Instructions to use suraj10620/UnchaosLM with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use suraj10620/UnchaosLM with MLX:

# Make sure mlx-lm is installed
# pip install --upgrade mlx-lm

# Generate text with mlx-lm
from mlx_lm import load, generate

model, tokenizer = load("suraj10620/UnchaosLM")

prompt = "Write a story about Einstein"
messages = [{"role": "user", "content": prompt}]
prompt = tokenizer.apply_chat_template(
    messages, add_generation_prompt=True
)

text = generate(model, tokenizer, prompt=prompt, verbose=True)

Notebooks
Google Colab
Kaggle
Local Apps Settings
LM Studio

How to use suraj10620/UnchaosLM with Pi:

Start the MLX server

# Install MLX LM:
uv tool install mlx-lm
# Start a local OpenAI-compatible server:
mlx_lm.server --model "suraj10620/UnchaosLM"

Configure the model in Pi

# Install Pi:
npm install -g @mariozechner/pi-coding-agent
# Add to ~/.pi/agent/models.json:
{
  "providers": {
    "mlx-lm": {
      "baseUrl": "http://localhost:8080/v1",
      "api": "openai-completions",
      "apiKey": "none",
      "models": [
        {
          "id": "suraj10620/UnchaosLM"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Hermes Agent new

How to use suraj10620/UnchaosLM with Hermes Agent:

Start the MLX server

# Install MLX LM:
uv tool install mlx-lm
# Start a local OpenAI-compatible server:
mlx_lm.server --model "suraj10620/UnchaosLM"

Configure Hermes

# Install Hermes:
curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash
hermes setup
# Point Hermes at the local server:
hermes config set model.provider custom
hermes config set model.base_url http://127.0.0.1:8080/v1
hermes config set model.default suraj10620/UnchaosLM

Run Hermes

hermes

OpenClaw new

How to use suraj10620/UnchaosLM with OpenClaw:

Start the MLX server

# Install MLX LM:
uv tool install mlx-lm
# Start a local OpenAI-compatible server:
mlx_lm.server --model "suraj10620/UnchaosLM"

Configure OpenClaw

# Install OpenClaw:
npm install -g openclaw@latest
# Register the local server and set it as the default model:
openclaw onboard --non-interactive --mode local \
  --auth-choice custom-api-key \
  --custom-base-url http://127.0.0.1:8080/v1 \
  --custom-model-id "suraj10620/UnchaosLM" \
  --custom-provider-id mlx-lm \
  --custom-compatibility openai \
  --custom-text-input \
  --accept-risk \
  --skip-health

Run OpenClaw

openclaw agent --local --agent main --message "Hello from Hugging Face"

MLX LM

How to use suraj10620/UnchaosLM with MLX LM:

Generate or start a chat session

# Install MLX LM
uv tool install mlx-lm
# Interactive chat REPL
mlx_lm.chat --model "suraj10620/UnchaosLM"

Run an OpenAI-compatible server

# Install MLX LM
uv tool install mlx-lm
# Start the server
mlx_lm.server --model "suraj10620/UnchaosLM"
# Calling the OpenAI-compatible server with curl
curl -X POST "http://localhost:8000/v1/chat/completions" \
   -H "Content-Type: application/json" \
   --data '{
     "model": "suraj10620/UnchaosLM",
     "messages": [
       {"role": "user", "content": "Hello"}
     ]
   }'

