\n  hidden_intent:\n    - content: \"What benchmarks are commonly evaluated for think-with-image? Which benchmarks does DeepEyesV2 report?\"\n\n    - content: \"For commonly compared think-with-image methods such as Pixel-Reasoner, Thyme, DeepEyes, and DeepEyesV2, what scores does the DeepEyesV2 paper report on V* / V-star?\"\n\n    - content: \"Record leading-method performance on V* / V-star for later comparison: Pixel-Reasoner (84.3), Thyme (82.2), DeepEyes (85.6), DeepEyesV2 (81.8).\"\n\nmetadata:\n  difficulty: easy"}
{"schema_version": "v2", "role": "researcher", "task_id": "researcher_task_006", "user_id": "researcher", "environment_id": "env_researcher", "title": "ICLR 2026 Accepted Papers: Survey on Think-with-Image Related Topics", "display_title": "ICLR 2026 Accepted Papers: Survey on Think-with-Image Related Topics", "description": "For a small 'think-with-image' tool-use agent project starting from scratch, use the list of ICLR 2026 accepted papers to filter and organize the most relevant works on image-driven reasoning, visual chain-of-thought (visual CoT), and multimodal agent tool use, producing outputs that can directly support subsequent topic focusing.", "task_type": "long_term", "difficulty": "hard", "trigger_type": "user", "initial_input": "I wrote a local file at paper_list.txt that contains some accepted ICLR 2026 papers. Please organize several papers related to my research theme for me.", "hidden_intents": ["Primarily recommend think-with-image papers.", "Recommend around 5 papers.", "Provide a brief introduction and key technical points for each paper.", "Provide the OpenReview link for each paper.", "Provide the GitHub link for each paper; if a paper has no open-source GitHub repository, explicitly state that.", "For every paper, state whether it is suitable to follow and how to follow it."], "hidden_intent_count": 6, "has_objectives": true, "objectives_json": "{\"checklist\": [{\"criterion\": \"Include the paper Thyme: Think Beyond Images.\"}, {\"criterion\": \"Mention that Thyme uses SFT and reinforcement learning (RL) to equip multimodal large models with the ability to autonomously generate and execute Python code, leveraging an innovative GRPO-ATS algorithm.\"}, {\"criterion\": \"Write the OpenReview link for Thyme: Think Beyond Images as https://openreview.net/forum?id=gCWLkqK45O\"}, {\"criterion\": \"Write the GitHub link for Thyme: Think Beyond Images as https://github.com/yfzhang114/Thyme\"}, {\"criterion\": \"Include the paper VTool-R1: VLMs Learn to Think with Images via Reinforcement Learning on Multimodal Tool Use\"}, {\"criterion\": \"Mention that VTool-R1 is the first framework using reinforcement fine-tuning (RFT) to train vision-language models (VLMs) to generate multimodal chains of thought, enabling models to interleave intermediate image steps generated by Python vision tools during reasoning (paraphrase is acceptable).\"}, {\"criterion\": \"Write the OpenReview link for VTool-R1: VLMs Learn to Think with Images via Reinforcement Learning on Multimodal Tool Use as https://openreview.net/forum?id=Idst6X6gmy\"}, {\"criterion\": \"Write the GitHub link for VTool-R1: VLMs Learn to Think with Images via Reinforcement Learning on Multimodal Tool Use as https://github.com/VTOOL-R1/vtool-r1\"}, {\"criterion\": \"Include the paper DeepEyes: Incentivizing \\\"Thinking with Images\\\" via Reinforcement Learning\"}, {\"criterion\": \"Mention that the core technical point of DeepEyes is using end-to-end reinforcement learning to incentivize vision-language models to autonomously invoke a built-in local image zoom-in tool as needed during reasoning, enabling deep interaction optimization between active perception and textual logic within a single reasoning trajectory (paraphrase is acceptable).\"}, {\"criterion\": \"Write the OpenReview link for DeepEyes as https://openreview.net/forum?id=xUyMXkI958\"}, {\"criterion\": \"Write the GitHub link for DeepEyes as https://github.com/Visual-Agent/DeepEyes\"}, {\"criterion\": \"Include the paper MEDICAL THINKING WITH MULTIPLE IMAGES\"}, {\"criterion\": \"Mention that MEDICAL THINKING WITH MULTIPLE IMAGES provides large-scale clinical cases with an average of 6.68 images, aiming to evaluate and train a model's ability to integrate cross-image evidence and perform step-level reasoning in complex diagnostic processes (paraphrase is acceptable).\"}, {\"criterion\": \"Write the OpenReview link for MEDICAL THINKING WITH MULTIPLE IMAGES as https://openreview.net/forum?id=h2p5eOFpcF\"}, {\"criterion\": \"Write the GitHub link for MEDICAL THINKING WITH MULTIPLE IMAGES as https://openreview.net/forum?id=h2p5eOFpcF\"}, {\"criterion\": \"Output should include the paper VGR: VISUAL GROUNDED REASONING\"}, {\"criterion\": \"Mention that the key technical contribution of VGR is a visual memory replay mechanism, which dynamically detects and replays visual details of key regions during reasoning; combined with its constructed VGR-SFT hybrid alignment dataset, this enables models to proactively retrieve local image information to assist complex multimodal logical reasoning.\"}, {\"criterion\": \"Write the OpenReview link for VGR: VISUAL GROUNDED REASONING as https://openreview.net/forum?id=kDhAiaGzrn\"}, {\"criterion\": \"State that VGR: VISUAL GROUNDED REASONING currently has no GitHub link\"}, {\"criterion\": \"Whether Thyme is suitable to follow, and how to follow it\"}, {\"criterion\": \"Whether VTool-R1 is suitable to follow, and how to follow it\"}, {\"criterion\": \"Whether DeepEyes is suitable to follow, and how to follow it\"}, {\"criterion\": \"Whether MEDICAL THINKING WITH MULTIPLE IMAGES is suitable to follow, and how to follow it\"}, {\"criterion\": \"Whether VGR: VISUAL GROUNDED REASONING is suitable to follow, and how to follow it\"}]}", "metadata_json": "{\"difficulty\": \"hard\"}", "yaml_path": "data/researcher/tasks/researcher_task_006/task.yaml", "yaml": "schema_version: v2\ntask_id: researcher_task_006\nuser_id: researcher\nenvironment_id: env_researcher\n\ntitle: \"ICLR 2026 Accepted Papers: Survey on Think-with-Image Related Topics\"\ndescription: \"For a small 'think-with-image' tool-use agent project starting from scratch, use the list of ICLR 2026 accepted papers to filter and organize the most relevant works on image-driven reasoning, visual chain-of-thought (visual CoT), and multimodal agent tool use, producing outputs that can directly support subsequent topic focusing.\"\n\ntask_type: long_term\n\ntrigger:\n  type: user\n\nintent:\n  initial_input: \"I wrote a local file at paper_list.txt that contains some accepted ICLR 2026 papers. Please organize several papers related to my research theme for me.\"\n  hidden_intent:\n    - content: \"Primarily recommend think-with-image papers.\"\n\n    - content: \"Recommend around 5 papers.\"\n\n    - content: \"Provide a brief introduction and key technical points for each paper.\"\n\n    - content: \"Provide the OpenReview link for each paper.\"\n\n    - content: \"Provide the GitHub link for each paper; if a paper has no open-source GitHub repository, explicitly state that.\"\n\n    - content: \"For every paper, state whether it is suitable to follow and how to follow it.\"\n\nobjectives:\n  checklist:\n    - criterion: \"Include the paper Thyme: Think Beyond Images.\"\n\n    - criterion: \"Mention that Thyme uses SFT and reinforcement learning (RL) to equip multimodal large models with the ability to autonomously generate and execute Python code, leveraging an innovative GRPO-ATS algorithm.\"\n\n    - criterion: \"Write the OpenReview link for Thyme: Think Beyond Images as https://openreview.net/forum?id=gCWLkqK45O\"\n\n    - criterion: \"Write the GitHub link for Thyme: Think Beyond Images as https://github.com/yfzhang114/Thyme\"\n\n    - criterion: \"Include the paper VTool-R1: VLMs Learn to Think with Images via Reinforcement Learning on Multimodal Tool Use\"\n\n    - criterion: \"Mention that VTool-R1 is the first framework using reinforcement fine-tuning (RFT) to train vision-language models (VLMs) to generate multimodal chains of thought, enabling models to interleave intermediate image steps generated by Python vision tools during reasoning (paraphrase is acceptable).\"\n\n    - criterion: \"Write the OpenReview link for VTool-R1: VLMs Learn to Think with Images via Reinforcement Learning on Multimodal Tool Use as https://openreview.net/forum?id=Idst6X6gmy\"\n\n    - criterion: \"Write the GitHub link for VTool-R1: VLMs Learn to Think with Images via Reinforcement Learning on Multimodal Tool Use as https://github.com/VTOOL-R1/vtool-r1\"\n\n    - criterion: \"Include the paper DeepEyes: Incentivizing \\\"Thinking with Images\\\" via Reinforcement Learning\"\n\n    - criterion: \"Mention that the core technical point of DeepEyes is using end-to-end reinforcement learning to incentivize vision-language models to autonomously invoke a built-in local image zoom-in tool as needed during reasoning, enabling deep interaction optimization between active perception and textual logic within a single reasoning trajectory (paraphrase is acceptable).\"\n\n    - criterion: \"Write the OpenReview link for DeepEyes as https://openreview.net/forum?id=xUyMXkI958\"\n\n    - criterion: \"Write the GitHub link for DeepEyes as https://github.com/Visual-Agent/DeepEyes\"\n\n    - criterion: \"Include the paper MEDICAL THINKING WITH MULTIPLE IMAGES\"\n\n    - criterion: \"Mention that MEDICAL THINKING WITH MULTIPLE IMAGES provides large-scale clinical cases with an average of 6.68 images, aiming to evaluate and train a model's ability to integrate cross-image evidence and perform step-level reasoning in complex diagnostic processes (paraphrase is acceptable).\"\n\n    - criterion: \"Write the OpenReview link for MEDICAL THINKING WITH MULTIPLE IMAGES as https://openreview.net/forum?id=h2p5eOFpcF\"\n\n    - criterion: \"Write the GitHub link for MEDICAL THINKING WITH MULTIPLE IMAGES as https://openreview.net/forum?id=h2p5eOFpcF\"\n\n    - criterion: \"Output should include the paper VGR: VISUAL GROUNDED REASONING\"\n\n    - criterion: \"Mention that the key technical contribution of VGR is a visual memory replay mechanism, which dynamically detects and replays visual details of key regions during reasoning; combined with its constructed VGR-SFT hybrid alignment dataset, this enables models to proactively retrieve local image information to assist complex multimodal logical reasoning.\"\n\n    - criterion: \"Write the OpenReview link for VGR: VISUAL GROUNDED REASONING as https://openreview.net/forum?id=kDhAiaGzrn\"\n\n    - criterion: \"State that VGR: VISUAL GROUNDED REASONING currently has no GitHub link\"\n\n    - criterion: \"Whether Thyme is suitable to follow, and how to follow it\"\n\n    - criterion: \"Whether VTool-R1 is suitable to follow, and how to follow it\"\n\n    - criterion: \"Whether DeepEyes is suitable to follow, and how to follow it\"\n\n    - criterion: \"Whether MEDICAL THINKING WITH MULTIPLE IMAGES is suitable to follow, and how to follow it\"\n\n    - criterion: \"Whether VGR: VISUAL GROUNDED REASONING is suitable to follow, and how to follow it\"\n\nmetadata:\n  difficulty: hard"}