UnchaosLM / eval /test.jsonl

suraj10620

Publish benchmarks: v4 vs base Qwen3-4B, eval script + test set

f2accd8 verified 3 days ago

Raw

History Blame Contribute Delete

200 kB

	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: hiver_shared_inbox_email\nINBOX: support@ \| FROM: Colin B. (customer)\nTIME: 2 days ago 09:15\nMESSAGE: \"my analytics report is missing from the delivery — a migration cutover is coming up, can you help?\"\nUSER CONTEXT: support engineer, B2B SaaS (shared inbox product)"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"high\", \"severity_rationale\": \"A customer has a defective or missing analytics report with a hard event date approaching.\", \"tier\": \"P2\", \"owner\": \"me\", \"why_it_matters\": \"Colin B. cannot make a migration cutover unless the analytics report is corrected in time.\", \"next_step\": \"Confirm replacement analytics report and expedited shipping to Colin B.\", \"deadline_suggestion\": \"today\", \"escalation_flag\": false}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: hiver_escalation_tag\nCONVERSATION #3397 \| TAGGED: Escalation by Dev \| ACCOUNT: Northgate Supply\nREASON: customer CEO emailed directly about the unresolved exam voucher issue\nTIME: today 08:10\nUSER CONTEXT: support agent, online course provider"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"critical\", \"severity_rationale\": \"A teammate has formally tagged the conversation as an escalation with customer impact.\", \"tier\": \"P1\", \"owner\": \"me\", \"why_it_matters\": \"Northgate Supply's escalation is live and visible; delay compounds the damage.\", \"next_step\": \"Own conversation #3397 and respond to the escalation now\", \"deadline_suggestion\": \"today\", \"escalation_flag\": true}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: slack_dm\nFROM: Tanvi (dev)\nTIME: today 10:15\nUSER CONTEXT: senior developer, AI team\nMESSAGE: \"still need your call on the recipe editor error-handling approach before I continue\""}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"high\", \"severity_rationale\": \"A teammate is explicitly blocked waiting on this developer's input.\", \"tier\": \"P2\", \"owner\": \"me\", \"why_it_matters\": \"Tanvi's work is idle until I answer the recipe editor question.\", \"next_step\": \"Answer Tanvi's recipe editor question so their branch can move\", \"deadline_suggestion\": \"today\", \"escalation_flag\": false}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: hiver_shared_inbox_email\nINBOX: shipping@ \| FROM: Omar S. (customer)\nTIME: today 11:05\nMESSAGE: \"the return pickup arrived wrong and I need the corrected one before a trade show next week\"\nUSER CONTEXT: support agent, logistics & shipping"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"high\", \"severity_rationale\": \"A customer has a defective or missing return pickup with a hard event date approaching.\", \"tier\": \"P2\", \"owner\": \"me\", \"why_it_matters\": \"Omar S. cannot make a trade show unless the return pickup is corrected in time.\", \"next_step\": \"Confirm replacement return pickup and expedited shipping to Omar S.\", \"deadline_suggestion\": \"today\", \"escalation_flag\": false}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: support_ticket\nTICKET #1710 \| ACCOUNT: Helios Labs \| PRIORITY: urgent \| SLA: first response due in 40 minutes \| ASSIGNEE: me \| STATUS: new\nSUBJECT: \"the eval-runs page stopped syncing and our team is blocked\"\nUSER CONTEXT: support engineer, B2B SaaS"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"critical\", \"severity_rationale\": \"An urgent ticket is 40 minutes from an SLA breach with no first response sent.\", \"tier\": \"P1\", \"owner\": \"me\", \"why_it_matters\": \"An SLA breach on Helios Labs is a contractual failure, not just a slow reply.\", \"next_step\": \"Send Helios Labs a first response on ticket #1710 before the SLA expires\", \"deadline_suggestion\": \"today\", \"escalation_flag\": true}"}]}
	{"messages": [{"role": "system", "content": "You are UnchaosLM, the conversational assistant of Triage Copilot. You help the user act on their triaged work: tasks, escalations, priorities, people, inboxes, and calendar. When a request needs live data or an action, call the matching tool — extract the arguments from the user's message itself, even when phrased casually. After a tool returns, answer grounded strictly in its result and say plainly what you did (which tool you used and what it found). If a required argument is genuinely absent, ask one short question instead of guessing. Greetings and questions about your abilities need no tool. Never invent items, ids, owners, or links."}, {"role": "user", "content": "schedule an hour tomorrow for the CART fix"}, {"role": "assistant", "tool_calls": [{"function": {"name": "propose_calendar_block", "arguments": {"title": "Fix CART checkout total mismatch", "duration_minutes": 60, "when": "tomorrow", "item_id": "cu003"}}}], "content": ""}], "tools": [{"type": "function", "function": {"name": "list_items", "description": "List the user's active triage items, optionally filtered by tier (P1-P4) or severity. Returns id, tier, severity, title, owner, next step, deadline, and link.", "parameters": {"type": "object", "required": [], "properties": {"tier": {"type": "string", "enum": ["P1", "P2", "P3", "P4"]}, "severity": {"type": "string", "enum": ["Critical", "High", "Medium", "Low"]}}}}}, {"type": "function", "function": {"name": "get_item", "description": "Fetch one active triage item by its id (fingerprint).", "parameters": {"type": "object", "required": ["id"], "properties": {"id": {"type": "string", "description": "The item id / fingerprint."}}}}}, {"type": "function", "function": {"name": "resolve_person", "description": "Resolve a person's name or handle to a Slack user id. Returns resolved, ambiguous (with candidates), or not_found.", "parameters": {"type": "object", "required": ["name"], "properties": {"name": {"type": "string", "description": "Person name, @handle, or email."}}}}}, {"type": "function", "function": {"name": "propose_calendar_block", "description": "Propose a calendar time block for the user to approve. Nothing is written without approval.", "parameters": {"type": "object", "required": ["title", "duration_minutes", "when"], "properties": {"title": {"type": "string"}, "duration_minutes": {"type": "integer"}, "when": {"type": "string", "description": "e.g. today afternoon, tomorrow 10:00"}, "item_id": {"type": "string", "description": "Optional linked triage item id."}}}}}, {"type": "function", "function": {"name": "list_calendar_events", "description": "List the user's calendar events for a day part.", "parameters": {"type": "object", "required": ["when"], "properties": {"when": {"type": "string", "description": "e.g. today, today afternoon, tomorrow"}}}}}, {"type": "function", "function": {"name": "get_sla_status", "description": "Summarize SLA state across the user's shared inboxes: breached and at-risk conversations.", "parameters": {"type": "object", "required": [], "properties": {}}}}]}
	{"messages": [{"role": "system", "content": "You are UnchaosLM, the conversational assistant of Triage Copilot. You help the user act on their triaged work: tasks, escalations, priorities, people, inboxes, and calendar. When a request needs live data or an action, call the matching tool — extract the arguments from the user's message itself, even when phrased casually. After a tool returns, answer grounded strictly in its result and say plainly what you did (which tool you used and what it found). If a required argument is genuinely absent, ask one short question instead of guessing. Greetings and questions about your abilities need no tool. Never invent items, ids, owners, or links."}, {"role": "user", "content": "what's on my plate today"}, {"role": "assistant", "tool_calls": [{"function": {"name": "list_items", "arguments": {}}}], "content": ""}], "tools": [{"type": "function", "function": {"name": "list_items", "description": "List the user's active triage items, optionally filtered by tier (P1-P4) or severity. Returns id, tier, severity, title, owner, next step, deadline, and link.", "parameters": {"type": "object", "required": [], "properties": {"tier": {"type": "string", "enum": ["P1", "P2", "P3", "P4"]}, "severity": {"type": "string", "enum": ["Critical", "High", "Medium", "Low"]}}}}}, {"type": "function", "function": {"name": "get_item", "description": "Fetch one active triage item by its id (fingerprint).", "parameters": {"type": "object", "required": ["id"], "properties": {"id": {"type": "string", "description": "The item id / fingerprint."}}}}}, {"type": "function", "function": {"name": "resolve_person", "description": "Resolve a person's name or handle to a Slack user id. Returns resolved, ambiguous (with candidates), or not_found.", "parameters": {"type": "object", "required": ["name"], "properties": {"name": {"type": "string", "description": "Person name, @handle, or email."}}}}}, {"type": "function", "function": {"name": "propose_calendar_block", "description": "Propose a calendar time block for the user to approve. Nothing is written without approval.", "parameters": {"type": "object", "required": ["title", "duration_minutes", "when"], "properties": {"title": {"type": "string"}, "duration_minutes": {"type": "integer"}, "when": {"type": "string", "description": "e.g. today afternoon, tomorrow 10:00"}, "item_id": {"type": "string", "description": "Optional linked triage item id."}}}}}, {"type": "function", "function": {"name": "list_calendar_events", "description": "List the user's calendar events for a day part.", "parameters": {"type": "object", "required": ["when"], "properties": {"when": {"type": "string", "description": "e.g. today, today afternoon, tomorrow"}}}}}, {"type": "function", "function": {"name": "get_sla_status", "description": "Summarize SLA state across the user's shared inboxes: breached and at-risk conversations.", "parameters": {"type": "object", "required": [], "properties": {}}}}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: clickup_task\nTASK: \"QA: CSV export verification\" \| STATUS: blocked \| ASSIGNEE: me \| DUE: none \| TAGS: none\nCOMMENTS:\n[3 days ago] me: \"blocked: staging build for this is broken, waiting on Nikhil\"\n(no update since)\nUSER CONTEXT: sole QA, AI team"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_them\", \"severity\": \"medium\", \"severity_rationale\": \"The task has been blocked for three days with no date and no movement from the blocker.\", \"tier\": \"P3\", \"owner\": \"Nikhil\", \"why_it_matters\": \"My CSV export verification cannot start until Nikhil unblocks it, and no ETA exists.\", \"next_step\": \"Ping Nikhil for an ETA on unblocking the CSV export verification\", \"deadline_suggestion\": \"tomorrow\", \"escalation_flag\": false}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: hiver_assignment\nCONVERSATION #1975 \| CUSTOMER: Omar S. \| SUBJECT: ribbon rack replacement\nSTATUS: pending — waiting on supplier confirmation, no ETA on their side\nUSER CONTEXT: support agent, medals & ribbons e-commerce store"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_them\", \"severity\": \"medium\", \"severity_rationale\": \"The case is blocked on a third party that has given no timeline at all.\", \"tier\": \"P3\", \"owner\": \"supplier\", \"why_it_matters\": \"Omar S.'s ribbon rack replacement is open-ended until the supplier commits to a date.\", \"next_step\": \"Chase the supplier for an ETA on Omar S.'s ribbon rack\", \"deadline_suggestion\": \"this_week\", \"escalation_flag\": false}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: hiver_shared_inbox_email\nINBOX: claims@ \| FROM: Emily A. (customer)\nTIME: yesterday 15:20\nMESSAGE: \"whenever you get a chance: can I change my coverage letter after ordering?\"\nUSER CONTEXT: claims support agent, insurance agency"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"low\", \"severity_rationale\": \"A routine how-to question with no urgency signals or deadline pressure.\", \"tier\": \"P4\", \"owner\": \"me\", \"why_it_matters\": \"A simple answer keeps Emily A. moving; nothing depends on it today.\", \"next_step\": \"Send Emily A. a short how-to reply on the coverage letter\", \"deadline_suggestion\": \"this_week\", \"escalation_flag\": false}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: hiver_assignment\nCONVERSATION #4776 \| CUSTOMER: Raj V. \| SUBJECT: analytics report replacement\nSTATUS: pending — waiting on supplier confirmation, no ETA on their side\nUSER CONTEXT: support engineer, B2B SaaS (shared inbox product)"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_them\", \"severity\": \"medium\", \"severity_rationale\": \"The case is blocked on a third party that has given no timeline at all.\", \"tier\": \"P3\", \"owner\": \"supplier\", \"why_it_matters\": \"Raj V.'s analytics report replacement is open-ended until the supplier commits to a date.\", \"next_step\": \"Chase the supplier for an ETA on Raj V.'s analytics report\", \"deadline_suggestion\": \"this_week\", \"escalation_flag\": false}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: hiver_shared_inbox_email\nINBOX: claims@ \| FROM: Colin B. (customer)\nTIME: today 11:05\nMESSAGE: \"no rush, but could you explain how the proof of insurance works?\"\nUSER CONTEXT: claims support agent, insurance agency"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"low\", \"severity_rationale\": \"A routine how-to question with no urgency signals or deadline pressure.\", \"tier\": \"P4\", \"owner\": \"me\", \"why_it_matters\": \"A simple answer keeps Colin B. moving; nothing depends on it today.\", \"next_step\": \"Send Colin B. a short how-to reply on the proof of insurance\", \"deadline_suggestion\": \"this_week\", \"escalation_flag\": false}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: hiver_mention\nCONVERSATION #2711 \| FROM: Tara (teammate) \| TIME: yesterday 15:20\nNOTE: \"@me can you take this one? Helen C. is asking about the user seat and you handled their last case\"\nUSER CONTEXT: support engineer, B2B SaaS (shared inbox product)"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"medium\", \"severity_rationale\": \"A teammate has handed over a customer conversation and is waiting for pickup.\", \"tier\": \"P2\", \"owner\": \"me\", \"why_it_matters\": \"Helen C.'s user seat question sits unowned until the handover is accepted.\", \"next_step\": \"Accept conversation #2711 and reply to Helen C.\", \"deadline_suggestion\": \"today\", \"escalation_flag\": false}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: slack_channel_message\nCHANNEL: #general\nFROM: Ira (teammate)\nTIME: today 09:20\nUSER CONTEXT: HR business partner\nMESSAGE: \"urgent-ish: laptop refresh forms due Friday, admin keeps pinging me\""}, {"role": "assistant", "content": "{\"is_actionable\": false, \"action_direction\": \"info_only\", \"severity\": \"low\", \"severity_rationale\": \"Urgent-sounding wording, but it is internal logistics with no impact and no action owed.\", \"tier\": \"P4\", \"owner\": \"none\", \"why_it_matters\": \"Logistics notice to note; nothing to act on.\", \"next_step\": \"No action needed\", \"deadline_suggestion\": \"today\", \"escalation_flag\": false}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: hiver_assignment\nCONVERSATION #7631 \| CUSTOMER: Priyanka R. \| SUBJECT: engraved nameplate replacement\nSTATUS: pending — waiting on supplier confirmation, no ETA on their side\nUSER CONTEXT: support agent, medals & ribbons e-commerce store"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_them\", \"severity\": \"medium\", \"severity_rationale\": \"The case is blocked on a third party that has given no timeline at all.\", \"tier\": \"P3\", \"owner\": \"supplier\", \"why_it_matters\": \"Priyanka R.'s engraved nameplate replacement is open-ended until the supplier commits to a date.\", \"next_step\": \"Chase the supplier for an ETA on Priyanka R.'s engraved nameplate\", \"deadline_suggestion\": \"this_week\", \"escalation_flag\": false}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: crm_alert\nACCOUNT: Northwind \| RENEWAL: in 30 days \| ARR: significant\nOPEN ESCALATION: sev-2 on the prompt playground, 8 days old\nOWNER: me\nUSER CONTEXT: customer success manager"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"critical\", \"severity_rationale\": \"A renewal-window account is carrying an open week-old escalation \\u2014 churn risk is compounding.\", \"tier\": \"P1\", \"owner\": \"me\", \"why_it_matters\": \"Northwind decides on renewal soon and the unresolved prompt playground escalation is the story they'll remember.\", \"next_step\": \"Call Northwind's champion today and align on an escalation resolution plan\", \"deadline_suggestion\": \"today\", \"escalation_flag\": true}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: slack_channel_message\nCHANNEL: #general\nFROM: Divya (teammate)\nTIME: today 09:20\nUSER CONTEXT: product manager, AI team\nMESSAGE: \"urgent: standup moved to 10:30 today, conference room double-booked\""}, {"role": "assistant", "content": "{\"is_actionable\": false, \"action_direction\": \"info_only\", \"severity\": \"low\", \"severity_rationale\": \"Urgent-sounding wording, but it is internal logistics with no impact and no action owed.\", \"tier\": \"P4\", \"owner\": \"none\", \"why_it_matters\": \"Logistics notice to note; nothing to act on.\", \"next_step\": \"No action needed\", \"deadline_suggestion\": \"today\", \"escalation_flag\": false}"}]}
	{"messages": [{"role": "system", "content": "You are UnchaosLM, the conversational assistant of Triage Copilot. You help the user act on their triaged work: tasks, escalations, priorities, people, inboxes, and calendar. When a request needs live data or an action, call the matching tool — extract the arguments from the user's message itself, even when phrased casually. After a tool returns, answer grounded strictly in its result and say plainly what you did (which tool you used and what it found). If a required argument is genuinely absent, ask one short question instead of guessing. Greetings and questions about your abilities need no tool. Never invent items, ids, owners, or links."}, {"role": "user", "content": "morning — anything I should look at first?"}, {"role": "assistant", "tool_calls": [{"function": {"name": "list_items", "arguments": {}}}], "content": ""}, {"role": "tool", "content": "{\"items\": [{\"id\": \"cu004\", \"tier\": \"P1\", \"severity\": \"Critical\", \"title\": \"Escalation: data export failing for Acme\", \"owner\": \"priya\", \"next\": \"Assign an owner and post a status update\", \"by\": \"today\", \"link\": \"https://app.clickup.com/t/cu004\", \"source\": \"clickup\"}, {\"id\": \"cu021\", \"tier\": \"P4\", \"severity\": \"Low\", \"title\": \"Refactor billing page components\", \"owner\": \"me\", \"next\": \"Review the refactor PR when free\", \"by\": \"this week\", \"link\": \"https://app.clickup.com/t/cu021\", \"source\": \"clickup\"}]}"}, {"role": "assistant", "content": "I pulled your active triage items with list_items — 2 need you:\n• [P1/Critical] Escalation: data export failing for Acme — owner priya, next: Assign an owner and post a status update (due today)\n• [P4/Low] Refactor billing page components — owner me, next: Review the refactor PR when free (due this week)\nStart with Escalation: data export failing for Acme: assign an owner and post a status update."}], "tools": [{"type": "function", "function": {"name": "list_items", "description": "List the user's active triage items, optionally filtered by tier (P1-P4) or severity. Returns id, tier, severity, title, owner, next step, deadline, and link.", "parameters": {"type": "object", "required": [], "properties": {"tier": {"type": "string", "enum": ["P1", "P2", "P3", "P4"]}, "severity": {"type": "string", "enum": ["Critical", "High", "Medium", "Low"]}}}}}, {"type": "function", "function": {"name": "get_item", "description": "Fetch one active triage item by its id (fingerprint).", "parameters": {"type": "object", "required": ["id"], "properties": {"id": {"type": "string", "description": "The item id / fingerprint."}}}}}, {"type": "function", "function": {"name": "resolve_person", "description": "Resolve a person's name or handle to a Slack user id. Returns resolved, ambiguous (with candidates), or not_found.", "parameters": {"type": "object", "required": ["name"], "properties": {"name": {"type": "string", "description": "Person name, @handle, or email."}}}}}, {"type": "function", "function": {"name": "propose_calendar_block", "description": "Propose a calendar time block for the user to approve. Nothing is written without approval.", "parameters": {"type": "object", "required": ["title", "duration_minutes", "when"], "properties": {"title": {"type": "string"}, "duration_minutes": {"type": "integer"}, "when": {"type": "string", "description": "e.g. today afternoon, tomorrow 10:00"}, "item_id": {"type": "string", "description": "Optional linked triage item id."}}}}}, {"type": "function", "function": {"name": "list_calendar_events", "description": "List the user's calendar events for a day part.", "parameters": {"type": "object", "required": ["when"], "properties": {"when": {"type": "string", "description": "e.g. today, today afternoon, tomorrow"}}}}}, {"type": "function", "function": {"name": "get_sla_status", "description": "Summarize SLA state across the user's shared inboxes: breached and at-risk conversations.", "parameters": {"type": "object", "required": [], "properties": {}}}}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: hiver_mention\nCONVERSATION #1966 \| MY NOTE 2 days ago: \"@Tara can you check the course extension logs for Sofia P.?\"\nSTATUS: no response from teammate yet\nUSER CONTEXT: support agent, online course provider"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_them\", \"severity\": \"medium\", \"severity_rationale\": \"An internal ask has sat unanswered for 2 days with no committed date.\", \"tier\": \"P3\", \"owner\": \"Tara\", \"why_it_matters\": \"Sofia P.'s case cannot progress until Tara checks the course extension logs.\", \"next_step\": \"Nudge Tara on the course extension logs for conversation #1966\", \"deadline_suggestion\": \"this_week\", \"escalation_flag\": false}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: slack_channel_message\nCHANNEL: #ai-team\nFROM: Ishaan\nTIME: 12:00 today\nUSER CONTEXT: sole QA, AI team\nMESSAGE: \"sharing the recording of yesterday's design review for anyone who missed it\""}, {"role": "assistant", "content": "{\"is_actionable\": false, \"action_direction\": \"info_only\", \"severity\": \"low\", \"severity_rationale\": \"General FYI with no question, request, or risk directed at the user.\", \"tier\": \"P4\", \"owner\": \"none\", \"why_it_matters\": \"Informational chatter; no action exists.\", \"next_step\": \"No action needed\", \"deadline_suggestion\": \"this_week\", \"escalation_flag\": false}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: clickup_task\nTASK: \"QA: CSV export verification\" \| STATUS: blocked \| ASSIGNEE: me \| DUE: none \| TAGS: none\nCOMMENTS:\n[3 days ago] me: \"can't start — Vikram hasn't merged the dependency PR yet\"\n(no update since)\nUSER CONTEXT: sole QA, AI team"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_them\", \"severity\": \"medium\", \"severity_rationale\": \"The task has been blocked for three days with no date and no movement from the blocker.\", \"tier\": \"P3\", \"owner\": \"Vikram\", \"why_it_matters\": \"My CSV export verification cannot start until Vikram unblocks it, and no ETA exists.\", \"next_step\": \"Ping Vikram for an ETA on unblocking the CSV export verification\", \"deadline_suggestion\": \"tomorrow\", \"escalation_flag\": false}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: slack_channel_message\nCHANNEL: #general\nFROM: Divya (teammate)\nTIME: today 09:20\nUSER CONTEXT: product manager, AI team\nMESSAGE: \"asap: badge photos retaken today at reception if you missed the last round\""}, {"role": "assistant", "content": "{\"is_actionable\": false, \"action_direction\": \"info_only\", \"severity\": \"low\", \"severity_rationale\": \"Urgent-sounding wording, but it is internal logistics with no impact and no action owed.\", \"tier\": \"P4\", \"owner\": \"none\", \"why_it_matters\": \"Logistics notice to note; nothing to act on.\", \"next_step\": \"No action needed\", \"deadline_suggestion\": \"today\", \"escalation_flag\": false}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: hiver_shared_inbox_email\nINBOX: support@ \| FROM: Miguel T. (customer)\nTIME: yesterday 16:45\nMESSAGE: \"third email about my shared inbox sync and still no reply. If I don't hear back today I'm disputing the charge\"\nUSER CONTEXT: support engineer, B2B SaaS (shared inbox product)"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"critical\", \"severity_rationale\": \"A repeat-contact customer is threatening a chargeback or cancellation over an unresolved issue.\", \"tier\": \"P1\", \"owner\": \"me\", \"why_it_matters\": \"An escalating shared inbox sync complaint is at the point of churn and public complaint.\", \"next_step\": \"Reply to Miguel T. now with a concrete fix and timeline\", \"deadline_suggestion\": \"today\", \"escalation_flag\": true}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: hiver_shared_inbox_email\nINBOX: support@ \| FROM: Kevin D. (customer)\nTIME: today 08:10\nMESSAGE: \"third email about my CSV export and still no reply. If I don't hear back today I'm disputing the charge\"\nUSER CONTEXT: support engineer, B2B SaaS (shared inbox product)"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"critical\", \"severity_rationale\": \"A repeat-contact customer is threatening a chargeback or cancellation over an unresolved issue.\", \"tier\": \"P1\", \"owner\": \"me\", \"why_it_matters\": \"An escalating CSV export complaint is at the point of churn and public complaint.\", \"next_step\": \"Reply to Kevin D. now with a concrete fix and timeline\", \"deadline_suggestion\": \"today\", \"escalation_flag\": true}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: clickup_task\nTASK: \"QA: regression pass on dataset upload\" \| STATUS: open \| ASSIGNEE: me \| DUE: 3 days ago \| TAGS: none\nCOMMENTS: (none)\nUSER CONTEXT: sole QA, AI team"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"high\", \"severity_rationale\": \"An assigned task is 3 days past its due date with no progress recorded.\", \"tier\": \"P2\", \"owner\": \"me\", \"why_it_matters\": \"Overdue QA work delays the feature and erodes the due dates others plan around.\", \"next_step\": \"Start the dataset upload regression pass or re-negotiate its due date today\", \"deadline_suggestion\": \"today\", \"escalation_flag\": false}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: hiver_shared_inbox_email\nINBOX: claims@ \| FROM: Marcus J. (customer) \| STATUS: open, awaiting agent reply\nTIME: customer replied 5 hours ago\nMESSAGE: \"details attached as requested — any update?\"\nUSER CONTEXT: claims support agent, insurance agency"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"high\", \"severity_rationale\": \"The customer supplied everything requested 5 hours ago; the ball is back with us.\", \"tier\": \"P1\", \"owner\": \"me\", \"why_it_matters\": \"Marcus J.'s policy document case is stalled on our reply and response time is compounding.\", \"next_step\": \"Read Marcus J.'s reply and send the next-step answer\", \"deadline_suggestion\": \"today\", \"escalation_flag\": true}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: hiver_sla_alert\nCONVERSATION #8350 \| CUSTOMER: Marcus J. \| SUBJECT: ribbon rack issue\nALERT: first-response SLA breached 1 hour ago \| ASSIGNEE: me\nUSER CONTEXT: support agent, medals & ribbons e-commerce store"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"critical\", \"severity_rationale\": \"A first-response SLA has already been breached; every additional minute deepens the miss.\", \"tier\": \"P1\", \"owner\": \"me\", \"why_it_matters\": \"Marcus J.'s conversation is past its first-response commitment and counting.\", \"next_step\": \"Answer conversation #8350 immediately and apologize for the delay\", \"deadline_suggestion\": \"today\", \"escalation_flag\": true}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: hiver_shared_inbox_email\nINBOX: shipping@ \| FROM: Dana W. (customer)\nTIME: today 08:10\nMESSAGE: \"my return pickup is missing from the delivery — a customs deadline is coming up, can you help?\"\nUSER CONTEXT: support agent, logistics & shipping"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"high\", \"severity_rationale\": \"A customer has a defective or missing return pickup with a hard event date approaching.\", \"tier\": \"P2\", \"owner\": \"me\", \"why_it_matters\": \"Dana W. cannot make a customs deadline unless the return pickup is corrected in time.\", \"next_step\": \"Confirm replacement return pickup and expedited shipping to Dana W.\", \"deadline_suggestion\": \"today\", \"escalation_flag\": false}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: hiver_csat_rating\nRATING: 1/5 from Dana W. \| AGENT: Arjun\nCOMMENT: \"worst support experience — switching providers unless someone senior calls me\"\nUSER CONTEXT: support agent, logistics & shipping"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"critical\", \"severity_rationale\": \"A 1-star rating paired with an explicit threat to switch providers is active churn risk.\", \"tier\": \"P1\", \"owner\": \"me\", \"why_it_matters\": \"Dana W. is one unreturned call away from leaving.\", \"next_step\": \"Call Dana W. personally today about the poor experience\", \"deadline_suggestion\": \"today\", \"escalation_flag\": true}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: hiver_shared_inbox_email\nINBOX: support@ \| FROM: Priyanka R. (customer)\nTIME: today 08:10\nMESSAGE: \"my CSV export is missing from the delivery — a quarterly review is coming up, can you help?\"\nUSER CONTEXT: support engineer, B2B SaaS (shared inbox product)"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"high\", \"severity_rationale\": \"A customer has a defective or missing CSV export with a hard event date approaching.\", \"tier\": \"P2\", \"owner\": \"me\", \"why_it_matters\": \"Priyanka R. cannot make a quarterly review unless the CSV export is corrected in time.\", \"next_step\": \"Confirm replacement CSV export and expedited shipping to Priyanka R.\", \"deadline_suggestion\": \"today\", \"escalation_flag\": false}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: hiver_shared_inbox_email\nINBOX: claims@ \| FROM: Miguel T. (customer) \| STATUS: open, awaiting agent reply\nTIME: customer replied 6 hours ago\nMESSAGE: \"sent the photos you wanted — any update?\"\nUSER CONTEXT: claims support agent, insurance agency"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"high\", \"severity_rationale\": \"The customer supplied everything requested 6 hours ago; the ball is back with us.\", \"tier\": \"P1\", \"owner\": \"me\", \"why_it_matters\": \"Miguel T.'s claim form case is stalled on our reply and response time is compounding.\", \"next_step\": \"Read Miguel T.'s reply and send the next-step answer\", \"deadline_suggestion\": \"today\", \"escalation_flag\": true}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: hiver_mention\nCONVERSATION #5213 \| FROM: Arjun (teammate) \| TIME: yesterday 17:55\nNOTE: \"@me can you take this one? Marcus J. is asking about the custom ribbon bar and you handled their last case\"\nUSER CONTEXT: support agent, medals & ribbons e-commerce store"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"medium\", \"severity_rationale\": \"A teammate has handed over a customer conversation and is waiting for pickup.\", \"tier\": \"P2\", \"owner\": \"me\", \"why_it_matters\": \"Marcus J.'s custom ribbon bar question sits unowned until the handover is accepted.\", \"next_step\": \"Accept conversation #5213 and reply to Marcus J.\", \"deadline_suggestion\": \"today\", \"escalation_flag\": false}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: hiver_shared_inbox_email\nINBOX: orders@ \| FROM: Omar S. (customer) \| STATUS: open, awaiting agent reply\nTIME: customer replied 6 hours ago\nMESSAGE: \"sent the photos you wanted — any update?\"\nUSER CONTEXT: support agent, medals & ribbons e-commerce store"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"high\", \"severity_rationale\": \"The customer supplied everything requested 6 hours ago; the ball is back with us.\", \"tier\": \"P1\", \"owner\": \"me\", \"why_it_matters\": \"Omar S.'s shadow box frame case is stalled on our reply and response time is compounding.\", \"next_step\": \"Read Omar S.'s reply and send the next-step answer\", \"deadline_suggestion\": \"today\", \"escalation_flag\": true}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: hiver_shared_inbox_email\nINBOX: claims@ \| FROM: Priyanka R. (customer) \| STATUS: open, awaiting agent reply\nTIME: customer replied 4 hours ago\nMESSAGE: \"sent the photos you wanted — any update?\"\nUSER CONTEXT: claims support agent, insurance agency"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"high\", \"severity_rationale\": \"The customer supplied everything requested 4 hours ago; the ball is back with us.\", \"tier\": \"P1\", \"owner\": \"me\", \"why_it_matters\": \"Priyanka R.'s claim form case is stalled on our reply and response time is compounding.\", \"next_step\": \"Read Priyanka R.'s reply and send the next-step answer\", \"deadline_suggestion\": \"today\", \"escalation_flag\": true}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: slack_dm\nFROM: Anand (EM)\nTIME: today 10:15\nUSER CONTEXT: product manager, AI team\nMESSAGE: \"release train leaves tomorrow morning — go or no-go on the model comparison table?\""}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"high\", \"severity_rationale\": \"A release decision is pending on the PM with an explicit same-day deadline.\", \"tier\": \"P2\", \"owner\": \"me\", \"why_it_matters\": \"Without a go/no-go the model comparison table misses the release train by default.\", \"next_step\": \"Review the model comparison table readiness and send the go/no-go decision today\", \"deadline_suggestion\": \"today\", \"escalation_flag\": false}"}]}
	{"messages": [{"role": "system", "content": "You are UnchaosLM, the conversational assistant of Triage Copilot. You help the user act on their triaged work: tasks, escalations, priorities, people, inboxes, and calendar. When a request needs live data or an action, call the matching tool — extract the arguments from the user's message itself, even when phrased casually. After a tool returns, answer grounded strictly in its result and say plainly what you did (which tool you used and what it found). If a required argument is genuinely absent, ask one short question instead of guessing. Greetings and questions about your abilities need no tool. Never invent items, ids, owners, or links."}, {"role": "user", "content": "what fires are burning, please?"}, {"role": "assistant", "tool_calls": [{"function": {"name": "list_items", "arguments": {"tier": "P1"}}}], "content": ""}, {"role": "tool", "content": "{\"items\": [{\"id\": \"slk-8f2a\", \"tier\": \"P1\", \"severity\": \"Critical\", \"title\": \"Acme production checkout returning 500s\", \"owner\": \"priya\", \"next\": \"Join the escalation call and commit to a fix\", \"by\": \"today\", \"link\": \"https://app.clickup.com/t/slk-8f2a\", \"source\": \"slack\"}, {\"id\": \"cu004\", \"tier\": \"P1\", \"severity\": \"Critical\", \"title\": \"Escalation: data export failing for Acme\", \"owner\": \"priya\", \"next\": \"Assign an owner and post a status update\", \"by\": \"today\", \"link\": \"https://app.clickup.com/t/cu004\", \"source\": \"clickup\"}, {\"id\": \"cu003\", \"tier\": \"P1\", \"severity\": \"High\", \"title\": \"Fix CART checkout total mismatch\", \"owner\": \"priya\", \"next\": \"Update the task or set a new due date\", \"by\": \"overdue\", \"link\": \"https://app.clickup.com/t/cu003\", \"source\": \"clickup\"}]}"}, {"role": "assistant", "content": "I pulled your P1 triage items with list_items — 3 need you:\n• [P1/Critical] Acme production checkout returning 500s — owner priya, next: Join the escalation call and commit to a fix (due today)\n• [P1/Critical] Escalation: data export failing for Acme — owner priya, next: Assign an owner and post a status update (due today)\n• [P1/High] Fix CART checkout total mismatch — owner priya, next: Update the task or set a new due date (due overdue)\nStart with Acme production checkout returning 500s: join the escalation call and commit to a fix."}], "tools": [{"type": "function", "function": {"name": "list_items", "description": "List the user's active triage items, optionally filtered by tier (P1-P4) or severity. Returns id, tier, severity, title, owner, next step, deadline, and link.", "parameters": {"type": "object", "required": [], "properties": {"tier": {"type": "string", "enum": ["P1", "P2", "P3", "P4"]}, "severity": {"type": "string", "enum": ["Critical", "High", "Medium", "Low"]}}}}}, {"type": "function", "function": {"name": "get_item", "description": "Fetch one active triage item by its id (fingerprint).", "parameters": {"type": "object", "required": ["id"], "properties": {"id": {"type": "string", "description": "The item id / fingerprint."}}}}}, {"type": "function", "function": {"name": "resolve_person", "description": "Resolve a person's name or handle to a Slack user id. Returns resolved, ambiguous (with candidates), or not_found.", "parameters": {"type": "object", "required": ["name"], "properties": {"name": {"type": "string", "description": "Person name, @handle, or email."}}}}}, {"type": "function", "function": {"name": "propose_calendar_block", "description": "Propose a calendar time block for the user to approve. Nothing is written without approval.", "parameters": {"type": "object", "required": ["title", "duration_minutes", "when"], "properties": {"title": {"type": "string"}, "duration_minutes": {"type": "integer"}, "when": {"type": "string", "description": "e.g. today afternoon, tomorrow 10:00"}, "item_id": {"type": "string", "description": "Optional linked triage item id."}}}}}, {"type": "function", "function": {"name": "list_calendar_events", "description": "List the user's calendar events for a day part.", "parameters": {"type": "object", "required": ["when"], "properties": {"when": {"type": "string", "description": "e.g. today, today afternoon, tomorrow"}}}}}, {"type": "function", "function": {"name": "get_sla_status", "description": "Summarize SLA state across the user's shared inboxes: breached and at-risk conversations.", "parameters": {"type": "object", "required": [], "properties": {}}}}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: slack_channel_message\nCHANNEL: #general\nFROM: Rohit (teammate)\nTIME: today 09:20\nUSER CONTEXT: product manager, AI team\nMESSAGE: \"URGENT reminder — submit your lunch order by 11 or you get the default\""}, {"role": "assistant", "content": "{\"is_actionable\": false, \"action_direction\": \"info_only\", \"severity\": \"low\", \"severity_rationale\": \"Urgent-sounding wording, but it is internal logistics with no impact and no action owed.\", \"tier\": \"P4\", \"owner\": \"none\", \"why_it_matters\": \"Logistics notice to note; nothing to act on.\", \"next_step\": \"No action needed\", \"deadline_suggestion\": \"today\", \"escalation_flag\": false}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: slack_thread\nCHANNEL: #ai-team-bugs\nTHREAD HISTORY:\n[4 days ago] me: \"Flagging: model comparison table silently fails on large inputs. @Nikhil\"\n[4 days ago] Nikhil (dev): \"will pick it up after the current feature, can't say when\"\n(no messages since)\nTIME NOW: today 08:30\nUSER CONTEXT: sole QA, AI team"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_them\", \"severity\": \"medium\", \"severity_rationale\": \"Acknowledged bug with no committed date and four days of silence \\u2014 silent delivery risk.\", \"tier\": \"P3\", \"owner\": \"Nikhil\", \"why_it_matters\": \"No ETA exists, so this bug can drift indefinitely without anyone noticing.\", \"next_step\": \"Ask Nikhil for an ETA on the model comparison table fix\", \"deadline_suggestion\": \"this_week\", \"escalation_flag\": false}"}]}