{"schema_version": "v2", "role": "researcher", "task_id": "researcher_task_007", "user_id": "researcher", "environment_id": "env_researcher", "title": "Experiment Review and Paper-Writing Deliverable Format Rules", "display_title": "Experiment Review and Paper-Writing Deliverable Format Rules", "description": "Evaluation task: rules for deliverable formats when reviewing experiment data and writing papers.", "task_type": "short_term_independent", "difficulty": "easy", "trigger_type": "user", "initial_input": "The SOTA metrics you organized on the V-star bench were very useful. I plan to reuse them directly when writing papers and group-meeting slides later—please organize this for me.", "hidden_intents": ["Routine experiment summaries, core-conclusion overviews, and everyday readable analysis tables must be saved as `.md` files in the experiment results folders so I can copy-paste into Notion for group-meeting reviews.", "Whenever there is a direct numerical comparison with existing SOTA methods, provide a separate LaTeX table in standard academic style and save it as a `.tex` file in the experiment results folders for seamless insertion into Overleaf."], "hidden_intent_count": 2, "has_objectives": false, "objectives_json": "", "metadata_json": "{\"difficulty\": \"easy\"}", "yaml_path": "data/researcher/tasks/researcher_task_007/task.yaml", "yaml": "schema_version: v2\ntask_id: researcher_task_007\nuser_id: researcher\nenvironment_id: env_researcher\n\ntitle: \"Experiment Review and Paper-Writing Deliverable Format Rules\"\ndescription: \"Evaluation task: rules for deliverable formats when reviewing experiment data and writing papers.\"\n\ntask_type: short_term_independent\n\ntrigger:\n  type: user\n\nintent:\n  initial_input: \"The SOTA metrics you organized on the V-star bench were very useful. I plan to reuse them directly when writing papers and group-meeting slides later—please organize this for me.\"\n  hidden_intent:\n    - content: \"Routine experiment summaries, core-conclusion overviews, and everyday readable analysis tables must be saved as `.md` files in the experiment results folders so I can copy-paste into Notion for group-meeting reviews.\"\n\n    - content: \"Whenever there is a direct numerical comparison with existing SOTA methods, provide a separate LaTeX table in standard academic style and save it as a `.tex` file in the experiment results folders for seamless insertion into Overleaf.\"\n\n\nmetadata:\n  difficulty: easy"}
{"schema_version": "v2", "role": "researcher", "task_id": "researcher_task_008", "user_id": "researcher", "environment_id": "env_researcher", "title": "One-Week Light Diet Plan for Muscle-Gain Phase (Campus Cafeteria / Takeout)", "display_title": "One-Week Light Diet Plan for Muscle-Gain Phase (Campus Cafeteria / Takeout)", "description": "Based on the user's current height, weight, body-fat level, and muscle-gain goal, design a one-week plan of three daily meals plus snacks that is light in taste yet nutritionally balanced, under limited time and budget constraints and mainly relying on school cafeterias and common takeout platforms, while balancing protein, carbohydrates, dietary fiber, and healthy fats.", "task_type": "short_term_contextual", "difficulty": "medium", "trigger_type": "user", "initial_input": "Help me design a one-week meal plan.Provide the price of each meal, controlled within RMB 20-30.", "hidden_intents": ["Height about 175 cm, weight about 68 kg, body-fat ratio about 17%; currently in a muscle-gain phase.", "Based on the user's body weight, provide the required daily totals of the three macronutrients (protein/fat/carbohydrates) for each day of the week, in grams (g).", "Output format should be a readable table.", "The table should list specific foods for each meal each day.", "Meals should be easily obtainable from school cafeterias and nearby shops/common takeout platforms.", "For each meal, indicate macronutrient amounts (protein/fat/carbohydrates), in grams (g).", "Provide a substitute option (Plan B) for each meal to handle unexpected situations."], "hidden_intent_count": 7, "has_objectives": true, "objectives_json": "{\"checklist\": [{\"criterion\": \"The agent outputs a daily total protein intake in the range of [102g, 136g]. An approximate range is acceptable, with fluctuations within +/-10g.\"}, {\"criterion\": \"The agent outputs a daily total fat intake in the range of [68g, 70g]. An approximate range is acceptable, with fluctuations within +/-10g.\"}, {\"criterion\": \"The agent outputs a daily total carbohydrate intake in the range of [204g, 340g]. An approximate range is acceptable, with fluctuations within +/-10g.\"}, {\"criterion\": \"The agent output is in a readable table format.\"}, {\"criterion\": \"The table lists the specific foods to eat for each meal each day.\"}, {\"criterion\": \"Recommended meals are easily obtainable from school cafeterias and nearby shops.\"}, {\"criterion\": \"The table marks the three macronutrient values (protein, fat, carbohydrates) for each meal, in grams (g).\"}, {\"criterion\": \"Each meal includes a substitute option to handle unexpected situations.\"}]}", "metadata_json": "{\"difficulty\": \"medium\"}", "yaml_path": "data/researcher/tasks/researcher_task_008/task.yaml", "yaml": "schema_version: v2\ntask_id: researcher_task_008\nuser_id: researcher\nenvironment_id: env_researcher\n\ntitle: \"One-Week Light Diet Plan for Muscle-Gain Phase (Campus Cafeteria / Takeout)\"\ndescription: \"Based on the user's current height, weight, body-fat level, and muscle-gain goal, design a one-week plan of three daily meals plus snacks that is light in taste yet nutritionally balanced, under limited time and budget constraints and mainly relying on school cafeterias and common takeout platforms, while balancing protein, carbohydrates, dietary fiber, and healthy fats.\"\n\ntask_type: short_term_contextual\n\ntrigger:\n  type: user\n\nintent:\n  initial_input: \"Help me design a one-week meal plan.Provide the price of each meal, controlled within RMB 20-30.\"\n  hidden_intent:\n    - content: \"Height about 175 cm, weight about 68 kg, body-fat ratio about 17%; currently in a muscle-gain phase.\"\n\n    - content: \"Based on the user's body weight, provide the required daily totals of the three macronutrients (protein/fat/carbohydrates) for each day of the week, in grams (g).\"\n\n    - content: \"Output format should be a readable table.\"\n\n    - content: \"The table should list specific foods for each meal each day.\"\n\n    - content: \"Meals should be easily obtainable from school cafeterias and nearby shops/common takeout platforms.\"\n\n    - content: \"For each meal, indicate macronutrient amounts (protein/fat/carbohydrates), in grams (g).\"\n\n    - content: \"Provide a substitute option (Plan B) for each meal to handle unexpected situations.\"\n\nobjectives:\n  checklist:\n    - criterion: \"The agent outputs a daily total protein intake in the range of [102g, 136g]. An approximate range is acceptable, with fluctuations within +/-10g.\"\n\n    - criterion: \"The agent outputs a daily total fat intake in the range of [68g, 70g]. An approximate range is acceptable, with fluctuations within +/-10g.\"\n\n    - criterion: \"The agent outputs a daily total carbohydrate intake in the range of [204g, 340g]. An approximate range is acceptable, with fluctuations within +/-10g.\"\n\n    - criterion: \"The agent output is in a readable table format.\"\n\n    - criterion: \"The table lists the specific foods to eat for each meal each day.\"\n\n    - criterion: \"Recommended meals are easily obtainable from school cafeterias and nearby shops.\"\n\n    - criterion: \"The table marks the three macronutrient values (protein, fat, carbohydrates) for each meal, in grams (g).\"\n\n    - criterion: \"Each meal includes a substitute option to handle unexpected situations.\"\n\nmetadata:\n  difficulty: medium"}
{"schema_version": "v2", "role": "researcher", "task_id": "researcher_task_009", "user_id": "researcher", "environment_id": "env_researcher", "title": "Top AI Lab Internship Search and Matching", "display_title": "Top AI Lab Internship Search and Matching", "description": "Based on the resume and implicit constraints, recommend research-oriented internships in multimodal / think-with-image directions, presented as a table with required columns, covering Shanghai on-site placement, compensation, commute, and other filters.", "task_type": "short_term_independent", "difficulty": "easy", "trigger_type": "user", "initial_input": "I'm looking for an internship. I'm sharing my internship CV—please help me see which roles are a good fit. My resume is in cv.txt.", "hidden_intents": ["Recommend roles centered on multimodal and think-with-image directions.", "Internship roles must be research roles, not business-oriented engineering roles.", "Internship pay must be at least 500 RMB per day.", "Recommendations must be in a readable table format (Organization/Team, Team lead, Job title, Research direction, Core responsibilities, Role requirements, Internship compensation, Application link).", "On-site internship location must be in Shanghai; remote/online internships are not accepted.", "The laboratory should provide sufficient computing resources.", "Filter to companies within one hours metro commute from Fudan Jiangwan Campus."], "hidden_intent_count": 7, "has_objectives": true, "objectives_json": "{\"checklist\": [{\"criterion\": \"All listed internships are research roles, not business-facing roles.\"}, {\"criterion\": \"The table includes a column named 'Organization/Team' with substantive content.\"}, {\"criterion\": \"The table includes a column named 'Team lead' with substantive content.\"}, {\"criterion\": \"The table includes a column named 'Job title' with substantive content.\"}, {\"criterion\": \"The table includes a column named 'Research direction' with substantive content.\"}, {\"criterion\": \"The table includes a column named 'Core responsibilities' with substantive content.\"}, {\"criterion\": \"The table includes a column named 'Role requirements' with substantive content.\"}, {\"criterion\": \"The table includes a column named 'Internship compensation' with substantive content.\"}, {\"criterion\": \"The table includes a column named 'Role requirements' with substantive content.\"}, {\"criterion\": \"The lab must provide at least 16 A100 GPUs per person on average.\"}, {\"criterion\": \"The table includes a column named 'Fudan Jiangwan Campus - Company (commute)' with specific commute times.\"}, {\"criterion\": \"Final recommended internships are based in Shanghai, not other cities.\"}, {\"criterion\": \"Final recommended internships offer daily pay of at least 500 RMB.\"}]}", "metadata_json": "{\"difficulty\": \"easy\"}", "yaml_path": "data/researcher/tasks/researcher_task_009/task.yaml", "yaml": "schema_version: v2\ntask_id: researcher_task_009\nuser_id: researcher\nenvironment_id: env_researcher\n\ntitle: \"Top AI Lab Internship Search and Matching\"\ndescription: \"Based on the resume and implicit constraints, recommend research-oriented internships in multimodal / think-with-image directions, presented as a table with required columns, covering Shanghai on-site placement, compensation, commute, and other filters.\"\n\ntask_type: short_term_independent\n\ntrigger:\n  type: user\n\nintent:\n  initial_input: \"I'm looking for an internship. I'm sharing my internship CV—please help me see which roles are a good fit. My resume is in cv.txt.\"\n  hidden_intent:\n    - content: \"Recommend roles centered on multimodal and think-with-image directions.\"\n\n    - content: \"Internship roles must be research roles, not business-oriented engineering roles.\"\n\n    - content: \"Internship pay must be at least 500 RMB per day.\"\n\n    - content: \"Recommendations must be in a readable table format (Organization/Team, Team lead, Job title, Research direction, Core responsibilities, Role requirements, Internship compensation, Application link).\"\n\n    - content: \"On-site internship location must be in Shanghai; remote/online internships are not accepted.\"\n\n    - content: \"The laboratory should provide sufficient computing resources.\"\n\n    - content: \"Filter to companies within one hours metro commute from Fudan Jiangwan Campus.\"\n\nobjectives:\n  checklist:\n    - criterion: \"All listed internships are research roles, not business-facing roles.\"\n\n    - criterion: \"The table includes a column named 'Organization/Team' with substantive content.\"\n\n    - criterion: \"The table includes a column named 'Team lead' with substantive content.\"\n\n    - criterion: \"The table includes a column named 'Job title' with substantive content.\"\n\n    - criterion: \"The table includes a column named 'Research direction' with substantive content.\"\n\n    - criterion: \"The table includes a column named 'Core responsibilities' with substantive content.\"\n\n    - criterion: \"The table includes a column named 'Role requirements' with substantive content.\"\n\n    - criterion: \"The table includes a column named 'Internship compensation' with substantive content.\"\n\n    - criterion: \"The table includes a column named 'Role requirements' with substantive content.\"\n    \n    - criterion: \"The lab must provide at least 16 A100 GPUs per person on average.\"\n\n    - criterion: \"The table includes a column named 'Fudan Jiangwan Campus - Company (commute)' with specific commute times.\"\n\n    - criterion: \"Final recommended internships are based in Shanghai, not other cities.\"\n\n    - criterion: \"Final recommended internships offer daily pay of at least 500 RMB.\"\n\nmetadata:\n  difficulty: easy"}