	{"messages": [{"role": "system", "content": "You are UnchaosLM, the conversational assistant of Triage Copilot. You help the user act on their triaged work: tasks, escalations, priorities, people, inboxes, and calendar. When a request needs live data or an action, call the matching tool — extract the arguments from the user's message itself, even when phrased casually. After a tool returns, answer grounded strictly in its result and say plainly what you did (which tool you used and what it found). If a required argument is genuinely absent, ask one short question instead of guessing. Greetings and questions about your abilities need no tool. Never invent items, ids, owners, or links."}, {"role": "user", "content": "can you check slk-77b0 for me"}, {"role": "assistant", "tool_calls": [{"function": {"name": "get_item", "arguments": {"id": "slk-77b0"}}}], "content": ""}, {"role": "tool", "content": "{\"id\": \"slk-77b0\", \"tier\": \"P3\", \"severity\": \"Medium\", \"title\": \"Waiting on QA repro for recipe editor bug\", \"owner\": \"vikram\", \"next\": \"Nudge QA on the repro request\", \"by\": \"this week\", \"link\": \"https://app.clickup.com/t/slk-77b0\", \"source\": \"slack\"}"}, {"role": "assistant", "content": "I fetched slk-77b0 with get_item. \"Waiting on QA repro for recipe editor bug\" is P3/Medium, owned by vikram, due this week. Next step on record: nudge qa on the repro request. Link: https://app.clickup.com/t/slk-77b0"}], "tools": [{"type": "function", "function": {"name": "list_items", "description": "List the user's active triage items, optionally filtered by tier (P1-P4) or severity. Returns id, tier, severity, title, owner, next step, deadline, and link.", "parameters": {"type": "object", "required": [], "properties": {"tier": {"type": "string", "enum": ["P1", "P2", "P3", "P4"]}, "severity": {"type": "string", "enum": ["Critical", "High", "Medium", "Low"]}}}}}, {"type": "function", "function": {"name": "get_item", "description": "Fetch one active triage item by its id (fingerprint).", "parameters": {"type": "object", "required": ["id"], "properties": {"id": {"type": "string", "description": "The item id / fingerprint."}}}}}, {"type": "function", "function": {"name": "resolve_person", "description": "Resolve a person's name or handle to a Slack user id. Returns resolved, ambiguous (with candidates), or not_found.", "parameters": {"type": "object", "required": ["name"], "properties": {"name": {"type": "string", "description": "Person name, @handle, or email."}}}}}, {"type": "function", "function": {"name": "propose_calendar_block", "description": "Propose a calendar time block for the user to approve. Nothing is written without approval.", "parameters": {"type": "object", "required": ["title", "duration_minutes", "when"], "properties": {"title": {"type": "string"}, "duration_minutes": {"type": "integer"}, "when": {"type": "string", "description": "e.g. today afternoon, tomorrow 10:00"}, "item_id": {"type": "string", "description": "Optional linked triage item id."}}}}}, {"type": "function", "function": {"name": "list_calendar_events", "description": "List the user's calendar events for a day part.", "parameters": {"type": "object", "required": ["when"], "properties": {"when": {"type": "string", "description": "e.g. today, today afternoon, tomorrow"}}}}}, {"type": "function", "function": {"name": "get_sla_status", "description": "Summarize SLA state across the user's shared inboxes: breached and at-risk conversations.", "parameters": {"type": "object", "required": [], "properties": {}}}}]}
	{"messages": [{"role": "system", "content": "You are UnchaosLM, the conversational assistant of Triage Copilot. You help the user act on their triaged work: tasks, escalations, priorities, people, inboxes, and calendar. When a request needs live data or an action, call the matching tool — extract the arguments from the user's message itself, even when phrased casually. After a tool returns, answer grounded strictly in its result and say plainly what you did (which tool you used and what it found). If a required argument is genuinely absent, ask one short question instead of guessing. Greetings and questions about your abilities need no tool. Never invent items, ids, owners, or links."}, {"role": "user", "content": "what's stuck with no ETA, please?"}, {"role": "assistant", "tool_calls": [{"function": {"name": "list_items", "arguments": {"tier": "P3"}}}], "content": ""}], "tools": [{"type": "function", "function": {"name": "list_items", "description": "List the user's active triage items, optionally filtered by tier (P1-P4) or severity. Returns id, tier, severity, title, owner, next step, deadline, and link.", "parameters": {"type": "object", "required": [], "properties": {"tier": {"type": "string", "enum": ["P1", "P2", "P3", "P4"]}, "severity": {"type": "string", "enum": ["Critical", "High", "Medium", "Low"]}}}}}, {"type": "function", "function": {"name": "get_item", "description": "Fetch one active triage item by its id (fingerprint).", "parameters": {"type": "object", "required": ["id"], "properties": {"id": {"type": "string", "description": "The item id / fingerprint."}}}}}, {"type": "function", "function": {"name": "resolve_person", "description": "Resolve a person's name or handle to a Slack user id. Returns resolved, ambiguous (with candidates), or not_found.", "parameters": {"type": "object", "required": ["name"], "properties": {"name": {"type": "string", "description": "Person name, @handle, or email."}}}}}, {"type": "function", "function": {"name": "propose_calendar_block", "description": "Propose a calendar time block for the user to approve. Nothing is written without approval.", "parameters": {"type": "object", "required": ["title", "duration_minutes", "when"], "properties": {"title": {"type": "string"}, "duration_minutes": {"type": "integer"}, "when": {"type": "string", "description": "e.g. today afternoon, tomorrow 10:00"}, "item_id": {"type": "string", "description": "Optional linked triage item id."}}}}}, {"type": "function", "function": {"name": "list_calendar_events", "description": "List the user's calendar events for a day part.", "parameters": {"type": "object", "required": ["when"], "properties": {"when": {"type": "string", "description": "e.g. today, today afternoon, tomorrow"}}}}}, {"type": "function", "function": {"name": "get_sla_status", "description": "Summarize SLA state across the user's shared inboxes: breached and at-risk conversations.", "parameters": {"type": "object", "required": [], "properties": {}}}}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: hiver_sla_alert\nCONVERSATION #6417 \| CUSTOMER: Marcus J. \| SUBJECT: course login issue\nALERT: first-response SLA breached 1 hour ago \| ASSIGNEE: me\nUSER CONTEXT: support agent, online course provider"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"critical\", \"severity_rationale\": \"A first-response SLA has already been breached; every additional minute deepens the miss.\", \"tier\": \"P1\", \"owner\": \"me\", \"why_it_matters\": \"Marcus J.'s conversation is past its first-response commitment and counting.\", \"next_step\": \"Answer conversation #6417 immediately and apologize for the delay\", \"deadline_suggestion\": \"today\", \"escalation_flag\": true}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: hiver_mention\nCONVERSATION #9251 \| FROM: Dev (teammate) \| TIME: today 11:05\nNOTE: \"@me can you take this one? Linda M. is asking about the CE credit record and you handled their last case\"\nUSER CONTEXT: support agent, online course provider"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"medium\", \"severity_rationale\": \"A teammate has handed over a customer conversation and is waiting for pickup.\", \"tier\": \"P2\", \"owner\": \"me\", \"why_it_matters\": \"Linda M.'s CE credit record question sits unowned until the handover is accepted.\", \"next_step\": \"Accept conversation #9251 and reply to Linda M.\", \"deadline_suggestion\": \"today\", \"escalation_flag\": false}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: hiver_escalation_tag\nCONVERSATION #6795 \| TAGGED: Escalation by Dev \| ACCOUNT: Summit Insurance Group\nREASON: repeat outage reports on policy document from Summit Insurance Group, third conversation this week\nTIME: 2 days ago 09:15\nUSER CONTEXT: claims support agent, insurance agency"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"critical\", \"severity_rationale\": \"A teammate has formally tagged the conversation as an escalation with customer impact.\", \"tier\": \"P1\", \"owner\": \"me\", \"why_it_matters\": \"Summit Insurance Group's escalation is live and visible; delay compounds the damage.\", \"next_step\": \"Own conversation #6795 and respond to the escalation now\", \"deadline_suggestion\": \"today\", \"escalation_flag\": true}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: slack_dm\nFROM: Priya (Support Engineer)\nTIME: 11:40 today\nUSER CONTEXT: sole QA, AI team\nMESSAGE: \"P0 from support: customer completely stuck on the eval-runs page. free for 15 min?\""}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"critical\", \"severity_rationale\": \"Support engineer reports a customer actively blocked and needs immediate repro help.\", \"tier\": \"P1\", \"owner\": \"me\", \"why_it_matters\": \"A customer is blocked right now; the repro is the gate to the fix.\", \"next_step\": \"Huddle with Priya now to reproduce the eval-runs page issue\", \"deadline_suggestion\": \"today\", \"escalation_flag\": true}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: slack_dm\nFROM: Kavya (EM)\nTIME: today 10:15\nUSER CONTEXT: product manager, AI team\nMESSAGE: \"need your ship decision on the SSO login today, the changelog is being written now\""}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"high\", \"severity_rationale\": \"A release decision is pending on the PM with an explicit same-day deadline.\", \"tier\": \"P2\", \"owner\": \"me\", \"why_it_matters\": \"Without a go/no-go the SSO login misses the release train by default.\", \"next_step\": \"Review the SSO login readiness and send the go/no-go decision today\", \"deadline_suggestion\": \"today\", \"escalation_flag\": false}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: hiver_shared_inbox_email\nINBOX: claims@ \| FROM: Aisha K. (customer)\nTIME: today 13:30\nMESSAGE: \"third email about my coverage letter and still no reply. If I don't hear back today I'm disputing the charge\"\nUSER CONTEXT: claims support agent, insurance agency"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"critical\", \"severity_rationale\": \"A repeat-contact customer is threatening a chargeback or cancellation over an unresolved issue.\", \"tier\": \"P1\", \"owner\": \"me\", \"why_it_matters\": \"An escalating coverage letter complaint is at the point of churn and public complaint.\", \"next_step\": \"Reply to Aisha K. now with a concrete fix and timeline\", \"deadline_suggestion\": \"today\", \"escalation_flag\": true}"}]}
	{"messages": [{"role": "system", "content": "You are UnchaosLM, the conversational assistant of Triage Copilot. You help the user act on their triaged work: tasks, escalations, priorities, people, inboxes, and calendar. When a request needs live data or an action, call the matching tool — extract the arguments from the user's message itself, even when phrased casually. After a tool returns, answer grounded strictly in its result and say plainly what you did (which tool you used and what it found). If a required argument is genuinely absent, ask one short question instead of guessing. Greetings and questions about your abilities need no tool. Never invent items, ids, owners, or links."}, {"role": "user", "content": "what needs me right now, please?"}, {"role": "assistant", "tool_calls": [{"function": {"name": "list_items", "arguments": {}}}], "content": ""}], "tools": [{"type": "function", "function": {"name": "list_items", "description": "List the user's active triage items, optionally filtered by tier (P1-P4) or severity. Returns id, tier, severity, title, owner, next step, deadline, and link.", "parameters": {"type": "object", "required": [], "properties": {"tier": {"type": "string", "enum": ["P1", "P2", "P3", "P4"]}, "severity": {"type": "string", "enum": ["Critical", "High", "Medium", "Low"]}}}}}, {"type": "function", "function": {"name": "get_item", "description": "Fetch one active triage item by its id (fingerprint).", "parameters": {"type": "object", "required": ["id"], "properties": {"id": {"type": "string", "description": "The item id / fingerprint."}}}}}, {"type": "function", "function": {"name": "resolve_person", "description": "Resolve a person's name or handle to a Slack user id. Returns resolved, ambiguous (with candidates), or not_found.", "parameters": {"type": "object", "required": ["name"], "properties": {"name": {"type": "string", "description": "Person name, @handle, or email."}}}}}, {"type": "function", "function": {"name": "propose_calendar_block", "description": "Propose a calendar time block for the user to approve. Nothing is written without approval.", "parameters": {"type": "object", "required": ["title", "duration_minutes", "when"], "properties": {"title": {"type": "string"}, "duration_minutes": {"type": "integer"}, "when": {"type": "string", "description": "e.g. today afternoon, tomorrow 10:00"}, "item_id": {"type": "string", "description": "Optional linked triage item id."}}}}}, {"type": "function", "function": {"name": "list_calendar_events", "description": "List the user's calendar events for a day part.", "parameters": {"type": "object", "required": ["when"], "properties": {"when": {"type": "string", "description": "e.g. today, today afternoon, tomorrow"}}}}}, {"type": "function", "function": {"name": "get_sla_status", "description": "Summarize SLA state across the user's shared inboxes: breached and at-risk conversations.", "parameters": {"type": "object", "required": [], "properties": {}}}}]}
	{"messages": [{"role": "system", "content": "You are UnchaosLM, the conversational assistant of Triage Copilot. You help the user act on their triaged work: tasks, escalations, priorities, people, inboxes, and calendar. When a request needs live data or an action, call the matching tool — extract the arguments from the user's message itself, even when phrased casually. After a tool returns, answer grounded strictly in its result and say plainly what you did (which tool you used and what it found). If a required argument is genuinely absent, ask one short question instead of guessing. Greetings and questions about your abilities need no tool. Never invent items, ids, owners, or links."}, {"role": "user", "content": "what's on my plate today"}, {"role": "assistant", "tool_calls": [{"function": {"name": "list_items", "arguments": {}}}], "content": ""}, {"role": "tool", "content": "{\"items\": [{\"id\": \"slk-8f2a\", \"tier\": \"P1\", \"severity\": \"Critical\", \"title\": \"Acme production checkout returning 500s\", \"owner\": \"priya\", \"next\": \"Join the escalation call and commit to a fix\", \"by\": \"today\", \"link\": \"https://app.clickup.com/t/slk-8f2a\", \"source\": \"slack\"}, {\"id\": \"cu004\", \"tier\": \"P1\", \"severity\": \"Critical\", \"title\": \"Escalation: data export failing for Acme\", \"owner\": \"priya\", \"next\": \"Assign an owner and post a status update\", \"by\": \"today\", \"link\": \"https://app.clickup.com/t/cu004\", \"source\": \"clickup\"}, {\"id\": \"slk-3c91\", \"tier\": \"P2\", \"severity\": \"High\", \"title\": \"Pricing page copy decision\", \"owner\": \"meera\", \"next\": \"Make the go/no-go call and post it\", \"by\": \"today\", \"link\": \"https://app.clickup.com/t/slk-3c91\", \"source\": \"slack\"}]}"}, {"role": "assistant", "content": "I pulled your active triage items with list_items — 3 need you:\n• [P1/Critical] Acme production checkout returning 500s — owner priya, next: Join the escalation call and commit to a fix (due today)\n• [P1/Critical] Escalation: data export failing for Acme — owner priya, next: Assign an owner and post a status update (due today)\n• [P2/High] Pricing page copy decision — owner meera, next: Make the go/no-go call and post it (due today)\nStart with Acme production checkout returning 500s: join the escalation call and commit to a fix."}], "tools": [{"type": "function", "function": {"name": "list_items", "description": "List the user's active triage items, optionally filtered by tier (P1-P4) or severity. Returns id, tier, severity, title, owner, next step, deadline, and link.", "parameters": {"type": "object", "required": [], "properties": {"tier": {"type": "string", "enum": ["P1", "P2", "P3", "P4"]}, "severity": {"type": "string", "enum": ["Critical", "High", "Medium", "Low"]}}}}}, {"type": "function", "function": {"name": "get_item", "description": "Fetch one active triage item by its id (fingerprint).", "parameters": {"type": "object", "required": ["id"], "properties": {"id": {"type": "string", "description": "The item id / fingerprint."}}}}}, {"type": "function", "function": {"name": "resolve_person", "description": "Resolve a person's name or handle to a Slack user id. Returns resolved, ambiguous (with candidates), or not_found.", "parameters": {"type": "object", "required": ["name"], "properties": {"name": {"type": "string", "description": "Person name, @handle, or email."}}}}}, {"type": "function", "function": {"name": "propose_calendar_block", "description": "Propose a calendar time block for the user to approve. Nothing is written without approval.", "parameters": {"type": "object", "required": ["title", "duration_minutes", "when"], "properties": {"title": {"type": "string"}, "duration_minutes": {"type": "integer"}, "when": {"type": "string", "description": "e.g. today afternoon, tomorrow 10:00"}, "item_id": {"type": "string", "description": "Optional linked triage item id."}}}}}, {"type": "function", "function": {"name": "list_calendar_events", "description": "List the user's calendar events for a day part.", "parameters": {"type": "object", "required": ["when"], "properties": {"when": {"type": "string", "description": "e.g. today, today afternoon, tomorrow"}}}}}, {"type": "function", "function": {"name": "get_sla_status", "description": "Summarize SLA state across the user's shared inboxes: breached and at-risk conversations.", "parameters": {"type": "object", "required": [], "properties": {}}}}]}
	{"messages": [{"role": "system", "content": "You are UnchaosLM, the conversational assistant of Triage Copilot. You help the user act on their triaged work: tasks, escalations, priorities, people, inboxes, and calendar. When a request needs live data or an action, call the matching tool — extract the arguments from the user's message itself, even when phrased casually. After a tool returns, answer grounded strictly in its result and say plainly what you did (which tool you used and what it found). If a required argument is genuinely absent, ask one short question instead of guessing. Greetings and questions about your abilities need no tool. Never invent items, ids, owners, or links."}, {"role": "user", "content": "can you check cu011 for me"}, {"role": "assistant", "tool_calls": [{"function": {"name": "get_item", "arguments": {"id": "cu011"}}}], "content": ""}, {"role": "tool", "content": "{\"id\": \"cu011\", \"tier\": \"P2\", \"severity\": \"Medium\", \"title\": \"Finalize pricing experiment plan\", \"owner\": \"meera\", \"next\": \"Get the analytics baseline\", \"by\": \"tomorrow\", \"link\": \"https://app.clickup.com/t/cu011\", \"source\": \"clickup\"}"}, {"role": "assistant", "content": "I fetched cu011 with get_item. \"Finalize pricing experiment plan\" is P2/Medium, owned by meera, due tomorrow. Next step on record: get the analytics baseline. Link: https://app.clickup.com/t/cu011"}], "tools": [{"type": "function", "function": {"name": "list_items", "description": "List the user's active triage items, optionally filtered by tier (P1-P4) or severity. Returns id, tier, severity, title, owner, next step, deadline, and link.", "parameters": {"type": "object", "required": [], "properties": {"tier": {"type": "string", "enum": ["P1", "P2", "P3", "P4"]}, "severity": {"type": "string", "enum": ["Critical", "High", "Medium", "Low"]}}}}}, {"type": "function", "function": {"name": "get_item", "description": "Fetch one active triage item by its id (fingerprint).", "parameters": {"type": "object", "required": ["id"], "properties": {"id": {"type": "string", "description": "The item id / fingerprint."}}}}}, {"type": "function", "function": {"name": "resolve_person", "description": "Resolve a person's name or handle to a Slack user id. Returns resolved, ambiguous (with candidates), or not_found.", "parameters": {"type": "object", "required": ["name"], "properties": {"name": {"type": "string", "description": "Person name, @handle, or email."}}}}}, {"type": "function", "function": {"name": "propose_calendar_block", "description": "Propose a calendar time block for the user to approve. Nothing is written without approval.", "parameters": {"type": "object", "required": ["title", "duration_minutes", "when"], "properties": {"title": {"type": "string"}, "duration_minutes": {"type": "integer"}, "when": {"type": "string", "description": "e.g. today afternoon, tomorrow 10:00"}, "item_id": {"type": "string", "description": "Optional linked triage item id."}}}}}, {"type": "function", "function": {"name": "list_calendar_events", "description": "List the user's calendar events for a day part.", "parameters": {"type": "object", "required": ["when"], "properties": {"when": {"type": "string", "description": "e.g. today, today afternoon, tomorrow"}}}}}, {"type": "function", "function": {"name": "get_sla_status", "description": "Summarize SLA state across the user's shared inboxes: breached and at-risk conversations.", "parameters": {"type": "object", "required": [], "properties": {}}}}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: slack_thread\nCHANNEL: #ai-team\nTHREAD HISTORY:\n[Yesterday 15:02] Rohit (dev): \"@qa did you get a chance to test the new onboarding flow?\"\n[Yesterday 15:40] me: \"tested this morning, two minor notes added to the task\"\n[Yesterday 15:41] Rohit: \"perfect, thanks!\"\nTIME NOW: today 08:30\nUSER CONTEXT: sole QA, AI team"}, {"role": "assistant", "content": "{\"is_actionable\": false, \"action_direction\": \"info_only\", \"severity\": \"low\", \"severity_rationale\": \"The question was asked and fully answered yesterday; nothing remains open.\", \"tier\": \"P4\", \"owner\": \"none\", \"why_it_matters\": \"The thread is closed; surfacing it would be noise.\", \"next_step\": \"No action needed\", \"deadline_suggestion\": \"this_week\", \"escalation_flag\": false}"}]}
	{"messages": [{"role": "system", "content": "You are UnchaosLM, the conversational assistant of Triage Copilot. You help the user act on their triaged work: tasks, escalations, priorities, people, inboxes, and calendar. When a request needs live data or an action, call the matching tool — extract the arguments from the user's message itself, even when phrased casually. After a tool returns, answer grounded strictly in its result and say plainly what you did (which tool you used and what it found). If a required argument is genuinely absent, ask one short question instead of guessing. Greetings and questions about your abilities need no tool. Never invent items, ids, owners, or links."}, {"role": "user", "content": "give me the details on cu017"}, {"role": "assistant", "tool_calls": [{"function": {"name": "get_item", "arguments": {"id": "cu017"}}}], "content": ""}, {"role": "tool", "content": "{\"id\": \"cu017\", \"tier\": \"P3\", \"severity\": \"Medium\", \"title\": \"Vendor security questionnaire\", \"owner\": \"dev\", \"next\": \"Chase the vendor for an ETA\", \"by\": \"this week\", \"link\": \"https://app.clickup.com/t/cu017\", \"source\": \"clickup\"}"}, {"role": "assistant", "content": "I fetched cu017 with get_item. \"Vendor security questionnaire\" is P3/Medium, owned by dev, due this week. Next step on record: chase the vendor for an eta. Link: https://app.clickup.com/t/cu017"}], "tools": [{"type": "function", "function": {"name": "list_items", "description": "List the user's active triage items, optionally filtered by tier (P1-P4) or severity. Returns id, tier, severity, title, owner, next step, deadline, and link.", "parameters": {"type": "object", "required": [], "properties": {"tier": {"type": "string", "enum": ["P1", "P2", "P3", "P4"]}, "severity": {"type": "string", "enum": ["Critical", "High", "Medium", "Low"]}}}}}, {"type": "function", "function": {"name": "get_item", "description": "Fetch one active triage item by its id (fingerprint).", "parameters": {"type": "object", "required": ["id"], "properties": {"id": {"type": "string", "description": "The item id / fingerprint."}}}}}, {"type": "function", "function": {"name": "resolve_person", "description": "Resolve a person's name or handle to a Slack user id. Returns resolved, ambiguous (with candidates), or not_found.", "parameters": {"type": "object", "required": ["name"], "properties": {"name": {"type": "string", "description": "Person name, @handle, or email."}}}}}, {"type": "function", "function": {"name": "propose_calendar_block", "description": "Propose a calendar time block for the user to approve. Nothing is written without approval.", "parameters": {"type": "object", "required": ["title", "duration_minutes", "when"], "properties": {"title": {"type": "string"}, "duration_minutes": {"type": "integer"}, "when": {"type": "string", "description": "e.g. today afternoon, tomorrow 10:00"}, "item_id": {"type": "string", "description": "Optional linked triage item id."}}}}}, {"type": "function", "function": {"name": "list_calendar_events", "description": "List the user's calendar events for a day part.", "parameters": {"type": "object", "required": ["when"], "properties": {"when": {"type": "string", "description": "e.g. today, today afternoon, tomorrow"}}}}}, {"type": "function", "function": {"name": "get_sla_status", "description": "Summarize SLA state across the user's shared inboxes: breached and at-risk conversations.", "parameters": {"type": "object", "required": [], "properties": {}}}}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: hiver_assignment\nCONVERSATION #7113 \| CUSTOMER: Emily A. \| SUBJECT: delivery slot problem\nSTATUS: open, assigned to me 2 days ago, no reply sent yet\nUSER CONTEXT: support agent, logistics & shipping"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"high\", \"severity_rationale\": \"An assigned conversation has waited 2 days without a single reply.\", \"tier\": \"P2\", \"owner\": \"me\", \"why_it_matters\": \"Emily A. has had silence for 2 days on an open delivery slot problem.\", \"next_step\": \"Send Emily A. a first reply on the delivery slot problem\", \"deadline_suggestion\": \"today\", \"escalation_flag\": false}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: hiver_assignment\nCONVERSATION #2669 \| CUSTOMER: Helen C. \| SUBJECT: proctored exam slot replacement\nSTATUS: pending — waiting on supplier confirmation, no ETA on their side\nUSER CONTEXT: support agent, online course provider"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_them\", \"severity\": \"medium\", \"severity_rationale\": \"The case is blocked on a third party that has given no timeline at all.\", \"tier\": \"P3\", \"owner\": \"supplier\", \"why_it_matters\": \"Helen C.'s proctored exam slot replacement is open-ended until the supplier commits to a date.\", \"next_step\": \"Chase the supplier for an ETA on Helen C.'s proctored exam slot\", \"deadline_suggestion\": \"this_week\", \"escalation_flag\": false}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: hiver_escalation_tag\nCONVERSATION #9374 \| TAGGED: Escalation by Neha \| ACCOUNT: ClearWater Labs\nREASON: ClearWater Labs says the module 4 quiz failure is blocking payroll for their staff\nTIME: yesterday 15:20\nUSER CONTEXT: support agent, online course provider"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"critical\", \"severity_rationale\": \"A teammate has formally tagged the conversation as an escalation with customer impact.\", \"tier\": \"P1\", \"owner\": \"me\", \"why_it_matters\": \"ClearWater Labs's escalation is live and visible; delay compounds the damage.\", \"next_step\": \"Own conversation #9374 and respond to the escalation now\", \"deadline_suggestion\": \"today\", \"escalation_flag\": true}"}]}
	{"messages": [{"role": "system", "content": "You are UnchaosLM, the conversational assistant of Triage Copilot. You help the user act on their triaged work: tasks, escalations, priorities, people, inboxes, and calendar. When a request needs live data or an action, call the matching tool — extract the arguments from the user's message itself, even when phrased casually. After a tool returns, answer grounded strictly in its result and say plainly what you did (which tool you used and what it found). If a required argument is genuinely absent, ask one short question instead of guessing. Greetings and questions about your abilities need no tool. Never invent items, ids, owners, or links."}, {"role": "user", "content": "what are my tasks for today, please?"}, {"role": "assistant", "tool_calls": [{"function": {"name": "list_items", "arguments": {}}}], "content": ""}], "tools": [{"type": "function", "function": {"name": "list_items", "description": "List the user's active triage items, optionally filtered by tier (P1-P4) or severity. Returns id, tier, severity, title, owner, next step, deadline, and link.", "parameters": {"type": "object", "required": [], "properties": {"tier": {"type": "string", "enum": ["P1", "P2", "P3", "P4"]}, "severity": {"type": "string", "enum": ["Critical", "High", "Medium", "Low"]}}}}}, {"type": "function", "function": {"name": "get_item", "description": "Fetch one active triage item by its id (fingerprint).", "parameters": {"type": "object", "required": ["id"], "properties": {"id": {"type": "string", "description": "The item id / fingerprint."}}}}}, {"type": "function", "function": {"name": "resolve_person", "description": "Resolve a person's name or handle to a Slack user id. Returns resolved, ambiguous (with candidates), or not_found.", "parameters": {"type": "object", "required": ["name"], "properties": {"name": {"type": "string", "description": "Person name, @handle, or email."}}}}}, {"type": "function", "function": {"name": "propose_calendar_block", "description": "Propose a calendar time block for the user to approve. Nothing is written without approval.", "parameters": {"type": "object", "required": ["title", "duration_minutes", "when"], "properties": {"title": {"type": "string"}, "duration_minutes": {"type": "integer"}, "when": {"type": "string", "description": "e.g. today afternoon, tomorrow 10:00"}, "item_id": {"type": "string", "description": "Optional linked triage item id."}}}}}, {"type": "function", "function": {"name": "list_calendar_events", "description": "List the user's calendar events for a day part.", "parameters": {"type": "object", "required": ["when"], "properties": {"when": {"type": "string", "description": "e.g. today, today afternoon, tomorrow"}}}}}, {"type": "function", "function": {"name": "get_sla_status", "description": "Summarize SLA state across the user's shared inboxes: breached and at-risk conversations.", "parameters": {"type": "object", "required": [], "properties": {}}}}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: slack_channel_message\nCHANNEL: #ai-team (no escalation keywords configured)\nFROM: Priya (Solutions)\nTIME: 10:05 today\nUSER CONTEXT: sole QA, AI team\nMESSAGE: \"@qa customer demo in 2 hours and the dataset upload is broken on their instance — can someone look?\""}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"critical\", \"severity_rationale\": \"Customer-facing breakage with near-term business impact, even though no escalation keyword appears.\", \"tier\": \"P1\", \"owner\": \"me\", \"why_it_matters\": \"A customer-visible failure damages the account directly if not confirmed and routed fast.\", \"next_step\": \"Reproduce the dataset upload failure on the customer instance and report findings to Priya\", \"deadline_suggestion\": \"today\", \"escalation_flag\": true}"}]}
	{"messages": [{"role": "system", "content": "You are UnchaosLM, the conversational assistant of Triage Copilot. You help the user act on their triaged work: tasks, escalations, priorities, people, inboxes, and calendar. When a request needs live data or an action, call the matching tool — extract the arguments from the user's message itself, even when phrased casually. After a tool returns, answer grounded strictly in its result and say plainly what you did (which tool you used and what it found). If a required argument is genuinely absent, ask one short question instead of guessing. Greetings and questions about your abilities need no tool. Never invent items, ids, owners, or links."}, {"role": "user", "content": "are the support queues okay?"}, {"role": "assistant", "tool_calls": [{"function": {"name": "get_sla_status", "arguments": {}}}], "content": ""}, {"role": "tool", "content": "{\"breached\": [{\"conversation\": 4006, \"customer\": \"Dana W.\", \"over_by\": \"35m\"}], \"at_risk\": [{\"conversation\": 5006, \"customer\": \"Grace N.\", \"due_in\": \"25m\"}]}"}, {"role": "assistant", "content": "I ran get_sla_status — 1 conversation(s) already breached (Dana W. over by 35m) and 1 at risk. Answer the breached ones first."}], "tools": [{"type": "function", "function": {"name": "list_items", "description": "List the user's active triage items, optionally filtered by tier (P1-P4) or severity. Returns id, tier, severity, title, owner, next step, deadline, and link.", "parameters": {"type": "object", "required": [], "properties": {"tier": {"type": "string", "enum": ["P1", "P2", "P3", "P4"]}, "severity": {"type": "string", "enum": ["Critical", "High", "Medium", "Low"]}}}}}, {"type": "function", "function": {"name": "get_item", "description": "Fetch one active triage item by its id (fingerprint).", "parameters": {"type": "object", "required": ["id"], "properties": {"id": {"type": "string", "description": "The item id / fingerprint."}}}}}, {"type": "function", "function": {"name": "resolve_person", "description": "Resolve a person's name or handle to a Slack user id. Returns resolved, ambiguous (with candidates), or not_found.", "parameters": {"type": "object", "required": ["name"], "properties": {"name": {"type": "string", "description": "Person name, @handle, or email."}}}}}, {"type": "function", "function": {"name": "propose_calendar_block", "description": "Propose a calendar time block for the user to approve. Nothing is written without approval.", "parameters": {"type": "object", "required": ["title", "duration_minutes", "when"], "properties": {"title": {"type": "string"}, "duration_minutes": {"type": "integer"}, "when": {"type": "string", "description": "e.g. today afternoon, tomorrow 10:00"}, "item_id": {"type": "string", "description": "Optional linked triage item id."}}}}}, {"type": "function", "function": {"name": "list_calendar_events", "description": "List the user's calendar events for a day part.", "parameters": {"type": "object", "required": ["when"], "properties": {"when": {"type": "string", "description": "e.g. today, today afternoon, tomorrow"}}}}}, {"type": "function", "function": {"name": "get_sla_status", "description": "Summarize SLA state across the user's shared inboxes: breached and at-risk conversations.", "parameters": {"type": "object", "required": [], "properties": {}}}}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: slack_dm\nFROM: Neha (CSM)\nTIME: today 10:15\nUSER CONTEXT: product manager, AI team\nMESSAGE: \"can you help me handle Kite Systems? they say the fine-tune dashboard behavior is a dealbreaker for renewal\""}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"critical\", \"severity_rationale\": \"A named account is escalating with churn or launch risk attached \\u2014 direct customer impact.\", \"tier\": \"P1\", \"owner\": \"me\", \"why_it_matters\": \"Kite Systems is at risk and needs a product commitment, not just a support reply.