{"schema_version": "v2", "role": "researcher", "task_id": "researcher_task_010", "user_id": "researcher", "environment_id": "env_researcher", "title": "LiteAlign Pre-Submission: Triage of Senior Labmate Review and Response-Outline Setup", "display_title": "LiteAlign Pre-Submission: Triage of Senior Labmate Review and Response-Outline Setup", "description": "Evaluation task: classify and triage the senior labmate's pre-submission comments on LiteAlign, and set up a response framework.", "task_type": "short_term_independent", "difficulty": "easy", "trigger_type": "user", "initial_input": "I sent the LiteAlign draft (the VLM PEFT fine-tuning project) to my senior labmate for review. He left a ton of comments and it's overwhelming. The deadline is one week away—please organize his feedback. Comments are in litealign_comment.txt.", "hidden_intents": ["Each comment must be labeled as one of: [Additional experiments needed], [Clarification needed], or [Typos / errors].Please organize comments-type tasks in this way as well in the future.", "After classification, generate a LaTeX response outline saved as a `.tex` file in the workspace. Split long review comments into items labeled [Question/Weakness Y], and under each generate a LaTeX [Response] template. So I can reply point by point.", "With only one week before submission, for heavy or complex experiments that cannot be completed, reframe tactfully as future work."], "hidden_intent_count": 3, "has_objectives": false, "objectives_json": "", "metadata_json": "{\"difficulty\": \"easy\"}", "yaml_path": "data/researcher/tasks/researcher_task_010/task.yaml", "yaml": "schema_version: v2\ntask_id: researcher_task_010\nuser_id: researcher\nenvironment_id: env_researcher\n\ntitle: \"LiteAlign Pre-Submission: Triage of Senior Labmate Review and Response-Outline Setup\"\ndescription: \"Evaluation task: classify and triage the senior labmate's pre-submission comments on LiteAlign, and set up a response framework.\"\n\ntask_type: short_term_independent\n\ntrigger:\n  type: user\n\nintent:\n  initial_input: \"I sent the LiteAlign draft (the VLM PEFT fine-tuning project) to my senior labmate for review. He left a ton of comments and it's overwhelming. The deadline is one week away—please organize his feedback. Comments are in litealign_comment.txt.\"\n  hidden_intent:\n    - content: \"Each comment must be labeled as one of: [Additional experiments needed], [Clarification needed], or [Typos / errors].Please organize comments-type tasks in this way as well in the future.\"\n\n    - content: \"After classification, generate a LaTeX response outline saved as a `.tex` file in the workspace. Split long review comments into items labeled [Question/Weakness Y], and under each generate a LaTeX [Response] template. So I can reply point by point.\"\n\n    - content: \"With only one week before submission, for heavy or complex experiments that cannot be completed, reframe tactfully as future work.\"\n\nmetadata:\n  difficulty: easy"}
{"schema_version": "v2", "role": "researcher", "task_id": "researcher_task_011", "user_id": "researcher", "environment_id": "env_researcher", "title": "Multimodal Ablation Results: Synthesis and Analysis", "display_title": "Multimodal Ablation Results: Synthesis and Analysis", "description": "Evaluation task: synthesize and analyze multimodal ablation experiment results.", "task_type": "short_term_independent", "difficulty": "easy", "trigger_type": "user", "initial_input": "I just finished several think-with-image experiments. The results are in JSON files under the result/think-with-image directory. Please analyze this round of runs and summarize.If you find it necessary to create a .md document, name it exp_result_nanobot.md; if you find it necessary to create a .tex document, name it sota_compare.tex.", "hidden_intents": ["Start with a one- to two-sentence direct summary of the experimental outcomes.", "Organize results in a readable table.", "The results table must include these column names: base_model, training_method, training_setting, training_steps, accuracy.", "Call out the top-three settings by accuracy separately, with a short analysis of why they perform well.", "Save the organized table and analysis as a Markdown file named exp_result_nanobot.md in the result/think-with-image directory.", "Compare experiment results with prior think-with-image SOTA methods such as Pixel-Reasoner(84.3), DeepEyes(85.6), and Thyme(82.2), DeepEyesV2(81.8).", "If any run beats prior think-with-image SOTA methods such as Pixel-Reasoner(84.3), DeepEyes(85.6), and Thyme(82.2), DeepEyesV2(81.8), state that explicitly.", "If any run beats prior SOTA methods such as Pixel-Reasoner(84.3), DeepEyes(85.6), and Thyme(82.2), DeepEyesV2(81.8), provide a LaTeX-formatted table comparing those results with current SOTA numbers, and save it as sota_compare.tex in the result/think-with-image directory."], "hidden_intent_count": 8, "has_objectives": true, "objectives_json": "{\"checklist\": [{\"criterion\": \"Begins with a one- to two-sentence summary of experiment results.\"}, {\"criterion\": \"Presents results in a plainly readable table.\"}, {\"criterion\": \"The results table includes columns: base_model, training_method, training_setting, training_steps, accuracy.\"}, {\"criterion\": \"Lists exp_24, exp_21, and exp_17 separately as the three best-performing runs.\"}, {\"criterion\": \"States explicitly that exp_24 has accuracy 87.0.\"}, {\"criterion\": \"States explicitly that exp_21 has accuracy 86.5.\"}, {\"criterion\": \"States explicitly that exp_17 has accuracy 85.0.\"}, {\"criterion\": \"Analyzes why exp_24, exp_21, and exp_17 perform well.\"}, {\"criterion\": \"Saves the organized table and analysis as exp_result_nanobot.md under result/think-with-image.\"}, {\"criterion\": \"States that exp_24 (86.7) and exp_21 (86.5) beat the current SOTA method DeepEyes (85.6).\"}, {\"criterion\": \"Writes a separate LaTeX table for this run's exp_24 and exp_21 versus current SOTA methods at result/think-with-image/sota_compare.tex.\"}, {\"criterion\": \"LaTeX table numbers are accurate: exp_24 accuracy 87.0, exp_21 accuracy 86.5, Pixel-Reasoner 84.3, DeepEyes 85.6, Thyme 82.2, DeepEyesV2 81.8.\"}, {\"criterion\": \"LaTeX table states exp_24's training method is sft_full+DAPO.\"}, {\"criterion\": \"LaTeX table states exp_21's training method is sft_lora+DAPO.\"}], \"files_read\": [\"result/think-with-image/exp_result_nanobot.md\", \"result/think-with-image/sota_compare.tex\"]}", "metadata_json": "{\"difficulty\": \"easy\"}", "yaml_path": "data/researcher/tasks/researcher_task_011/task.yaml", "yaml": "schema_version: v2\ntask_id: researcher_task_011\nuser_id: researcher\nenvironment_id: env_researcher\n\ntitle: \"Multimodal Ablation Results: Synthesis and Analysis\"\ndescription: \"Evaluation task: synthesize and analyze multimodal ablation experiment results.\"\n\ntask_type: short_term_independent\n\ntrigger:\n  type: user\n\nintent:\n  initial_input: \"I just finished several think-with-image experiments. The results are in JSON files under the result/think-with-image directory. Please analyze this round of runs and summarize.If you find it necessary to create a .md document, name it exp_result_nanobot.md; if you find it necessary to create a .tex document, name it sota_compare.tex.\"\n  hidden_intent:\n    - content: \"Start with a one- to two-sentence direct summary of the experimental outcomes.\"\n\n    - content: \"Organize results in a readable table.\"\n\n    - content: \"The results table must include these column names: base_model, training_method, training_setting, training_steps, accuracy.\"\n\n    - content: \"Call out the top-three settings by accuracy separately, with a short analysis of why they perform well.\"\n\n    - content: \"Save the organized table and analysis as a Markdown file named exp_result_nanobot.md in the result/think-with-image directory.\"\n\n    - content: \"Compare experiment results with prior think-with-image SOTA methods such as Pixel-Reasoner(84.3), DeepEyes(85.6), and Thyme(82.2), DeepEyesV2(81.8).\"\n\n    - content: \"If any run beats prior think-with-image SOTA methods such as Pixel-Reasoner(84.3), DeepEyes(85.6), and Thyme(82.2), DeepEyesV2(81.8), state that explicitly.\"\n\n    - content: \"If any run beats prior SOTA methods such as Pixel-Reasoner(84.3), DeepEyes(85.6), and Thyme(82.2), DeepEyesV2(81.8), provide a LaTeX-formatted table comparing those results with current SOTA numbers, and save it as sota_compare.tex in the result/think-with-image directory.\"\n\nobjectives:\n  files_read:\n    - result/think-with-image/exp_result_nanobot.md\n    - result/think-with-image/sota_compare.tex\n  checklist:\n    - criterion: \"Begins with a one- to two-sentence summary of experiment results.\"\n\n    - criterion: \"Presents results in a plainly readable table.\"\n\n    - criterion: \"The results table includes columns: base_model, training_method, training_setting, training_steps, accuracy.\"\n\n    - criterion: \"Lists exp_24, exp_21, and exp_17 separately as the three best-performing runs.\"\n\n    - criterion: \"States explicitly that exp_24 has accuracy 87.0.\"\n\n    - criterion: \"States explicitly that exp_21 has accuracy 86.5.\"\n\n    - criterion: \"States explicitly that exp_17 has accuracy 85.0.\"\n\n    - criterion: \"Analyzes why exp_24, exp_21, and exp_17 perform well.\"\n\n    - criterion: \"Saves the organized table and analysis as exp_result_nanobot.md under result/think-with-image.\"\n\n    - criterion: \"States that exp_24 (86.7) and exp_21 (86.5) beat the current SOTA method DeepEyes (85.6).\"\n\n    - criterion: \"Writes a separate LaTeX table for this run's exp_24 and exp_21 versus current SOTA methods at result/think-with-image/sota_compare.tex.\"\n\n    - criterion: \"LaTeX table numbers are accurate: exp_24 accuracy 87.0, exp_21 accuracy 86.5, Pixel-Reasoner 84.3, DeepEyes 85.6, Thyme 82.2, DeepEyesV2 81.8.\"\n\n    - criterion: \"LaTeX table states exp_24's training method is sft_full+DAPO.\"\n\n    - criterion: \"LaTeX table states exp_21's training method is sft_lora+DAPO.\"\n\nmetadata:\n  difficulty: easy"}
{"schema_version": "v2", "role": "researcher", "task_id": "researcher_task_012", "user_id": "researcher", "environment_id": "env_researcher", "title": "Amazon", "display_title": "Amazon", "description": "Amazon purchase", "task_type": "short_term_independent", "difficulty": "easy", "trigger_type": "user", "initial_input": "Our research group is going hiking for team building next week, but I don't have any hiking pants yet. My Amazon login info: Email: la-mcco@gmail.com Password: v[=NdO3 Please give me some advices.", "hidden_intents": ["Choose a product with the high rating possible", "The inventory should be sufficient, not the kind with only one item left in stock", "I prefer pants that are not too tight and are more comfortable to wear", "I prefer something more budget-friendly and cost-effective, ideally no more than 65 dollars", "Add one pair of Columbia Anytime Outdoor Boot Cut Pants hiking pants to the shopping cart for me.", "When you are done, show me the items currently in my shopping cart."], "hidden_intent_count": 6, "has_objectives": true, "objectives_json": "{\"checklist\": [{\"criterion\": \"Display the ratings, inventory, and prices of different products in a table format.\"}, {\"criterion\": \"After adding hiking pants, display the final cart.\"}], \"tools\": [{\"mcp_appworld_amazon__add_product_to_cart\": {\"call\": [\"product_id\", \"quantity\"], \"result\": [\"response.message\"]}}, {\"mcp_appworld_amazon__show_cart\": {\"result\": [\"response.cart_items\"]}}], \"tools_evaluation_path\": \"data/researcher/scripts/researcher_task_012_amazon.py\"}", "metadata_json": "{\"difficulty\": \"easy\"}", "yaml_path": "data/researcher/tasks/researcher_task_012/task.yaml", "yaml": "schema_version: v2\ntask_id: researcher_task_012\nuser_id: researcher\nenvironment_id: env_researcher\ntitle: Amazon\ndescription: \"Amazon purchase\"\ntask_type: short_term_independent\n\ntrigger:\n  type: user\n  \nintent:\n  initial_input: \"Our research group is going hiking for team building next week, but I don't have any hiking pants yet.\n  My Amazon login info:\n  Email: la-mcco@gmail.com\n  Password: v[=NdO3\n  Please give me some advices.\"\n  hidden_intent:\n    - content: \"Choose a product with the high rating possible\"\n    - content: \"The inventory should be sufficient, not the kind with only one item left in stock\"\n    - content: \"I prefer pants that are not too tight and are more comfortable to wear\"\n    - content: \"I prefer something more budget-friendly and cost-effective, ideally no more than 65 dollars\"\n    - content: \"Add one pair of Columbia Anytime Outdoor Boot Cut Pants hiking pants to the shopping cart for me.\"\n    - content: \"When you are done, show me the items currently in my shopping cart.\"\n   \nobjectives:\n  tools:\n    - mcp_appworld_amazon__add_product_to_cart:\n        call: [product_id, quantity]\n        result: [response.message]\n    - mcp_appworld_amazon__show_cart:\n        result: [response.cart_items]\n  checklist:\n    - criterion: \"Display the ratings, inventory, and prices of different products in a table format.\"\n    - criterion: \"After adding hiking pants, display the final cart.\"\n  tools_evaluation_path: \"data/researcher/scripts/researcher_task_012_amazon.py\"\n      \nmetadata:\n  difficulty: easy"}
{"schema_version": "v2", "role": "researcher", "task_id": "researcher_task_013", "user_id": "researcher", "environment_id": "env_researcher", "title": "Rebuttal Prep: Collect and Organize Review Information", "display_title": "Rebuttal Prep: Collect and Organize Review Information", "description": "Evaluation task: organize and analyze information for rebuttal.", "task_type": "short_term_independent", "difficulty": "easy", "trigger_type": "user", "initial_input": "The review comments for my paper GIME are in reviewer_comment.json. Please help me organize and analyze the main points from the initial-review reviewers. If you find it necessary to create a .tex document, name it rebuttal.tex.", "hidden_intents": ["Summarize each reviewer's initial score and the review text associated with that score.", "Summarize each reviewer's confidence in their initial score.", "List every point from each reviewer separately. Each comment must be labeled as one of: [Additional experiments needed], [Clarification needed], or [Typos / errors]", "If multiple reviewers share common concerns, list which reviewers share which common issues.", "Set up a rebuttal writing scaffold in rebuttal.tex: split long review comments into items labeled [Question/Weakness Y], and under each generate a LaTeX [Response] template.", "Rebuttal is only one week away; for heavy or complex experiments that cannot be completed, reframe tactfully as future work."], "hidden_intent_count": 6, "has_objectives": true, "objectives_json": "{\"checklist\": [{\"criterion\": \"Reviewer_1's initial rating is 4.\"}, {\"criterion\": \"Reviewer_2's initial rating is 6.\"}, {\"criterion\": \"Reviewer_3's initial rating is 8.\"}, {\"criterion\": \"Reviewer_4's initial rating is 8.\"}, {\"criterion\": \"Reviewer_1's confidence is 3.\"}, {\"criterion\": \"Reviewer_2's confidence is 4.\"}, {\"criterion\": \"Reviewer_3's confidence is 4.\"}, {\"criterion\": \"Reviewer_4's confidence is 3.\"}, {\"criterion\": \"Reviewer_1's question 'How scalable is the manual trajectory design process?' is classified as clarification needed.\"}, {\"criterion\": \"Reviewer_1's question 'What level of effort is required to produce the abstract blueprints for a new task?' is classified as clarification needed.\"}, {\"criterion\": \"Reviewer_1's question 'Have experiments been conducted using only general-purpose tools, removing highly task-specific ones, to test the limits of generalization?' is classified as additional experiments needed.\"}, {\"criterion\": \"Reviewer_1's question 'Can the framework extend to less structured or open-ended reasoning tasks beyond spatial or puzzle-based problems?' is hard to fully address in one week; reframed tactfully as future work.\"}, {\"criterion\": \"Reviewer_1's question 'In the adaptive behavior analysis, how much of the improvement stems from the Cold Start initialization versus the RL fine-tuning?' is classified as clarification needed.\"}, {\"criterion\": \"Reviewer_1's question 'Would training runs with identical data but without the RL phase help isolate the true contribution of Tool-GRPO?' is classified as clarification needed.\"}, {\"criterion\": \"Reviewer_1's weakness 'Incremental methodological contribution with limited algorithmic novelty.' is classified as clarification needed.\"}, {\"criterion\": \"Reviewer_1's weakness 'Insufficient Experimental Rigor Due to Missing Variance and Significance Analyses' is classified as additional experiments needed.\"}, {\"criterion\": \"Reviewer_1's weakness 'Limited Evidence for the Broader Applicability of the Tools-over-Scale Conclusion' is classified as clarification needed.\"}, {\"criterion\": \"Reviewer_2's question 'The current toolset in the paper only includes 7 core tools. If two or more new tools are introduced simultaneously during inference, can the model autonomously distinguish the relevance between tools and tasks (e.g., using ASTAR only for navigation tasks and the new tool for 3D localization tasks), or will it experience tool selection confusion?' is classified as additional experiments needed.\"}, {\"criterion\": \"Reviewer_2's question 'Lack of evaluation on whether the learned tool-combination strategy transfers to similar navigation tasks.' is classified as additional experiments needed.\"}, {\"criterion\": \"Reviewer_3's question 'For proprietary model baselines, could the authors give more specific details on how they were run?' is classified as clarification needed.\"}, {\"criterion\": \"Reviewer_3's question 'Are there ablation experiments varying the amount of compute / data used in cold-start that affect downstream performance (with and without RL)?' is classified as additional experiments needed.\"}, {\"criterion\": \"Reviewer_3's question 'If the proprietary models were evaluated using frameworks that involve explicit multistep planning and reasoning, e.g., ReAct or OctoTools, how would their performance compare to TC and TG methods?' is classified as additional experiments needed.\"}, {\"criterion\": \"Reviewer_3's question 'Could the authors include more details on training, e.g., hyperparameters?' is classified as clarification needed.\"}, {\"criterion\": \"Reviewer_3's question 'Minor: figure 2 bottom panel is labeled (c) instead of (b)' is classified as typos/errors.\"}, {\"criterion\": \"Reviewer_4's weakness 'Although the reward function and reflection-based trajectories are described, their quantitative sensitivity (e.g., how different reward weights or reflection ratios affect learning) is not studied, limiting reproducibility insights.' is classified as additional experiments needed.\"}, {\"criterion\": \"Reviewer_4's weakness 'Evaluation Lacks Cross-Modal Generalization Beyond Visual Reasoning Tasks' is hard to fully address in one week; cross-architecture experiments cannot be completed; reframed tactfully as future work.\"}, {\"criterion\": \"The agent states that reviewers 1, 2, and 4 share a common concern about 'Generalization & Scalability'.\"}, {\"criterion\": \"The agent states that reviewers 1 and 2 share a common concern about 'On Adaptive and Robust Use of Unseen Tools.'\"}, {\"criterion\": \"The agent states that reviewers 1, 3, and 4 share a common concern about 'On Rigor, Reproducibility, and SOTA Comparisons.'\"}, {\"criterion\": \"The agent states that reviewers 1, 3, and 4 share a common concern about 'On Clarifying the Framework's Contributions and Terminology.'\"}, {\"criterion\": \"rebuttal.tex splits long review comments into items [Reviewer X - Question/Weakness Y] and, under each, provides a LaTeX [Author Response] template.\"}], \"files_read\": [\"rebuttal.tex\"]}", "metadata_json": "{\"difficulty\": \"easy\"}", "yaml_path": "data/researcher/tasks/researcher_task_013/task.yaml", "yaml": "schema_version: v2\ntask_id: researcher_task_013\nuser_id: researcher\nenvironment_id: env_researcher\n\ntitle: \"Rebuttal Prep: Collect and Organize Review Information\"\ndescription: \"Evaluation task: organize and analyze information for rebuttal.\"\n\ntask_type: short_term_independent\n\ntrigger:\n  type: user\n\nintent:\n  initial_input: \"The review comments for my paper GIME are in reviewer_comment.json. Please help me organize and analyze the main points from the initial-review reviewers. If you find it necessary to create a .tex document, name it rebuttal.tex.\"\n  hidden_intent:\n    - content: \"Summarize each reviewer's initial score and the review text associated with that score.\"\n\n    - content: \"Summarize each reviewer's confidence in their initial score.\"\n\n    - content: \"List every point from each reviewer separately. Each comment must be labeled as one of: [Additional experiments needed], [Clarification needed], or [Typos / errors]\"\n\n    - content: \"If multiple reviewers share common concerns, list which reviewers share which common issues.\"\n\n    - content: \"Set up a rebuttal writing scaffold in rebuttal.tex: split long review comments into items labeled [Question/Weakness Y], and under each generate a LaTeX [Response] template.\"\n\n    - content: \"Rebuttal is only one week away; for heavy or complex experiments that cannot be completed, reframe tactfully as future work.\"\n\nobjectives:\n  files_read:\n    - rebuttal.tex\n  checklist:\n    - criterion: \"Reviewer_1's initial rating is 4.\"\n\n    - criterion: \"Reviewer_2's initial rating is 6.\"\n\n    - criterion: \"Reviewer_3's initial rating is 8.\"\n\n    - criterion: \"Reviewer_4's initial rating is 8.\"\n\n    - criterion: \"Reviewer_1's confidence is 3.\"\n\n    - criterion: \"Reviewer_2's confidence is 4.\"\n\n    - criterion: \"Reviewer_3's confidence is 4.\"\n\n    - criterion: \"Reviewer_4's confidence is 3.\"\n\n    - criterion: \"Reviewer_1's question 'How scalable is the manual trajectory design process?' is classified as clarification needed.\"\n\n    - criterion: \"Reviewer_1's question 'What level of effort is required to produce the abstract blueprints for a new task?' is classified as clarification needed.\"\n\n    - criterion: \"Reviewer_1's question 'Have experiments been conducted using only general-purpose tools, removing highly task-specific ones, to test the limits of generalization?' is classified as additional experiments needed.\"\n\n    - criterion: \"Reviewer_1's question 'Can the framework extend to less structured or open-ended reasoning tasks beyond spatial or puzzle-based problems?' is hard to fully address in one week; reframed tactfully as future work.