\", \"next_step\": \"Join the Kite Systems escalation call and commit to a fine-tune dashboard plan\", \"deadline_suggestion\": \"today\", \"escalation_flag\": true}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: hiver_sla_alert\nCONVERSATION #9453 \| CUSTOMER: Kevin D. \| SUBJECT: custom ribbon bar issue\nALERT: resolution SLA breached 3 hours ago \| ASSIGNEE: me\nUSER CONTEXT: support agent, medals & ribbons e-commerce store"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"critical\", \"severity_rationale\": \"A resolution SLA has already been breached; every additional minute deepens the miss.\", \"tier\": \"P1\", \"owner\": \"me\", \"why_it_matters\": \"Kevin D.'s conversation is past its resolution commitment and counting.\", \"next_step\": \"Answer conversation #9453 immediately and apologize for the delay\", \"deadline_suggestion\": \"today\", \"escalation_flag\": true}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: slack_channel_message\nCHANNEL: #ai-team\nFROM: Nikhil (dev)\nTIME: today 11:30\nUSER CONTEXT: product manager, AI team\nMESSAGE: \"the PRD doesn't say what happens when the chat memory gets empty input — which behavior is intended?\""}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"medium\", \"severity_rationale\": \"Repeated clarifying questions mean the spec is ambiguous and developers are building on guesses.\", \"tier\": \"P2\", \"owner\": \"me\", \"why_it_matters\": \"An ambiguous chat memory spec turns into rework and delivery slip if not resolved now.\", \"next_step\": \"Answer the chat memory question in thread and update the PRD section\", \"deadline_suggestion\": \"today\", \"escalation_flag\": false}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: hiver_csat_rating\nRATING: 5/5 from Dana W. \| AGENT: Meera\nCOMMENT: \"really appreciated the quick turnaround\"\nUSER CONTEXT: claims support agent, insurance agency"}, {"role": "assistant", "content": "{\"is_actionable\": false, \"action_direction\": \"info_only\", \"severity\": \"low\", \"severity_rationale\": \"A positive rating with praise; no request or follow-up implied.\", \"tier\": \"P4\", \"owner\": \"none\", \"why_it_matters\": \"Nice signal for the team; nothing to act on.\", \"next_step\": \"No action needed\", \"deadline_suggestion\": \"this_week\", \"escalation_flag\": false}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: slack_channel_message\nCHANNEL: #ai-team\nFROM: Divya\nTIME: 12:00 today\nUSER CONTEXT: sole QA, AI team\nMESSAGE: \"reminder that sprint demo notes live in the shared doc\""}, {"role": "assistant", "content": "{\"is_actionable\": false, \"action_direction\": \"info_only\", \"severity\": \"low\", \"severity_rationale\": \"General FYI with no question, request, or risk directed at the user.\", \"tier\": \"P4\", \"owner\": \"none\", \"why_it_matters\": \"Informational chatter; no action exists.\", \"next_step\": \"No action needed\", \"deadline_suggestion\": \"this_week\", \"escalation_flag\": false}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: slack_dm\nFROM: Neha (CSM)\nTIME: today 10:15\nUSER CONTEXT: product manager, AI team\nMESSAGE: \"heads up, Kite Systems raised the eval-runs page issue to their leadership; we need a plan today\""}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"critical\", \"severity_rationale\": \"A named account is escalating with churn or launch risk attached \\u2014 direct customer impact.\", \"tier\": \"P1\", \"owner\": \"me\", \"why_it_matters\": \"Kite Systems is at risk and needs a product commitment, not just a support reply.\", \"next_step\": \"Join the Kite Systems escalation call and commit to a eval-runs page plan\", \"deadline_suggestion\": \"today\", \"escalation_flag\": true}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: linked_ticket\nENG TASK: \"SSO login bug for Orbita\" \| STATUS: open \| ASSIGNEE: Rohit (eng) \| LAST ENG UPDATE: 3 days ago\nCUSTOMER: Orbita has chased twice since\nUSER CONTEXT: support engineer, B2B SaaS"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_them\", \"severity\": \"high\", \"severity_rationale\": \"A customer-impacting bug has no engineering update in three days while the customer chases.\", \"tier\": \"P1\", \"owner\": \"Rohit\", \"why_it_matters\": \"Orbita is waiting on a fix and silence from us reads as neglect.\", \"next_step\": \"Push Rohit for an ETA and send Orbita an honest status update\", \"deadline_suggestion\": \"today\", \"escalation_flag\": true}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: slack_thread\nCHANNEL: #ai-team-bugs\nTHREAD HISTORY:\n[Mon 14:30] me: \"Flagging: prompt playground shows stale data after a re-run. Repro steps in the task. cc @Meera\"\n[Tue 11:20] Meera (dev): \"pushed to staging, should work now\"\nTIME NOW: Wed 08:30\nUSER CONTEXT: sole QA, AI team; I have not replied since Meera's message"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"medium\", \"severity_rationale\": \"The fix is claimed but unverified; the bug stays possibly-live until QA confirms it.\", \"tier\": \"P2\", \"owner\": \"me\", \"why_it_matters\": \"An unverified fix can ship broken; verification is the open action and it is mine.\", \"next_step\": \"Verify Meera's fix for the prompt playground bug and reply on the thread\", \"deadline_suggestion\": \"today\", \"escalation_flag\": false}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: slack_dm\nFROM: Meera (dev)\nTIME: today 10:15\nUSER CONTEXT: senior developer, AI team\nMESSAGE: \"please prioritize my PR — it's the CART-1734 customer fix and the window closes at 2\""}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"high\", \"severity_rationale\": \"A customer-impact fix is blocked solely on this reviewer's approval.\", \"tier\": \"P1\", \"owner\": \"me\", \"why_it_matters\": \"Every hour of review delay extends Kite Systems's outage exposure.\", \"next_step\": \"Review and approve the CART-1734 fix PR this morning\", \"deadline_suggestion\": \"today\", \"escalation_flag\": true}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: hiver_shared_inbox_email\nINBOX: admissions@ \| FROM: Emily A. (customer)\nTIME: 2 days ago 09:15\nMESSAGE: \"the module 4 quiz arrived wrong and I need the corrected one before an exam sitting next week\"\nUSER CONTEXT: support agent, online course provider"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"high\", \"severity_rationale\": \"A customer has a defective or missing module 4 quiz with a hard event date approaching.\", \"tier\": \"P2\", \"owner\": \"me\", \"why_it_matters\": \"Emily A. cannot make an exam sitting unless the module 4 quiz is corrected in time.\", \"next_step\": \"Confirm replacement module 4 quiz and expedited shipping to Emily A.\", \"deadline_suggestion\": \"today\", \"escalation_flag\": false}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: hiver_shared_inbox_email\nINBOX: admissions@ \| FROM: Emily A. (customer)\nTIME: 2 days ago 09:15\nMESSAGE: \"third email about my course login and still no reply. If I don't hear back today I'm disputing the charge\"\nUSER CONTEXT: support agent, online course provider"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"critical\", \"severity_rationale\": \"A repeat-contact customer is threatening a chargeback or cancellation over an unresolved issue.\", \"tier\": \"P1\", \"owner\": \"me\", \"why_it_matters\": \"An escalating course login complaint is at the point of churn and public complaint.\", \"next_step\": \"Reply to Emily A. now with a concrete fix and timeline\", \"deadline_suggestion\": \"today\", \"escalation_flag\": true}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: hiver_shared_inbox_email\nINBOX: claims@ \| FROM: Raj V. (customer)\nTIME: today 10:40\nMESSAGE: \"just confirming I received the claim form — no action needed\"\nUSER CONTEXT: claims support agent, insurance agency"}, {"role": "assistant", "content": "{\"is_actionable\": false, \"action_direction\": \"info_only\", \"severity\": \"low\", \"severity_rationale\": \"A courtesy confirmation with explicit no-action language and no open request.\", \"tier\": \"P4\", \"owner\": \"none\", \"why_it_matters\": \"Positive close-out note; nothing to act on.\", \"next_step\": \"No action needed\", \"deadline_suggestion\": \"this_week\", \"escalation_flag\": false}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: slack_channel_message\nCHANNEL: #general\nFROM: Ayaan (teammate)\nTIME: today 09:20\nUSER CONTEXT: support engineer, B2B SaaS\nMESSAGE: \"urgent: standup moved to 10:30 today, conference room double-booked\""}, {"role": "assistant", "content": "{\"is_actionable\": false, \"action_direction\": \"info_only\", \"severity\": \"low\", \"severity_rationale\": \"Urgent-sounding wording, but it is internal logistics with no impact and no action owed.\", \"tier\": \"P4\", \"owner\": \"none\", \"why_it_matters\": \"Logistics notice to note; nothing to act on.\", \"next_step\": \"No action needed\", \"deadline_suggestion\": \"today\", \"escalation_flag\": false}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: hiver_mention\nCONVERSATION #8054 \| FROM: Neha (teammate) \| TIME: today, 20 minutes ago\nNOTE: \"@me urgent — Helen C. says the beneficiary update failure is blocking their whole team, can you jump in?\"\nUSER CONTEXT: claims support agent, insurance agency"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"critical\", \"severity_rationale\": \"A teammate flags a customer-blocking failure and asks for immediate help.\", \"tier\": \"P1\", \"owner\": \"me\", \"why_it_matters\": \"Helen C.'s team is blocked on the beneficiary update failure right now.\", \"next_step\": \"Join conversation #8054 and triage the beneficiary update failure\", \"deadline_suggestion\": \"today\", \"escalation_flag\": true}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: hiver_shared_inbox_email\nINBOX: admissions@ \| FROM: Dana W. (customer)\nTIME: today 10:40\nMESSAGE: \"received a damaged certificate download; I need a replacement in time for a license renewal deadline\"\nUSER CONTEXT: support agent, online course provider"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"high\", \"severity_rationale\": \"A customer has a defective or missing certificate download with a hard event date approaching.\", \"tier\": \"P2\", \"owner\": \"me\", \"why_it_matters\": \"Dana W. cannot make a license renewal deadline unless the certificate download is corrected in time.\", \"next_step\": \"Confirm replacement certificate download and expedited shipping to Dana W.\", \"deadline_suggestion\": \"today\", \"escalation_flag\": false}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: hiver_shared_inbox_email\nINBOX: orders@ \| FROM: Jenny F. (customer)\nTIME: today 08:10\nMESSAGE: \"my shadow box frame is missing from the delivery — a memorial service is coming up, can you help?\"\nUSER CONTEXT: support agent, medals & ribbons e-commerce store"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"high\", \"severity_rationale\": \"A customer has a defective or missing shadow box frame with a hard event date approaching.\", \"tier\": \"P2\", \"owner\": \"me\", \"why_it_matters\": \"Jenny F. cannot make a memorial service unless the shadow box frame is corrected in time.\", \"next_step\": \"Confirm replacement shadow box frame and expedited shipping to Jenny F.\", \"deadline_suggestion\": \"today\", \"escalation_flag\": false}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: clickup_task\nTASK: \"chat memory regression\" \| STATUS: in-progress \| ASSIGNEE: Sana (dev) \| DUE: none \| TAGS: none\nCOMMENTS:\n[yesterday 16:10] Sana: \"@qa was this on staging or prod data? and which browser?\"\n(no reply from me)\nUSER CONTEXT: sole QA, AI team"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"medium\", \"severity_rationale\": \"The assigned dev is blocked on details only I can provide about the repro.\", \"tier\": \"P2\", \"owner\": \"me\", \"why_it_matters\": \"Sana cannot progress the chat memory fix until I share the repro details.\", \"next_step\": \"Comment the requested repro details on Sana's task\", \"deadline_suggestion\": \"today\", \"escalation_flag\": false}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: clickup_task\nTASK: \"QA: fine-tune dashboard test pass\" \| STATUS: in-progress \| ASSIGNEE: me \| DUE: yesterday \| TAGS: none\nCOMMENTS:\n[2 days ago] me: \"completed the full run, no regressions found\"\nUSER CONTEXT: sole QA, AI team; standup at 10:00; the work is done but the status was never updated"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"low\", \"severity_rationale\": \"Work is done; only the status update is missing, but it misleads the board before standup.\", \"tier\": \"P4\", \"owner\": \"me\", \"why_it_matters\": \"A stale in-progress status makes finished work look late at standup.\", \"next_step\": \"Mark the fine-tune dashboard test task complete in ClickUp before standup\", \"deadline_suggestion\": \"today\", \"escalation_flag\": false}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: slack_thread\nCHANNEL: #ai-team\nTHREAD HISTORY:\n[Yesterday 15:02] Nikhil (dev): \"@qa any blockers from QA side for tomorrow's release?\"\n[Yesterday 15:40] me: \"no blockers, everything on the list is verified\"\n[Yesterday 15:41] Nikhil: \"perfect, thanks!\"\nTIME NOW: today 08:30\nUSER CONTEXT: sole QA, AI team"}, {"role": "assistant", "content": "{\"is_actionable\": false, \"action_direction\": \"info_only\", \"severity\": \"low\", \"severity_rationale\": \"The question was asked and fully answered yesterday; nothing remains open.\", \"tier\": \"P4\", \"owner\": \"none\", \"why_it_matters\": \"The thread is closed; surfacing it would be noise.\", \"next_step\": \"No action needed\", \"deadline_suggestion\": \"this_week\", \"escalation_flag\": false}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: hiver_shared_inbox_email\nINBOX: admissions@ \| FROM: Robert L. (customer)\nTIME: yesterday 17:55\nMESSAGE: \"my proctored exam slot is missing from the delivery — a course expiry date is coming up, can you help?\"\nUSER CONTEXT: support agent, online course provider"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"high\", \"severity_rationale\": \"A customer has a defective or missing proctored exam slot with a hard event date approaching.\", \"tier\": \"P2\", \"owner\": \"me\", \"why_it_matters\": \"Robert L. cannot make a course expiry date unless the proctored exam slot is corrected in time.\", \"next_step\": \"Confirm replacement proctored exam slot and expedited shipping to Robert L.\", \"deadline_suggestion\": \"today\", \"escalation_flag\": false}"}]}
	{"messages": [{"role": "system", "content": "You are UnchaosLM, the conversational assistant of Triage Copilot. You help the user act on their triaged work: tasks, escalations, priorities, people, inboxes, and calendar. When a request needs live data or an action, call the matching tool — extract the arguments from the user's message itself, even when phrased casually. After a tool returns, answer grounded strictly in its result and say plainly what you did (which tool you used and what it found). If a required argument is genuinely absent, ask one short question instead of guessing. Greetings and questions about your abilities need no tool. Never invent items, ids, owners, or links."}, {"role": "user", "content": "what fires are burning"}, {"role": "assistant", "tool_calls": [{"function": {"name": "list_items", "arguments": {"tier": "P1"}}}], "content": ""}], "tools": [{"type": "function", "function": {"name": "list_items", "description": "List the user's active triage items, optionally filtered by tier (P1-P4) or severity. Returns id, tier, severity, title, owner, next step, deadline, and link.", "parameters": {"type": "object", "required": [], "properties": {"tier": {"type": "string", "enum": ["P1", "P2", "P3", "P4"]}, "severity": {"type": "string", "enum": ["Critical", "High", "Medium", "Low"]}}}}}, {"type": "function", "function": {"name": "get_item", "description": "Fetch one active triage item by its id (fingerprint).", "parameters": {"type": "object", "required": ["id"], "properties": {"id": {"type": "string", "description": "The item id / fingerprint."}}}}}, {"type": "function", "function": {"name": "resolve_person", "description": "Resolve a person's name or handle to a Slack user id. Returns resolved, ambiguous (with candidates), or not_found.", "parameters": {"type": "object", "required": ["name"], "properties": {"name": {"type": "string", "description": "Person name, @handle, or email."}}}}}, {"type": "function", "function": {"name": "propose_calendar_block", "description": "Propose a calendar time block for the user to approve. Nothing is written without approval.", "parameters": {"type": "object", "required": ["title", "duration_minutes", "when"], "properties": {"title": {"type": "string"}, "duration_minutes": {"type": "integer"}, "when": {"type": "string", "description": "e.g. today afternoon, tomorrow 10:00"}, "item_id": {"type": "string", "description": "Optional linked triage item id."}}}}}, {"type": "function", "function": {"name": "list_calendar_events", "description": "List the user's calendar events for a day part.", "parameters": {"type": "object", "required": ["when"], "properties": {"when": {"type": "string", "description": "e.g. today, today afternoon, tomorrow"}}}}}, {"type": "function", "function": {"name": "get_sla_status", "description": "Summarize SLA state across the user's shared inboxes: breached and at-risk conversations.", "parameters": {"type": "object", "required": [], "properties": {}}}}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: slack_dm\nFROM: Rahul (Support Engineer)\nTIME: 11:40 today\nUSER CONTEXT: sole QA, AI team\nMESSAGE: \"P0 from support: customer completely stuck on the dataset upload. free for 15 min?\""}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"critical\", \"severity_rationale\": \"Support engineer reports a customer actively blocked and needs immediate repro help.\", \"tier\": \"P1\", \"owner\": \"me\", \"why_it_matters\": \"A customer is blocked right now; the repro is the gate to the fix.\", \"next_step\": \"Huddle with Rahul now to reproduce the dataset upload issue\", \"deadline_suggestion\": \"today\", \"escalation_flag\": true}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: hiver_shared_inbox_email\nINBOX: claims@ \| FROM: Jenny F. (customer)\nTIME: today 09:25\nMESSAGE: \"received a damaged claim status; I need a replacement in time for a claim deadline\"\nUSER CONTEXT: claims support agent, insurance agency"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"high\", \"severity_rationale\": \"A customer has a defective or missing claim status with a hard event date approaching.\", \"tier\": \"P2\", \"owner\": \"me\", \"why_it_matters\": \"Jenny F. cannot make a claim deadline unless the claim status is corrected in time.\", \"next_step\": \"Confirm replacement claim status and expedited shipping to Jenny F.\", \"deadline_suggestion\": \"today\", \"escalation_flag\": false}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: health_score_alert\nACCOUNT: Northwind \| SCORE: dropped 82 -> 61 this month\nDRIVER: logins down 40%, CSV export usage near zero\nOWNER: me\nUSER CONTEXT: customer success manager"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"high\", \"severity_rationale\": \"A sharp health-score drop with collapsing usage is an early churn signal.