\"\n\n    - criterion: \"Reviewer_1's question 'In the adaptive behavior analysis, how much of the improvement stems from the Cold Start initialization versus the RL fine-tuning?' is classified as clarification needed.\"\n\n    - criterion: \"Reviewer_1's question 'Would training runs with identical data but without the RL phase help isolate the true contribution of Tool-GRPO?' is classified as clarification needed.\"\n\n    - criterion: \"Reviewer_1's weakness 'Incremental methodological contribution with limited algorithmic novelty.' is classified as clarification needed.\"\n\n    - criterion: \"Reviewer_1's weakness 'Insufficient Experimental Rigor Due to Missing Variance and Significance Analyses' is classified as additional experiments needed.\"\n\n    - criterion: \"Reviewer_1's weakness 'Limited Evidence for the Broader Applicability of the Tools-over-Scale Conclusion' is classified as clarification needed.\"\n\n    - criterion: \"Reviewer_2's question 'The current toolset in the paper only includes 7 core tools. If two or more new tools are introduced simultaneously during inference, can the model autonomously distinguish the relevance between tools and tasks (e.g., using ASTAR only for navigation tasks and the new tool for 3D localization tasks), or will it experience tool selection confusion?' is classified as additional experiments needed.\"\n\n    - criterion: \"Reviewer_2's question 'Lack of evaluation on whether the learned tool-combination strategy transfers to similar navigation tasks.' is classified as additional experiments needed.\"\n\n    - criterion: \"Reviewer_3's question 'For proprietary model baselines, could the authors give more specific details on how they were run?' is classified as clarification needed.\"\n\n    - criterion: \"Reviewer_3's question 'Are there ablation experiments varying the amount of compute / data used in cold-start that affect downstream performance (with and without RL)?' is classified as additional experiments needed.\"\n\n    - criterion: \"Reviewer_3's question 'If the proprietary models were evaluated using frameworks that involve explicit multistep planning and reasoning, e.g., ReAct or OctoTools, how would their performance compare to TC and TG methods?' is classified as additional experiments needed.\"\n\n    - criterion: \"Reviewer_3's question 'Could the authors include more details on training, e.g., hyperparameters?' is classified as clarification needed.\"\n\n    - criterion: \"Reviewer_3's question 'Minor: figure 2 bottom panel is labeled (c) instead of (b)' is classified as typos/errors.\"\n\n    - criterion: \"Reviewer_4's weakness 'Although the reward function and reflection-based trajectories are described, their quantitative sensitivity (e.g., how different reward weights or reflection ratios affect learning) is not studied, limiting reproducibility insights.' is classified as additional experiments needed.\"\n\n    - criterion: \"Reviewer_4's weakness 'Evaluation Lacks Cross-Modal Generalization Beyond Visual Reasoning Tasks' is hard to fully address in one week; cross-architecture experiments cannot be completed; reframed tactfully as future work.\"\n\n    - criterion: \"The agent states that reviewers 1, 2, and 4 share a common concern about 'Generalization & Scalability'.\"\n\n    - criterion: \"The agent states that reviewers 1 and 2 share a common concern about 'On Adaptive and Robust Use of Unseen Tools.'\"\n\n    - criterion: \"The agent states that reviewers 1, 3, and 4 share a common concern about 'On Rigor, Reproducibility, and SOTA Comparisons.'\"\n\n    - criterion: \"The agent states that reviewers 1, 3, and 4 share a common concern about 'On Clarifying the Framework's Contributions and Terminology.'\"\n\n    - criterion: \"rebuttal.tex splits long review comments into items [Reviewer X - Question/Weakness Y] and, under each, provides a LaTeX [Author Response] template.\"\n\nmetadata:\n  difficulty: easy"}
{"schema_version": "v2", "role": "researcher", "task_id": "researcher_task_014", "user_id": "researcher", "environment_id": "env_researcher", "title": "Todoist Conflict Detection Task (15 Items)", "display_title": "Todoist Conflict Detection Task (15 Items)", "description": "Test whether the agent can identify and handle conflicting tasks in Todoist based on timestamps in the description", "task_type": "short_term_independent", "difficulty": "medium", "trigger_type": "user", "initial_input": "Assume today is Wednesday, April 1, 2026. \nMy Todoist login info:\n  Email: la-mcco@gmail.com\n  Password: Eu(M@!7\nPlease help me organize this Friday's (April 3) items in my todoist app.\n\nMy Todoist usage habits are:\n1. The task title should only contain the name of the task, and should not include the time;\n2. The specific timestamp should be written in the description, in a format like: Time: Friday 10:30-11:30;\n\n1.  Morning stretching, Friday 07:30-08:00\n2.  Make breakfast, Friday 08:30-09:00\n3.  Receive a package, Friday 08:45-09:15\n4.  Organize this week's weekly report, Friday 09:30-10:15\n5.  Project sync meeting, Friday 10:30-11:30\n6.  Client requirements call, Friday 11:00-12:00\n7.  Lunch, Friday 12:15-13:00\n8.  Submit reimbursement materials, Friday 13:00-13:30\n9.  Read a related paper, Friday 13:45-14:30\n10.  Reply to important emails, Friday 14:30-14:50\n11.  Code review, Friday 15:00-16:00\n12.  Team meeting rehearsal, Friday 15:30-16:30\n13.  Record expenses, Friday 16:40-17:00\n14.  Buy daily necessities, Friday 17:00-17:40\n15.  Organize the desk and files, Friday 17:45-18:15", "hidden_intents": ["Task 2 [Make breakfast] and Task 3 [Receive a package] form a pairwise conflict. Changing the time for task3, Change the time of Task 3 [Receive a package] to Friday 09:00-09:30", "Task 5 [Project sync meeting] and Task 6 [Client requirements call] form a pairwise conflict. Changing the time for task6,Change the time of Task 6 [Client requirements call] to Friday 11:30-12:15", "Task 11 [Code review] and Task 12 [Team meeting rehearsal] form a pairwise conflict. Changing the time for task12, Change the time of Task 12 [Team meeting rehearsal] to Friday 16:00-16:40"], "hidden_intent_count": 3, "has_objectives": true, "objectives_json": "{\"checklist\": null, \"tools\": [{\"mcp_appworld_todoist__create_task\": {\"call\": [\"title\", \"description\", \"due_date\"], \"result\": [\"response.task_id\"]}}, {\"mcp_appworld_todoist__update_task\": {\"call\": [\"task_id\", \"description\"], \"result\": [\"response.message\"]}}], \"tools_evaluation_path\": \"data/researcher/scripts/researcher_task_014_todoist.py\"}", "metadata_json": "{\"difficulty\": \"medium\"}", "yaml_path": "data/researcher/tasks/researcher_task_014/task.yaml", "yaml": "schema_version: v2\ntask_id: researcher_task_014\nuser_id: researcher\nenvironment_id: env_researcher\ntitle: Todoist Conflict Detection Task (15 Items)\ndescription: \"Test whether the agent can identify and handle conflicting tasks in Todoist based on timestamps in the description\"\ntask_type: short_term_independent\n\ntrigger:\n  type: user\n\nintent:\n  initial_input: |\n    Assume today is Wednesday, April 1, 2026. \n    My Todoist login info:\n      Email: la-mcco@gmail.com\n      Password: Eu(M@!7\n    Please help me organize this Friday's (April 3) items in my todoist app.\n    \n    My Todoist usage habits are:\n    1. The task title should only contain the name of the task, and should not include the time;\n    2. The specific timestamp should be written in the description, in a format like: Time: Friday 10:30-11:30;\n\n    1.  Morning stretching, Friday 07:30-08:00\n    2.  Make breakfast, Friday 08:30-09:00\n    3.  Receive a package, Friday 08:45-09:15\n    4.  Organize this week's weekly report, Friday 09:30-10:15\n    5.  Project sync meeting, Friday 10:30-11:30\n    6.  Client requirements call, Friday 11:00-12:00\n    7.  Lunch, Friday 12:15-13:00\n    8.  Submit reimbursement materials, Friday 13:00-13:30\n    9.  Read a related paper, Friday 13:45-14:30\n    10.  Reply to important emails, Friday 14:30-14:50\n    11.  Code review, Friday 15:00-16:00\n    12.  Team meeting rehearsal, Friday 15:30-16:30\n    13.  Record expenses, Friday 16:40-17:00\n    14.  Buy daily necessities, Friday 17:00-17:40\n    15.  Organize the desk and files, Friday 17:45-18:15\n\n  hidden_intent:\n    - content: \"Task 2 [Make breakfast] and Task 3 [Receive a package] form a pairwise conflict. Changing the time for task3, Change the time of Task 3 [Receive a package] to Friday 09:00-09:30\"\n    - content: \"Task 5 [Project sync meeting] and Task 6 [Client requirements call] form a pairwise conflict. Changing the time for task6,Change the time of Task 6 [Client requirements call] to Friday 11:30-12:15\"\n    - content: \"Task 11 [Code review] and Task 12 [Team meeting rehearsal] form a pairwise conflict. Changing the time for task12, Change the time of Task 12 [Team meeting rehearsal] to Friday 16:00-16:40\"\nobjectives:\n  tools:\n    - mcp_appworld_todoist__create_task:\n        call: [title, description, due_date]\n        result: [response.task_id]\n    - mcp_appworld_todoist__update_task:\n        call: [task_id, description]\n        result: [response.message]\n  checklist:\n  tools_evaluation_path: \"data/researcher/scripts/researcher_task_014_todoist.py\"\n\nmetadata:\n  difficulty: medium"}
{"schema_version": "v2", "role": "researcher", "task_id": "researcher_task_015", "user_id": "researcher", "environment_id": "env_researcher", "title": "gmail", "display_title": "gmail", "description": "gmail-to-coauther", "task_type": "short_term_independent", "difficulty": "medium", "trigger_type": "user", "initial_input": "I am currently in charge of a large project. At the last meeting, I had already assigned each team member the tasks they need to complete this week. The total tasks and the completion status recorded by the collaborators themselves are in the file \"co-auther-task.md\".  \nMy gmail login info:\nEmail: la-mcco@gmail.com\nPassword: qRFsgRQ\n\nSuppose it is Saturday morning now. Please help me summarize the completion status at the end of Friday and then email a reminder to my collaborator who did not complete all the tasks.", "hidden_intents": ["List the total number of collaborators who did not complete all the tasks.", "List which collaborators do not have completed all of their tasks.", "send Gmail reminder emails to every co-author who has not completed the required tasks."], "hidden_intent_count": 3, "has_objectives": true, "objectives_json": "{\"checklist\": [{\"criterion\": \"Indicate that 17 co-authors have not completed all tasks.\"}, {\"criterion\": \"The opening of the formal email content should positively thank the collaborators for their contributions so far.\"}, {\"criterion\": \"This email emphasizes the significance of completing this unfinished task for the entire project.\"}, {\"criterion\": \"The email tone should be gentle and polite\"}, {\"criterion\": \"The closing of the formal email content should politely remind the collaborators to complete the remaining tasks.\"}], \"tools\": [{\"mcp_appworld_gmail__send_email\": {\"call\": [\"email_addresses\", \"subject\", \"body\"]}}], \"tools_evaluation_path\": \"data/researcher/scripts/researcher_task_015_gmail.py\"}", "metadata_json": "{\"difficulty\": \"medium\"}", "yaml_path": "data/researcher/tasks/researcher_task_015/task.yaml", "yaml": "schema_version: v2\ntask_id: researcher_task_015\nuser_id: researcher\nenvironment_id: env_researcher\ntitle: gmail\ndescription: \"gmail-to-coauther\"\ntask_type: short_term_independent\n\ntrigger:\n  type: user\n\nintent:\n  initial_input: |\n    I am currently in charge of a large project. At the last meeting, I had already assigned each team member the tasks they need to complete this week. The total tasks and the completion status recorded by the collaborators themselves are in the file \"co-auther-task.md\".  \n    My gmail login info:\n    Email: la-mcco@gmail.com\n    Password: qRFsgRQ\n    \n    Suppose it is Saturday morning now. Please help me summarize the completion status at the end of Friday and then email a reminder to my collaborator who did not complete all the tasks.\n\n  hidden_intent:\n    - content: \"List the total number of collaborators who did not complete all the tasks.\"\n    - content: \"List which collaborators do not have completed all of their tasks.\"\n    - content: \"send Gmail reminder emails to every co-author who has not completed the required tasks.\"\nobjectives:\n  tools:\n    - mcp_appworld_gmail__send_email:\n        call: [email_addresses, subject, body]\n  checklist:\n    - criterion: \"Indicate that 17 co-authors have not completed all tasks.\"\n    - criterion: \"The opening of the formal email content should positively thank the collaborators for their contributions so far.\"\n    - criterion: \"This email emphasizes the significance of completing this unfinished task for the entire project.\"\n    - criterion: \"The email tone should be gentle and polite\"\n    - criterion: \"The closing of the formal email content should politely remind the collaborators to complete the remaining tasks.\"\n  tools_evaluation_path: \"data/researcher/scripts/researcher_task_015_gmail.py\"\n\nmetadata:\n  difficulty: medium"}
{"schema_version": "v2", "role": "researcher", "task_id": "researcher_task_016", "user_id": "researcher", "environment_id": "env_researcher", "title": "splitwise", "display_title": "splitwise", "description": "splitwise-phone", "task_type": "short_term_independent", "difficulty": "medium", "trigger_type": "user", "initial_input": "Last week I organized a team-building trip for everyone in the research group. The file bill.txt contains some of the expenses and advance payments from that trip, you can read it. \nMy Splitwise login info:\nEmail: la-mcco@gmail.com\nPassword: T8H2[32\nPlease help handle the payment collection.", "hidden_intents": ["Create a Splitwise group with everyone who participated in this team-building trip", "List all the people who owe money to me", "Create a Splitwise payment record for two expenses I paid in advance: cabin deposit, parking and toll fees .", "Send text messages to other students who also have paid in advance, reminding them to create a Splitwise payment record for the expenses they have paid in advance.,  My phone number is 2873148336, and the phone password is AEQ@e[a"], "hidden_intent_count": 4, "has_objectives": true, "objectives_json": "{\"checklist\": [{\"criterion\": \"State that there are two expenses I paid in advance: the cabin deposit of 560 USD, and parking and toll fees of 30 USD\"}, {\"criterion\": \"Provide a complete list of who owes me money and how much each person owes me in total.\"}, {\"criterion\": \"The reminder text messages to the other classmates who also paid in advance should be polite and sincere\"}, {\"criterion\": \"The reminder text messages to the other classmates who also paid in advance should specify which payment requests they need to send and which people should be included in splitting each cost\"}, {\"criterion\": \"The reminder text messages to the other classmates who also paid in advance should thank them for their contribution to making this team-building trip a success\"}], \"tools\": [{\"mcp_appworld_splitwise__create_group\": {\"call\": [\"name\", \"member_emails\"], \"result\": [\"response.group_id\"]}}, {\"mcp_appworld_splitwise__add_member_to_group\": {\"call\": [\"group_id\", \"user_email\"]}}, {\"mcp_appworld_splitwise__record_expense\": {\"call\": [\"group_id\", \"paid_amount\", \"payer_email\", \"debtor_emails\", \"debt_amounts\"], \"result\": [\"response.expense_id\"]}}, {\"mcp_appworld_phone__send_text_message\": {\"call\": [\"phone_number\", \"message\"], \"result\": [\"response.text_message_id\"]}}], \"tools_evaluation_path\": \"data/researcher/scripts/researcher_task_016_splitwise_phone.py\"}", "metadata_json": "{\"difficulty\": \"medium\"}", "yaml_path": "data/researcher/tasks/researcher_task_016/task.yaml", "yaml": "schema_version: v2\ntask_id: researcher_task_016\nuser_id: researcher\nenvironment_id: env_researcher\ntitle: splitwise\ndescription: \"splitwise-phone\"\ntask_type: short_term_independent\n\ntrigger:\n  type: user\n\nintent:\n  initial_input: |\n    Last week I organized a team-building trip for everyone in the research group. The file bill.txt contains some of the expenses and advance payments from that trip, you can read it. \n    My Splitwise login info:\n    Email: la-mcco@gmail.com\n    Password: T8H2[32\n    Please help handle the payment collection.\n\n  hidden_intent:\n    - content: \"Create a Splitwise group with everyone who participated in this team-building trip\"\n    - content: \"List all the people who owe money to me\"\n    - content: \"Create a Splitwise payment record for two expenses I paid in advance: cabin deposit, parking and toll fees .\"\n    - content: \"Send text messages to other students who also have paid in advance, reminding them to create a Splitwise payment record for the expenses they have paid in advance.,  My phone number is 2873148336, and the phone password is AEQ@e[a   \"\n\nobjectives:\n  tools:\n    - mcp_appworld_splitwise__create_group:\n        call: [name, member_emails]\n        result: [response.group_id]\n    - mcp_appworld_splitwise__add_member_to_group:\n        call: [group_id, user_email]\n    - mcp_appworld_splitwise__record_expense:\n        call: [group_id, paid_amount, payer_email, debtor_emails, debt_amounts]\n        result: [response.expense_id]\n    - mcp_appworld_phone__send_text_message:\n        call: [phone_number, message]\n        result: [response.text_message_id]\n  checklist:\n    - criterion: \"State that there are two expenses I paid in advance: the cabin deposit of 560 USD, and parking and toll fees of 30 USD\"\n    - criterion: \"Provide a complete list of who owes me money and how much each person owes me in total.\"\n    - criterion: \"The reminder text messages to the other classmates who also paid in advance should be polite and sincere\"\n    - criterion: \"The reminder text messages to the other classmates who also paid in advance should specify which payment requests they need to send and which people should be included in splitting each cost\"\n    - criterion: \"The reminder text messages to the other classmates who also paid in advance should thank them for their contribution to making this team-building trip a success\"\n  tools_evaluation_path: \"data/researcher/scripts/researcher_task_016_splitwise_phone.py\"\n\nmetadata:\n  difficulty: medium"}
{"schema_version": "v2", "role": "researcher", "task_id": "researcher_task_017", "user_id": "researcher", "environment_id": "env_researcher", "title": "splitwise", "display_title": "splitwise", "description": "splitwise-phone", "task_type": "short_term_independent", "difficulty": "medium", "trigger_type": "user", "initial_input": "I've recently been enjoying listening to \"Heartstrings Symphony\" while working, \nMy Spotify login info:\nEmail: la-mcco@gmail.com\nPassword: wjhPa4l\nPlease help me write a review for this song and post it on Spotify and give more advices to share this songs .", "hidden_intents": ["Click to like this song", "Post a positive review for this song in Spotify, and show what review was posted.", "Follow the artists about the song in Spotify.", "Write out a way to recommend this song to friends."], "hidden_intent_count": 4, "has_objectives": true, "objectives_json": "{\"checklist\": [{\"criterion\": \"The review should write about the beautiful melody of 'Heartstrings Symphony,' which I've been listening to a lot while working.\"}, {\"criterion\": \"State that the song’s shareable_link is https://spotify.com/songs/54, and that it can be shared with friends to help expand the song’s reach.\"}], \"tools\": [{\"mcp_appworld_spotify__review_song\": {\"call\": [\"song_id\", \"rating\"], \"result\": [\"response\"]}}, {\"mcp_appworld_spotify__follow_artist\": {\"call\": [\"artist_id\"], \"result\": [\"response\"]}}, {\"mcp_appworld_spotify__like_song\": {\"call\": [\"song_id\"], \"result\": [\"response\"]}}], \"tools_evaluation_path\": \"data/researcher/scripts/researcher_task_017_spotify.py\"}", "metadata_json": "{\"difficulty\": \"medium\"}", "yaml_path": "data/researcher/tasks/researcher_task_017/task.yaml", "yaml": "schema_version: v2\ntask_id: researcher_task_017\nuser_id: researcher\nenvironment_id: env_researcher\ntitle: splitwise\ndescription: \"splitwise-phone\"\ntask_type: short_term_independent\n\ntrigger:\n  type: user\n\nintent:\n  initial_input: |\n    I've recently been enjoying listening to \"Heartstrings Symphony\" while working, \n    My Spotify login info:\n    Email: la-mcco@gmail.com\n    Password: wjhPa4l\n    Please help me write a review for this song and post it on Spotify and give more advices to share this songs .\n\n  hidden_intent:\n    - content: \"Click to like this song\"\n    - content: \"Post a positive review for this song in Spotify, and show what review was posted.\"\n    - content: \"Follow the artists about the song in Spotify.\"\n    - content: \"Write out a way to recommend this song to friends.\"\n\nobjectives:\n  tools:\n    - mcp_appworld_spotify__review_song:\n        call: [song_id, rating]\n        result: [response]\n    - mcp_appworld_spotify__follow_artist:\n        call: [artist_id]\n        result: [response]\n    - mcp_appworld_spotify__like_song:\n        call: [song_id]\n        result: [response]\n  checklist:\n    - criterion: \"The review should write about the beautiful melody of 'Heartstrings Symphony,' which I've been listening to a lot while working.\"\n    - criterion: \"State that the song’s shareable_link is https://spotify.com/songs/54, and that it can be shared with friends to help expand the song’s reach.\"\n  tools_evaluation_path: \"data/researcher/scripts/researcher_task_017_spotify.py\"\n\nmetadata:\n  difficulty: medium"}
{"schema_version": "v2", "role": "researcher", "task_id": "researcher_task_018", "user_id": "researcher", "environment_id": "env_researcher", "title": "OpenClaw Technical Q&A Preference Task", "display_title": "OpenClaw Technical Q&A Preference Task", "description": "The user is following the OpenClaw buzz and wants, beyond comparison with typical agents, implicit coverage of product-level pros/cons, introductory blog posts, and usage examples.", "task_type": "short_term_independent", "difficulty": "easy", "trigger_type": "user", "initial_input": "I've noticed OpenClaw getting popular lately and I'm curious. Can you explain how it differs from a regular agent?", "hidden_intents": ["Summarize how OpenClaw compares to existing agent products on the market, and list OpenClaw's strengths and weaknesses.", "Give some OpenClaw-related papers and the research institution and authors of these papers."], "hidden_intent_count": 2, "has_objectives": false, "objectives_json": "", "metadata_json": "{\"difficulty\": \"easy\"}", "yaml_path": "data/researcher/tasks/researcher_task_018/task.yaml", "yaml": "schema_version: v2\ntask_id: researcher_task_018\nuser_id: researcher\nenvironment_id: env_researcher\n\ntitle: \"OpenClaw Technical Q&A Preference Task\"\ndescription: \"The user is following the OpenClaw buzz and wants, beyond comparison with typical agents, implicit coverage of product-level pros/cons, introductory blog posts, and usage examples.\"\n\ntask_type: short_term_independent\n\ntrigger:\n  type: user\n\nintent:\n  initial_input: \"I've noticed OpenClaw getting popular lately and I'm curious. Can you explain how it differs from a regular agent?\"\n  hidden_intent:\n    - content: \"Summarize how OpenClaw compares to existing agent products on the market, and list OpenClaw's strengths and weaknesses.\"\n\n    - content: \"Give some OpenClaw-related papers and the research institution and authors of these papers.\"\n\n\nmetadata:\n  difficulty: easy"}