\", \"tier\": \"P1\", \"owner\": \"me\", \"why_it_matters\": \"Northwind's disengagement is measurable now; waiting turns a dip into a cancellation.\", \"next_step\": \"Investigate Northwind's usage drop and schedule a check-in call\", \"deadline_suggestion\": \"tomorrow\", \"escalation_flag\": true}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: slack_thread\nCHANNEL: #ai-team-bugs\nTHREAD HISTORY:\n[Mon 14:30] me: \"Flagging: webhook alerts shows stale data after a re-run. Repro steps in the task. cc @Tanvi\"\n[Tue 11:20] Tanvi (dev): \"should be good now\"\nTIME NOW: Wed 08:30\nUSER CONTEXT: sole QA, AI team; I have not replied since Tanvi's message"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"medium\", \"severity_rationale\": \"The fix is claimed but unverified; the bug stays possibly-live until QA confirms it.\", \"tier\": \"P2\", \"owner\": \"me\", \"why_it_matters\": \"An unverified fix can ship broken; verification is the open action and it is mine.\", \"next_step\": \"Verify Tanvi's fix for the webhook alerts bug and reply on the thread\", \"deadline_suggestion\": \"today\", \"escalation_flag\": false}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: hiver_sla_alert\nCONVERSATION #1431 \| CUSTOMER: Helen C. \| SUBJECT: automation rule issue\nALERT: first-response SLA due in 45 minutes \| ASSIGNEE: me\nUSER CONTEXT: support engineer, B2B SaaS (shared inbox product)"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"high\", \"severity_rationale\": \"The first-response SLA window closes in 45 minutes; a breach is avoidable but imminent.\", \"tier\": \"P2\", \"owner\": \"me\", \"why_it_matters\": \"Responding to Helen C. within 45 minutes keeps the SLA record clean.\", \"next_step\": \"Send a first response on conversation #1431 within 45 minutes\", \"deadline_suggestion\": \"today\", \"escalation_flag\": false}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: slack_dm\nFROM: Suresh (hiring manager)\nTIME: yesterday 15:10\nUSER CONTEXT: talent acquisition recruiter\nMESSAGE: \"please schedule S. Iyer's final round soon, we don't want to lose momentum\""}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"medium\", \"severity_rationale\": \"A hiring manager has asked for an interview loop to be scheduled this week.\", \"tier\": \"P2\", \"owner\": \"me\", \"why_it_matters\": \"Scheduling lag is the most common self-inflicted pipeline delay.\", \"next_step\": \"Send S. Iyer's panel invites and confirm the loop this week\", \"deadline_suggestion\": \"this_week\", \"escalation_flag\": false}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: hiver_mention\nCONVERSATION #5712 \| FROM: Kavya (teammate) \| TIME: today 13:30\nNOTE: \"@me can you take this one? Tom H. is asking about the claim status and you handled their last case\"\nUSER CONTEXT: claims support agent, insurance agency"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"medium\", \"severity_rationale\": \"A teammate has handed over a customer conversation and is waiting for pickup.\", \"tier\": \"P2\", \"owner\": \"me\", \"why_it_matters\": \"Tom H.'s claim status question sits unowned until the handover is accepted.\", \"next_step\": \"Accept conversation #5712 and reply to Tom H.\", \"deadline_suggestion\": \"today\", \"escalation_flag\": false}"}]}
	{"messages": [{"role": "system", "content": "You are UnchaosLM, the conversational assistant of Triage Copilot. You help the user act on their triaged work: tasks, escalations, priorities, people, inboxes, and calendar. When a request needs live data or an action, call the matching tool — extract the arguments from the user's message itself, even when phrased casually. After a tool returns, answer grounded strictly in its result and say plainly what you did (which tool you used and what it found). If a required argument is genuinely absent, ask one short question instead of guessing. Greetings and questions about your abilities need no tool. Never invent items, ids, owners, or links."}, {"role": "user", "content": "what's zubin's handle?"}, {"role": "assistant", "tool_calls": [{"function": {"name": "resolve_person", "arguments": {"name": "zubin"}}}], "content": ""}, {"role": "tool", "content": "{\"status\": \"not_found\"}"}, {"role": "assistant", "content": "I checked with resolve_person — no Slack user matching \"zubin\". Could you give the full name or email?"}], "tools": [{"type": "function", "function": {"name": "list_items", "description": "List the user's active triage items, optionally filtered by tier (P1-P4) or severity. Returns id, tier, severity, title, owner, next step, deadline, and link.", "parameters": {"type": "object", "required": [], "properties": {"tier": {"type": "string", "enum": ["P1", "P2", "P3", "P4"]}, "severity": {"type": "string", "enum": ["Critical", "High", "Medium", "Low"]}}}}}, {"type": "function", "function": {"name": "get_item", "description": "Fetch one active triage item by its id (fingerprint).", "parameters": {"type": "object", "required": ["id"], "properties": {"id": {"type": "string", "description": "The item id / fingerprint."}}}}}, {"type": "function", "function": {"name": "resolve_person", "description": "Resolve a person's name or handle to a Slack user id. Returns resolved, ambiguous (with candidates), or not_found.", "parameters": {"type": "object", "required": ["name"], "properties": {"name": {"type": "string", "description": "Person name, @handle, or email."}}}}}, {"type": "function", "function": {"name": "propose_calendar_block", "description": "Propose a calendar time block for the user to approve. Nothing is written without approval.", "parameters": {"type": "object", "required": ["title", "duration_minutes", "when"], "properties": {"title": {"type": "string"}, "duration_minutes": {"type": "integer"}, "when": {"type": "string", "description": "e.g. today afternoon, tomorrow 10:00"}, "item_id": {"type": "string", "description": "Optional linked triage item id."}}}}}, {"type": "function", "function": {"name": "list_calendar_events", "description": "List the user's calendar events for a day part.", "parameters": {"type": "object", "required": ["when"], "properties": {"when": {"type": "string", "description": "e.g. today, today afternoon, tomorrow"}}}}}, {"type": "function", "function": {"name": "get_sla_status", "description": "Summarize SLA state across the user's shared inboxes: breached and at-risk conversations.", "parameters": {"type": "object", "required": [], "properties": {}}}}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: hiver_sla_alert\nCONVERSATION #7026 \| CUSTOMER: Colin B. \| SUBJECT: freight quote issue\nALERT: resolution SLA breached 3 hours ago \| ASSIGNEE: me\nUSER CONTEXT: support agent, logistics & shipping"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"critical\", \"severity_rationale\": \"A resolution SLA has already been breached; every additional minute deepens the miss.\", \"tier\": \"P1\", \"owner\": \"me\", \"why_it_matters\": \"Colin B.'s conversation is past its resolution commitment and counting.\", \"next_step\": \"Answer conversation #7026 immediately and apologize for the delay\", \"deadline_suggestion\": \"today\", \"escalation_flag\": true}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: hiver_csat_rating\nRATING: 2/5 from Marcus J. \| AGENT: Nikhil\nCOMMENT: \"took too long and the course extension answer didn't solve my problem\"\nUSER CONTEXT: support agent, online course provider"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"medium\", \"severity_rationale\": \"A low rating with a concrete complaint is recoverable if followed up quickly.\", \"tier\": \"P2\", \"owner\": \"me\", \"why_it_matters\": \"Marcus J.'s course extension problem is still unsolved despite a closed conversation.\", \"next_step\": \"Reopen Marcus J.'s conversation and resolve the course extension problem properly\", \"deadline_suggestion\": \"tomorrow\", \"escalation_flag\": false}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: hiver_assignment\nCONVERSATION #3545 \| CUSTOMER: Miguel T. \| SUBJECT: lapel pin order problem\nSTATUS: open, assigned to me 5 days ago, no reply sent yet\nUSER CONTEXT: support agent, medals & ribbons e-commerce store"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"high\", \"severity_rationale\": \"An assigned conversation has waited 5 days without a single reply.\", \"tier\": \"P2\", \"owner\": \"me\", \"why_it_matters\": \"Miguel T. has had silence for 5 days on an open lapel pin order problem.\", \"next_step\": \"Send Miguel T. a first reply on the lapel pin order problem\", \"deadline_suggestion\": \"today\", \"escalation_flag\": false}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: hiver_sla_alert\nCONVERSATION #9056 \| CUSTOMER: Omar S. \| SUBJECT: CE credit record issue\nALERT: first-response SLA breached 3 hours ago \| ASSIGNEE: me\nUSER CONTEXT: support agent, online course provider"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"critical\", \"severity_rationale\": \"A first-response SLA has already been breached; every additional minute deepens the miss.\", \"tier\": \"P1\", \"owner\": \"me\", \"why_it_matters\": \"Omar S.'s conversation is past its first-response commitment and counting.\", \"next_step\": \"Answer conversation #9056 immediately and apologize for the delay\", \"deadline_suggestion\": \"today\", \"escalation_flag\": true}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: hiver_shared_inbox_email\nINBOX: support@ \| FROM: Robert L. (customer)\nTIME: today 13:30\nMESSAGE: \"this is unacceptable — the user seat issue has been open for a week. I want a manager and a refund\"\nUSER CONTEXT: support engineer, B2B SaaS (shared inbox product)"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"critical\", \"severity_rationale\": \"A repeat-contact customer is threatening a chargeback or cancellation over an unresolved issue.\", \"tier\": \"P1\", \"owner\": \"me\", \"why_it_matters\": \"An escalating user seat complaint is at the point of churn and public complaint.\", \"next_step\": \"Reply to Robert L. now with a concrete fix and timeline\", \"deadline_suggestion\": \"today\", \"escalation_flag\": true}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: clickup_task\nTASK: \"Verify fix: chat memory (CART-1473)\" \| STATUS: qa-ready \| ASSIGNEE: me \| DUE: today \| TAGS: CART, customer-impact\nCOMMENTS:\n[yesterday 18:05] Sana (dev): \"deployed to staging, ready for QA — customer is waiting on this in tomorrow's release\"\nUSER CONTEXT: sole QA, AI team; deployment window closes early afternoon"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"high\", \"severity_rationale\": \"CART-tagged customer-impact fix must be verified before the deployment window to make the release.\", \"tier\": \"P1\", \"owner\": \"me\", \"why_it_matters\": \"A waiting customer misses the release if verification slips past the deployment window.\", \"next_step\": \"Verify the chat memory fix on staging before the deployment window closes\", \"deadline_suggestion\": \"today\", \"escalation_flag\": true}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: slack_dm\nFROM: Suresh (hiring manager)\nTIME: yesterday 15:10\nUSER CONTEXT: talent acquisition recruiter\nMESSAGE: \"M. Kulkarni passed the screen — over to you for panel scheduling\""}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"medium\", \"severity_rationale\": \"A hiring manager has asked for an interview loop to be scheduled this week.\", \"tier\": \"P2\", \"owner\": \"me\", \"why_it_matters\": \"Scheduling lag is the most common self-inflicted pipeline delay.\", \"next_step\": \"Send M. Kulkarni's panel invites and confirm the loop this week\", \"deadline_suggestion\": \"this_week\", \"escalation_flag\": false}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: offboarding_checklist\nEMPLOYEE: Ritika \| LAST DAY: last Friday\nSTATUS: 2 of 6 access revocations UNCONFIRMED (VPN, admin console)\nUSER CONTEXT: IT administrator"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"high\", \"severity_rationale\": \"A departed employee still has unconfirmed access to two systems \\u2014 an open security hole.\", \"tier\": \"P1\", \"owner\": \"me\", \"why_it_matters\": \"Lingering access after departure is a textbook breach vector and audit finding.\", \"next_step\": \"Confirm and close all access revocations for Ritika today\", \"deadline_suggestion\": \"today\", \"escalation_flag\": true}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: hiver_mention\nCONVERSATION #6225 \| MY NOTE 4 days ago: \"@Tara can you check the ribbon rack logs for Dana W.?\"\nSTATUS: no response from teammate yet\nUSER CONTEXT: support agent, medals & ribbons e-commerce store"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_them\", \"severity\": \"medium\", \"severity_rationale\": \"An internal ask has sat unanswered for 4 days with no committed date.\", \"tier\": \"P3\", \"owner\": \"Tara\", \"why_it_matters\": \"Dana W.'s case cannot progress until Tara checks the ribbon rack logs.\", \"next_step\": \"Nudge Tara on the ribbon rack logs for conversation #6225\", \"deadline_suggestion\": \"this_week\", \"escalation_flag\": false}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: hiver_shared_inbox_email\nINBOX: admissions@ \| FROM: Tom H. (customer)\nTIME: today 09:25\nMESSAGE: \"whenever you get a chance: can I change my enrollment confirmation after ordering?\"\nUSER CONTEXT: support agent, online course provider"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"low\", \"severity_rationale\": \"A routine how-to question with no urgency signals or deadline pressure.\", \"tier\": \"P4\", \"owner\": \"me\", \"why_it_matters\": \"A simple answer keeps Tom H. moving; nothing depends on it today.\", \"next_step\": \"Send Tom H. a short how-to reply on the enrollment confirmation\", \"deadline_suggestion\": \"this_week\", \"escalation_flag\": false}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: slack_dm\nFROM: Vikram (dev)\nTIME: today 10:15\nUSER CONTEXT: product manager, AI team\nMESSAGE: \"whichever you pick is fine, but we need the webhook alerts decision today to stay on track\""}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"high\", \"severity_rationale\": \"The team is explicitly blocked on a decision only the PM can make, with near-term deadlines.\", \"tier\": \"P2\", \"owner\": \"me\", \"why_it_matters\": \"Every day the webhook alerts decision waits, the whole team idles or builds the wrong thing.\", \"next_step\": \"Make the webhook alerts decision and post it to the team today\", \"deadline_suggestion\": \"today\", \"escalation_flag\": false}"}]}
	{"messages": [{"role": "system", "content": "You are UnchaosLM, the conversational assistant of Triage Copilot. You help the user act on their triaged work: tasks, escalations, priorities, people, inboxes, and calendar. When a request needs live data or an action, call the matching tool — extract the arguments from the user's message itself, even when phrased casually. After a tool returns, answer grounded strictly in its result and say plainly what you did (which tool you used and what it found). If a required argument is genuinely absent, ask one short question instead of guessing. Greetings and questions about your abilities need no tool. Never invent items, ids, owners, or links."}, {"role": "user", "content": "anything low priority I can batch later, please?"}, {"role": "assistant", "tool_calls": [{"function": {"name": "list_items", "arguments": {"tier": "P4"}}}], "content": ""}, {"role": "tool", "content": "{\"items\": [{\"id\": \"cu021\", \"tier\": \"P4\", \"severity\": \"Low\", \"title\": \"Refactor billing page components\", \"owner\": \"me\", \"next\": \"Review the refactor PR when free\", \"by\": \"this week\", \"link\": \"https://app.clickup.com/t/cu021\", \"source\": \"clickup\"}]}"}, {"role": "assistant", "content": "I pulled your P4 triage items with list_items — 1 need you:\n• [P4/Low] Refactor billing page components — owner me, next: Review the refactor PR when free (due this week)\nStart with Refactor billing page components: review the refactor pr when free."}], "tools": [{"type": "function", "function": {"name": "list_items", "description": "List the user's active triage items, optionally filtered by tier (P1-P4) or severity. Returns id, tier, severity, title, owner, next step, deadline, and link.", "parameters": {"type": "object", "required": [], "properties": {"tier": {"type": "string", "enum": ["P1", "P2", "P3", "P4"]}, "severity": {"type": "string", "enum": ["Critical", "High", "Medium", "Low"]}}}}}, {"type": "function", "function": {"name": "get_item", "description": "Fetch one active triage item by its id (fingerprint).", "parameters": {"type": "object", "required": ["id"], "properties": {"id": {"type": "string", "description": "The item id / fingerprint."}}}}}, {"type": "function", "function": {"name": "resolve_person", "description": "Resolve a person's name or handle to a Slack user id. Returns resolved, ambiguous (with candidates), or not_found.", "parameters": {"type": "object", "required": ["name"], "properties": {"name": {"type": "string", "description": "Person name, @handle, or email."}}}}}, {"type": "function", "function": {"name": "propose_calendar_block", "description": "Propose a calendar time block for the user to approve. Nothing is written without approval.", "parameters": {"type": "object", "required": ["title", "duration_minutes", "when"], "properties": {"title": {"type": "string"}, "duration_minutes": {"type": "integer"}, "when": {"type": "string", "description": "e.g. today afternoon, tomorrow 10:00"}, "item_id": {"type": "string", "description": "Optional linked triage item id."}}}}}, {"type": "function", "function": {"name": "list_calendar_events", "description": "List the user's calendar events for a day part.", "parameters": {"type": "object", "required": ["when"], "properties": {"when": {"type": "string", "description": "e.g. today, today afternoon, tomorrow"}}}}}, {"type": "function", "function": {"name": "get_sla_status", "description": "Summarize SLA state across the user's shared inboxes: breached and at-risk conversations.", "parameters": {"type": "object", "required": [], "properties": {}}}}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: hiver_mention\nCONVERSATION #6703 \| FROM: Sameer (teammate) \| TIME: today, 20 minutes ago\nNOTE: \"@me urgent — Frank O. says the lapel pin order failure is blocking their whole team, can you jump in?\"\nUSER CONTEXT: support agent, medals & ribbons e-commerce store"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"critical\", \"severity_rationale\": \"A teammate flags a customer-blocking failure and asks for immediate help.\", \"tier\": \"P1\", \"owner\": \"me\", \"why_it_matters\": \"Frank O.'s team is blocked on the lapel pin order failure right now.\", \"next_step\": \"Join conversation #6703 and triage the lapel pin order failure\", \"deadline_suggestion\": \"today\", \"escalation_flag\": true}"}]}