{"schema_version": "v2", "role": "researcher", "task_id": "researcher_task_019", "user_id": "researcher", "environment_id": "env_researcher", "title": "Paper-Organization Preference Task (paper_record)", "display_title": "Paper-Organization Preference Task (paper_record)", "description": "Collect OpenClaw updates from news and WeChat official accounts, write openclaw_info.md in the workspace, and after each subsequent discussion briefly report additions/changes to paper-related entries in that file.", "task_type": "short_term_independent", "difficulty": "easy", "trigger_type": "user", "initial_input": "There's been a flood of OpenClaw-related news and blog posts lately. Search for a few relevant news articles or WeChat Official Account posts, and summarize the latest developments of OpenClaw.", "hidden_intents": ["Give GitHub links and arxiv for the OpenClaw-related papers."], "hidden_intent_count": 1, "has_objectives": false, "objectives_json": "", "metadata_json": "{\"difficulty\": \"easy\"}", "yaml_path": "data/researcher/tasks/researcher_task_019/task.yaml", "yaml": "schema_version: v2\ntask_id: researcher_task_019\nuser_id: researcher\nenvironment_id: env_researcher\n\ntitle: \"Paper-Organization Preference Task (paper_record)\"\ndescription: \"Collect OpenClaw updates from news and WeChat official accounts, write openclaw_info.md in the workspace, and after each subsequent discussion briefly report additions/changes to paper-related entries in that file.\"\n\ntask_type: short_term_independent\n\ntrigger:\n  type: user\n\nintent:\n  initial_input: \"There's been a flood of OpenClaw-related news and blog posts lately. Search for a few relevant news articles or WeChat Official Account posts, and summarize the latest developments of OpenClaw.\"\n  hidden_intent:\n    - content: \"Give GitHub links and arxiv for the OpenClaw-related papers.\"\nmetadata:\n  difficulty: easy"}
{"schema_version": "v2", "role": "researcher", "task_id": "researcher_task_020", "user_id": "researcher", "environment_id": "env_researcher", "title": "Filter Papers the User Cares About Most from Simulated Hugging Face Daily Papers", "display_title": "Filter Papers the User Cares About Most from Simulated Hugging Face Daily Papers", "description": "Simulate a Hugging Face Daily Papers feed: from the batch, pick OpenClaw-related papers, write recommendations per conventions (summary bullets, arXiv/GitHub links), and report document updates.", "task_type": "short_term_independent", "difficulty": "easy", "trigger_type": "env", "initial_input": "{\"source\":\"huggingface_hub\",\"event\":\"paper_recommendation_trigger\",\"target\":\"agent\",\"payload\":{\"papers\":[{\"id\":\"paper_01\",\"title\":\"Common Corpus: The Largest Collection of Ethical Data for LLM Pre-Training\"},{\"id\":\"paper_02\",\"title\":\"Q-RAG: Long Context Multi‑Step Retrieval via Value‑Based Embedder Training\"},{\"id\":\"paper_03\",\"title\":\"From movement to cognitive maps: recurrent neural networks reveal how locomotor development shapes hippocampal spatial coding\"},{\"id\":\"paper_04\",\"title\":\"FIRE: Frobenius-Isometry Reinitialization for Balancing the Stability–Plasticity Tradeoff\"},{\"id\":\"paper_05\",\"title\":\"Exchangeability of GNN Representations with Applications to Graph Retrieval\"},{\"id\":\"paper_06\",\"title\":\"A Trajectory-Based Safety Audit of Clawdbot (OpenClaw)\"},{\"id\":\"paper_07\",\"title\":\"Enhancing Generative Auto-bidding with Offline Reward Evaluation and Policy Search\"},{\"id\":\"paper_08\",\"title\":\"Why DPO is a Misspecified Estimator and How to Fix It\"},{\"id\":\"paper_09\",\"title\":\"WebDevJudge: Evaluating (M)LLMs as Critiques for Web Development Quality\"},{\"id\":\"paper_10\",\"title\":\"SafeDPO: A Simple Approach to Direct Preference Optimization with Enhanced Safety\"},{\"id\":\"paper_11\",\"title\":\"MedAgentGym: A Scalable Agentic Training Environment for Code-Centric Reasoning in Biomedical Data Science\"},{\"id\":\"paper_12\",\"title\":\"Optimistic Task Inference for Behavior Foundation Models\"},{\"id\":\"paper_13\",\"title\":\"CounselBench: A Large-Scale Expert Evaluation and Adversarial Benchmarking of Large Language Models in Mental Health Question Answering\"},{\"id\":\"paper_14\",\"title\":\"AstaBench: Rigorous Benchmarking of AI Agents with a Scientific Research Suite\"},{\"id\":\"paper_15\",\"title\":\"MetaClaw: Just Talk An Agent That Meta-Learns and Evolves in the Wild\"},{\"id\":\"paper_16\",\"title\":\"Neon: Negative Extrapolation From Self-Training Improves Image Generation\"},{\"id\":\"paper_17\",\"title\":\"Compositional Diffusion with Guided search for Long-Horizon Planning\"},{\"id\":\"paper_18\",\"title\":\"Visual symbolic mechanisms: Emergent symbol processing in Vision Language Models\"},{\"id\":\"paper_19\",\"title\":\"Addressing divergent representations from causal interventions on neural networks\"},{\"id\":\"paper_20\",\"title\":\"Cross-Domain Lossy Compression via Rate- and Classification-Constrained Optimal Transport\"},{\"id\":\"paper_21\",\"title\":\"Latent Fourier Transform\"},{\"id\":\"paper_22\",\"title\":\"GLASS Flows: Efficient Inference for Reward Alignment of Flow and Diffusion Models\"},{\"id\":\"paper_23\",\"title\":\"Latent Speech-Text Transformer\"},{\"id\":\"paper_24\",\"title\":\"LoongRL: Reinforcement Learning for Advanced Reasoning over Long Contexts\"},{\"id\":\"paper_25\",\"title\":\"Improving Developer Emotion Classification via LLM-Based Augmentation\"},{\"id\":\"paper_26\",\"title\":\"Revisiting Multilingual Data Mixtures in Language Model Pretraining\"},{\"id\":\"paper_27\",\"title\":\"One-Shot Style Personalization for RL Agents via Latent Discriminator\"},{\"id\":\"paper_28\",\"title\":\"OpenClaw-RL: Train Any Agent Simply by Talking\"},{\"id\":\"paper_29\",\"title\":\"Compositional HyperModules for Few-Shot Code Adaptation in Meta-Reinforcement Learning\"},{\"id\":\"paper_30\",\"title\":\"All-in-One: Boosting Basic Capabilities in one Omni-MLLM to Enhance Movie Understanding\"},{\"id\":\"paper_31\",\"title\":\"Contrastive Code Graph Embeddings for Reinforcement Learning-Based Automated Code Refactoring\"},{\"id\":\"paper_32\",\"title\":\"Soft Non-Diagonality Penalty Enables Latent Space-Level Interpretability of pLM at No Performance Cost\"},{\"id\":\"paper_33\",\"title\":\"Adaptive Mixing of Non-Invariant Information for Generalized Diffusion Policy\"},{\"id\":\"paper_34\",\"title\":\"Style2Shape: Image Style Guided 3D Shape Material Generation\"},{\"id\":\"paper_35\",\"title\":\"Contrastive-Aligned Knowledge Distillation for Collaborative Code Completion via Multi-Agent Reinforcement Learning\"},{\"id\":\"paper_36\",\"title\":\"Scaling Laws for Generative Reward Models\"},{\"id\":\"paper_37\",\"title\":\"Contrastive-Online-Meta (COM): A Dynamic Adaptation Mechanism for Instruction-Tuned CodeLLMs\"},{\"id\":\"paper_38\",\"title\":\"FedPAC: Consistent Representation Learning for Federated Unsupervised Learning under Data Heterogeneity\"},{\"id\":\"paper_39\",\"title\":\"Training as Computation: A Resource-Bounded Theory of Continual Self-Play Learning\"},{\"id\":\"paper_40\",\"title\":\"Cross-Modal Syntax-NL Attention for Multi-Agent Reinforcement Learning in Collaborative Coding\"},{\"id\":\"paper_41\",\"title\":\"Less is More: Improving Molecular Force Fields with Minimal Temporal Information\"},{\"id\":\"paper_42\",\"title\":\"Curricular Adversarial Training for Robust Code Generation via Hierarchical Reinforcement Learning\"},{\"id\":\"paper_43\",\"title\":\"\\\"Humans welcome to observe\\\": A First Look at the Agent Social Network Moltbook\"},{\"id\":\"paper_44\",\"title\":\"Triple-S: A Sticker Semantic Similarity Benchmark with General Sticker Encoder\"},{\"id\":\"paper_45\",\"title\":\"Unsupervised Dynamic Graph Multi-Model Representation Learning for Temporal Patterns Discovery: Uncovering Parkinson’s Disease Stages Using Cerebrospinal Fluid Longitudinal Profiles\"},{\"id\":\"paper_46\",\"title\":\"Causal Reasoning Favors Encoders: Limits of Decoder-Only Models\"},{\"id\":\"paper_47\",\"title\":\"Hierarchies over Pixels: A Benchmark for Cognitive Geospatial Reasoning for Agents\"},{\"id\":\"paper_48\",\"title\":\"But what is your honest answer? Aiding LLM-judges with honest alternatives using steering vectors\"},{\"id\":\"paper_49\",\"title\":\"LaMbDA: Local Latent Embedding Alignment for Cross-modal Time-Series Diffusion\"}]}}", "hidden_intents": ["Give list OpenClaw-related papers from this recommendation trigger.", "Give add a short introduction and key technical points for each OpenClaw-related paper.", "Give include arXiv links for the OpenClaw-related papers.", "Give GitHub links for the OpenClaw-related papers.", "Give the research institution and author for each OpenClaw-related paper."], "hidden_intent_count": 5, "has_objectives": true, "objectives_json": "{\"checklist\": [{\"criterion\": \"Recommends the paper A Trajectory-Based Safety Audit of Clawdbot.\"}, {\"criterion\": \"Recommends the paper Humans welcome to observe: A First Look at Moltbook.\"}, {\"criterion\": \"Recommends the paper MetaClaw: Just Talk - An Agent That Meta-Learns and Evolves in the Wild.\"}, {\"criterion\": \"Recommends the paper OpenClaw-RL: Train Any Agent Simply by Talking.\"}, {\"criterion\": \"States that the core idea of A Trajectory-Based Safety Audit of Clawdbot is a trajectory-based, six-dimensional safety-audit framework revealing structural security risks for tool agents like Clawdbot under ambiguous intent and adversarial instructions, stemming from permission sprawl/amplification (paraphrase acceptable).\"}, {\"criterion\": \"States that Humans welcome to observe: presents a large-scale empirical study of the first agent-native social network Moltbook, using topic categorization and toxicity metrics to analyze agent behavior evolution, group dynamics, and systemic risks in a native social setting (paraphrase acceptable).\"}, {\"criterion\": \"States that MetaClaw: Just Talk - An Agent That Meta-Learns and Evolves in the Wild proposes MetaClaw, a dual-loop framework combining fast skill evolution from failure trajectories with asynchronous policy optimization using users' idle time, enabling non-stop continual learning under real deployment (paraphrase acceptable).\"}, {\"criterion\": \"States that OpenClaw-RL: Train Any Agent Simply by Talking develops a general asynchronous RL framework turning next-state signals from agent interaction into an online learning source, combining process reward (PRM) and hindsight-guided distillation (OPD) to improve policy performance (paraphrase acceptable).\"}, {\"criterion\": \"Gives the GitHub link for A Trajectory-Based Safety Audit of Clawdbot as https://github.com/tychenn/clawdbot_report.\"}, {\"criterion\": \"Gives the GitHub link for Humans welcome to observe: A First Look at Moltbook as https://moltbookobserve.github.io/.\"}, {\"criterion\": \"Gives the GitHub link for MetaClaw: Just Talk - An Agent That Meta-Learns and Evolves in the Wild as https://github.com/aiming-lab/MetaClaw.\"}, {\"criterion\": \"Gives the GitHub link for OpenClaw-RL: Train Any Agent Simply by Talking as https://github.com/Gen-Verse/OpenClaw-RL.\"}, {\"criterion\": \"Gives the arXiv link for Humans welcome to observe: A First Look at the Agent Social Network Moltbook as https://arxiv.org/abs/2602.10127.\"}, {\"criterion\": \"Gives the arXiv link for A Trajectory-Based Safety Audit of Clawdbot as https://arxiv.org/abs/2602.14364.\"}, {\"criterion\": \"Gives the arXiv link for MetaClaw: Just Talk - An Agent That Meta-Learns and Evolves in the Wild as https://arxiv.org/abs/2603.17187.\"}, {\"criterion\": \"Gives the arXiv link for OpenClaw-RL: Train Any Agent Simply by Talking as https://arxiv.org/abs/2603.10165.\"}, {\"criterion\": \"OpenClaw-RL: Train Any Agent Simply by Talking is authored by Yinjie Wang, Xuyang Chen, Xiaolong Jin, Mengdi Wang, and Ling Yang. The affiliation is Princeton University.\"}, {\"criterion\": \"Humans Welcome to Observe: A First Look at the Agent Social Network Moltbook is authored by Yukun Jiang, Yage Zhang, Xinyue Shen, Michael Backes, and Yang Zhang. The affiliation is CISPA Helmholtz Center for Information Security.\"}, {\"criterion\": \"A Trajectory-Based Safety Audit of Clawdbot is authored by Tianyu Chen, Dongrui Liu, Xia Hu, Jingyi Yu, Wenjie Wang. The affiliations are ShanghaiTech University, Shanghai, China, and Shanghai Artificial Intelligence Laboratory, Shanghai, China.\"}, {\"criterion\": \"MetaClaw: Just Talk - An Agent That Meta-Learns and Evolves in the Wild is authored by Peng Xia1, Jianwen Chen1, Xinyu Yang2, Haoqin Tu3, Jiaqi Liu1, Kaiwen Xiong1, Siwei Han, Shi Qiu, Haonian Ji, Yuyin Zhou, Zeyu Zheng, Cihang Xie, and Huaxiu Yao. The affiliations are UNC-Chapel Hill, Carnegie Mellon University, UC Santa Cruz, and UC Berkeley.\"}]}", "metadata_json": "{\"difficulty\": \"easy\"}", "yaml_path": "data/researcher/tasks/researcher_task_020/task.yaml", "yaml": "schema_version: v2\ntask_id: researcher_task_020\nuser_id: researcher\nenvironment_id: env_researcher\n\ntitle: \"Filter Papers the User Cares About Most from Simulated Hugging Face Daily Papers\"\ndescription: \"Simulate a Hugging Face Daily Papers feed: from the batch, pick OpenClaw-related papers, write recommendations per conventions (summary bullets, arXiv/GitHub links), and report document updates.\"\n\ntask_type: short_term_independent\n\ntrigger:\n  type: env\n\nintent:\n  initial_input: |\n    {\"source\":\"huggingface_hub\",\"event\":\"paper_recommendation_trigger\",\"target\":\"agent\",\"payload\":{\"papers\":[{\"id\":\"paper_01\",\"title\":\"Common Corpus: The Largest Collection of Ethical Data for LLM Pre-Training\"},{\"id\":\"paper_02\",\"title\":\"Q-RAG: Long Context Multi‑Step Retrieval via Value‑Based Embedder Training\"},{\"id\":\"paper_03\",\"title\":\"From movement to cognitive maps: recurrent neural networks reveal how locomotor development shapes hippocampal spatial coding\"},{\"id\":\"paper_04\",\"title\":\"FIRE: Frobenius-Isometry Reinitialization for Balancing the Stability–Plasticity Tradeoff\"},{\"id\":\"paper_05\",\"title\":\"Exchangeability of GNN Representations with Applications to Graph Retrieval\"},{\"id\":\"paper_06\",\"title\":\"A Trajectory-Based Safety Audit of Clawdbot (OpenClaw)\"},{\"id\":\"paper_07\",\"title\":\"Enhancing Generative Auto-bidding with Offline Reward Evaluation and Policy Search\"},{\"id\":\"paper_08\",\"title\":\"Why DPO is a Misspecified Estimator and How to Fix It\"},{\"id\":\"paper_09\",\"title\":\"WebDevJudge: Evaluating (M)LLMs as Critiques for Web Development Quality\"},{\"id\":\"paper_10\",\"title\":\"SafeDPO: A Simple Approach to Direct Preference Optimization with Enhanced Safety\"},{\"id\":\"paper_11\",\"title\":\"MedAgentGym: A Scalable Agentic Training Environment for Code-Centric Reasoning in Biomedical Data Science\"},{\"id\":\"paper_12\",\"title\":\"Optimistic Task Inference for Behavior Foundation Models\"},{\"id\":\"paper_13\",\"title\":\"CounselBench: A Large-Scale Expert Evaluation and Adversarial Benchmarking of Large Language Models in Mental Health Question Answering\"},{\"id\":\"paper_14\",\"title\":\"AstaBench: Rigorous Benchmarking of AI Agents with a Scientific Research Suite\"},{\"id\":\"paper_15\",\"title\":\"MetaClaw: Just Talk An Agent That Meta-Learns and Evolves in the Wild\"},{\"id\":\"paper_16\",\"title\":\"Neon: Negative Extrapolation From Self-Training Improves Image Generation\"},{\"id\":\"paper_17\",\"title\":\"Compositional Diffusion with Guided search for Long-Horizon Planning\"},{\"id\":\"paper_18\",\"title\":\"Visual symbolic mechanisms: Emergent symbol processing in Vision Language Models\"},{\"id\":\"paper_19\",\"title\":\"Addressing divergent representations from causal interventions on neural networks\"},{\"id\":\"paper_20\",\"title\":\"Cross-Domain Lossy Compression via Rate- and Classification-Constrained Optimal Transport\"},{\"id\":\"paper_21\",\"title\":\"Latent Fourier Transform\"},{\"id\":\"paper_22\",\"title\":\"GLASS Flows: Efficient Inference for Reward Alignment of Flow and Diffusion Models\"},{\"id\":\"paper_23\",\"title\":\"Latent Speech-Text Transformer\"},{\"id\":\"paper_24\",\"title\":\"LoongRL: Reinforcement Learning for Advanced Reasoning over Long Contexts\"},{\"id\":\"paper_25\",\"title\":\"Improving Developer Emotion Classification via LLM-Based Augmentation\"},{\"id\":\"paper_26\",\"title\":\"Revisiting Multilingual Data Mixtures in Language Model Pretraining\"},{\"id\":\"paper_27\",\"title\":\"One-Shot Style Personalization for RL Agents via Latent Discriminator\"},{\"id\":\"paper_28\",\"title\":\"OpenClaw-RL: Train Any Agent Simply by Talking\"},{\"id\":\"paper_29\",\"title\":\"Compositional HyperModules for Few-Shot Code Adaptation in Meta-Reinforcement Learning\"},{\"id\":\"paper_30\",\"title\":\"All-in-One: Boosting Basic Capabilities in one Omni-MLLM to Enhance Movie Understanding\"},{\"id\":\"paper_31\",\"title\":\"Contrastive Code Graph Embeddings for Reinforcement Learning-Based Automated Code Refactoring\"},{\"id\":\"paper_32\",\"title\":\"Soft Non-Diagonality Penalty Enables Latent Space-Level Interpretability of pLM at No Performance Cost\"},{\"id\":\"paper_33\",\"title\":\"Adaptive Mixing of Non-Invariant Information for Generalized Diffusion Policy\"},{\"id\":\"paper_34\",\"title\":\"Style2Shape: Image Style Guided 3D Shape Material Generation\"},{\"id\":\"paper_35\",\"title\":\"Contrastive-Aligned Knowledge Distillation for Collaborative Code Completion via Multi-Agent Reinforcement Learning\"},{\"id\":\"paper_36\",\"title\":\"Scaling Laws for Generative Reward Models\"},{\"id\":\"paper_37\",\"title\":\"Contrastive-Online-Meta (COM): A Dynamic Adaptation Mechanism for Instruction-Tuned CodeLLMs\"},{\"id\":\"paper_38\",\"title\":\"FedPAC: Consistent Representation Learning for Federated Unsupervised Learning under Data Heterogeneity\"},{\"id\":\"paper_39\",\"title\":\"Training as Computation: A Resource-Bounded Theory of Continual Self-Play Learning\"},{\"id\":\"paper_40\",\"title\":\"Cross-Modal Syntax-NL Attention for Multi-Agent Reinforcement Learning in Collaborative Coding\"},{\"id\":\"paper_41\",\"title\":\"Less is More: Improving Molecular Force Fields with Minimal Temporal Information\"},{\"id\":\"paper_42\",\"title\":\"Curricular Adversarial Training for Robust Code Generation via Hierarchical Reinforcement Learning\"},{\"id\":\"paper_43\",\"title\":\"\\\"Humans welcome to observe\\\": A First Look at the Agent Social Network Moltbook\"},{\"id\":\"paper_44\",\"title\":\"Triple-S: A Sticker Semantic Similarity Benchmark with General Sticker Encoder\"},{\"id\":\"paper_45\",\"title\":\"Unsupervised Dynamic Graph Multi-Model Representation Learning for Temporal Patterns Discovery: Uncovering Parkinson’s Disease Stages Using Cerebrospinal Fluid Longitudinal Profiles\"},{\"id\":\"paper_46\",\"title\":\"Causal Reasoning Favors Encoders: Limits of Decoder-Only Models\"},{\"id\":\"paper_47\",\"title\":\"Hierarchies over Pixels: A Benchmark for Cognitive Geospatial Reasoning for Agents\"},{\"id\":\"paper_48\",\"title\":\"But what is your honest answer? Aiding LLM-judges with honest alternatives using steering vectors\"},{\"id\":\"paper_49\",\"title\":\"LaMbDA: Local Latent Embedding Alignment for Cross-modal Time-Series Diffusion\"}]}}\n\n  hidden_intent:\n    - content: \"Give list OpenClaw-related papers from this recommendation trigger.\"\n\n    - content: \"Give add a short introduction and key technical points for each OpenClaw-related paper.\"\n\n    - content: \"Give include arXiv links for the OpenClaw-related papers.\"\n\n    - content: \"Give GitHub links for the OpenClaw-related papers.\"\n\n    - content: \"Give the research institution and author for each OpenClaw-related paper.\"\n\nobjectives:\n\n  checklist:\n    - criterion: \"Recommends the paper A Trajectory-Based Safety Audit of Clawdbot.\"\n\n    - criterion: \"Recommends the paper Humans welcome to observe: A First Look at Moltbook.\"\n\n    - criterion: \"Recommends the paper MetaClaw: Just Talk - An Agent That Meta-Learns and Evolves in the Wild.\"\n\n    - criterion: \"Recommends the paper OpenClaw-RL: Train Any Agent Simply by Talking.\"\n\n    - criterion: \"States that the core idea of A Trajectory-Based Safety Audit of Clawdbot is a trajectory-based, six-dimensional safety-audit framework revealing structural security risks for tool agents like Clawdbot under ambiguous intent and adversarial instructions, stemming from permission sprawl/amplification (paraphrase acceptable).\"\n\n    - criterion: \"States that Humans welcome to observe: presents a large-scale empirical study of the first agent-native social network Moltbook, using topic categorization and toxicity metrics to analyze agent behavior evolution, group dynamics, and systemic risks in a native social setting (paraphrase acceptable).\"\n\n    - criterion: \"States that MetaClaw: Just Talk - An Agent That Meta-Learns and Evolves in the Wild proposes MetaClaw, a dual-loop framework combining fast skill evolution from failure trajectories with asynchronous policy optimization using users' idle time, enabling non-stop continual learning under real deployment (paraphrase acceptable).\"\n\n    - criterion: \"States that OpenClaw-RL: Train Any Agent Simply by Talking develops a general asynchronous RL framework turning next-state signals from agent interaction into an online learning source, combining process reward (PRM) and hindsight-guided distillation (OPD) to improve policy performance (paraphrase acceptable).\"\n\n    - criterion: \"Gives the GitHub link for A Trajectory-Based Safety Audit of Clawdbot as https://github.com/tychenn/clawdbot_report.\"\n\n    - criterion: \"Gives the GitHub link for Humans welcome to observe: A First Look at Moltbook as https://moltbookobserve.github.io/.\"\n\n    - criterion: \"Gives the GitHub link for MetaClaw: Just Talk - An Agent That Meta-Learns and Evolves in the Wild as https://github.com/aiming-lab/MetaClaw.\"\n\n    - criterion: \"Gives the GitHub link for OpenClaw-RL: Train Any Agent Simply by Talking as https://github.com/Gen-Verse/OpenClaw-RL.\"\n\n    - criterion: \"Gives the arXiv link for Humans welcome to observe: A First Look at the Agent Social Network Moltbook as https://arxiv.org/abs/2602.10127.\"\n\n    - criterion: \"Gives the arXiv link for A Trajectory-Based Safety Audit of Clawdbot as https://arxiv.org/abs/2602.14364.\"\n\n    - criterion: \"Gives the arXiv link for MetaClaw: Just Talk - An Agent That Meta-Learns and Evolves in the Wild as https://arxiv.org/abs/2603.17187.\"\n\n    - criterion: \"Gives the arXiv link for OpenClaw-RL: Train Any Agent Simply by Talking as https://arxiv.org/abs/2603.10165.\"\n\n    - criterion: \"OpenClaw-RL: Train Any Agent Simply by Talking is authored by Yinjie Wang, Xuyang Chen, Xiaolong Jin, Mengdi Wang, and Ling Yang. The affiliation is Princeton University.\"\n\n    - criterion: \"Humans Welcome to Observe: A First Look at the Agent Social Network Moltbook is authored by Yukun Jiang, Yage Zhang, Xinyue Shen, Michael Backes, and Yang Zhang. The affiliation is CISPA Helmholtz Center for Information Security.\"\n\n    - criterion: \"A Trajectory-Based Safety Audit of Clawdbot is authored by Tianyu Chen, Dongrui Liu, Xia Hu, Jingyi Yu, Wenjie Wang. The affiliations are ShanghaiTech University, Shanghai, China, and Shanghai Artificial Intelligence Laboratory, Shanghai, China.\"\n\n    - criterion: \"MetaClaw: Just Talk - An Agent That Meta-Learns and Evolves in the Wild is authored by Peng Xia1, Jianwen Chen1, Xinyu Yang2, Haoqin Tu3, Jiaqi Liu1, Kaiwen Xiong1, Siwei Han, Shi Qiu, Haonian Ji, Yuyin Zhou, Zeyu Zheng, Cihang Xie, and Huaxiu Yao. The affiliations are UNC-Chapel Hill, Carnegie Mellon University, UC Santa Cruz, and UC Berkeley.\"\n\nmetadata:\n  difficulty: easy"}