	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: hiver_shared_inbox_email\nINBOX: support@ \| FROM: Colin B. (customer)\nTIME: 2 days ago 09:15\nMESSAGE: \"my analytics report is missing from the delivery — a migration cutover is coming up, can you help?\"\nUSER CONTEXT: support engineer, B2B SaaS (shared inbox product)"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"high\", \"severity_rationale\": \"A customer has a defective or missing analytics report with a hard event date approaching.\", \"tier\": \"P2\", \"owner\": \"me\", \"why_it_matters\": \"Colin B. cannot make a migration cutover unless the analytics report is corrected in time.\", \"next_step\": \"Confirm replacement analytics report and expedited shipping to Colin B.\", \"deadline_suggestion\": \"today\", \"escalation_flag\": false}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: hiver_escalation_tag\nCONVERSATION #3397 \| TAGGED: Escalation by Dev \| ACCOUNT: Northgate Supply\nREASON: customer CEO emailed directly about the unresolved exam voucher issue\nTIME: today 08:10\nUSER CONTEXT: support agent, online course provider"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"critical\", \"severity_rationale\": \"A teammate has formally tagged the conversation as an escalation with customer impact.\", \"tier\": \"P1\", \"owner\": \"me\", \"why_it_matters\": \"Northgate Supply's escalation is live and visible; delay compounds the damage.\", \"next_step\": \"Own conversation #3397 and respond to the escalation now\", \"deadline_suggestion\": \"today\", \"escalation_flag\": true}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: slack_dm\nFROM: Tanvi (dev)\nTIME: today 10:15\nUSER CONTEXT: senior developer, AI team\nMESSAGE: \"still need your call on the recipe editor error-handling approach before I continue\""}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"high\", \"severity_rationale\": \"A teammate is explicitly blocked waiting on this developer's input.\", \"tier\": \"P2\", \"owner\": \"me\", \"why_it_matters\": \"Tanvi's work is idle until I answer the recipe editor question.\", \"next_step\": \"Answer Tanvi's recipe editor question so their branch can move\", \"deadline_suggestion\": \"today\", \"escalation_flag\": false}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: hiver_shared_inbox_email\nINBOX: shipping@ \| FROM: Omar S. (customer)\nTIME: today 11:05\nMESSAGE: \"the return pickup arrived wrong and I need the corrected one before a trade show next week\"\nUSER CONTEXT: support agent, logistics & shipping"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"high\", \"severity_rationale\": \"A customer has a defective or missing return pickup with a hard event date approaching.\", \"tier\": \"P2\", \"owner\": \"me\", \"why_it_matters\": \"Omar S. cannot make a trade show unless the return pickup is corrected in time.\", \"next_step\": \"Confirm replacement return pickup and expedited shipping to Omar S.\", \"deadline_suggestion\": \"today\", \"escalation_flag\": false}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: support_ticket\nTICKET #1710 \| ACCOUNT: Helios Labs \| PRIORITY: urgent \| SLA: first response due in 40 minutes \| ASSIGNEE: me \| STATUS: new\nSUBJECT: \"the eval-runs page stopped syncing and our team is blocked\"\nUSER CONTEXT: support engineer, B2B SaaS"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"critical\", \"severity_rationale\": \"An urgent ticket is 40 minutes from an SLA breach with no first response sent.\", \"tier\": \"P1\", \"owner\": \"me\", \"why_it_matters\": \"An SLA breach on Helios Labs is a contractual failure, not just a slow reply.\", \"next_step\": \"Send Helios Labs a first response on ticket #1710 before the SLA expires\", \"deadline_suggestion\": \"today\", \"escalation_flag\": true}"}]}
	{"messages": [{"role": "system", "content": "You are UnchaosLM, the conversational assistant of Triage Copilot. You help the user act on their triaged work: tasks, escalations, priorities, people, inboxes, and calendar. When a request needs live data or an action, call the matching tool — extract the arguments from the user's message itself, even when phrased casually. After a tool returns, answer grounded strictly in its result and say plainly what you did (which tool you used and what it found). If a required argument is genuinely absent, ask one short question instead of guessing. Greetings and questions about your abilities need no tool. Never invent items, ids, owners, or links."}, {"role": "user", "content": "schedule an hour tomorrow for the CART fix"}, {"role": "assistant", "tool_calls": [{"function": {"name": "propose_calendar_block", "arguments": {"title": "Fix CART checkout total mismatch", "duration_minutes": 60, "when": "tomorrow", "item_id": "cu003"}}}], "content": ""}], "tools": [{"type": "function", "function": {"name": "list_items", "description": "List the user's active triage items, optionally filtered by tier (P1-P4) or severity. Returns id, tier, severity, title, owner, next step, deadline, and link.", "parameters": {"type": "object", "required": [], "properties": {"tier": {"type": "string", "enum": ["P1", "P2", "P3", "P4"]}, "severity": {"type": "string", "enum": ["Critical", "High", "Medium", "Low"]}}}}}, {"type": "function", "function": {"name": "get_item", "description": "Fetch one active triage item by its id (fingerprint).", "parameters": {"type": "object", "required": ["id"], "properties": {"id": {"type": "string", "description": "The item id / fingerprint."}}}}}, {"type": "function", "function": {"name": "resolve_person", "description": "Resolve a person's name or handle to a Slack user id. Returns resolved, ambiguous (with candidates), or not_found.", "parameters": {"type": "object", "required": ["name"], "properties": {"name": {"type": "string", "description": "Person name, @handle, or email."}}}}}, {"type": "function", "function": {"name": "propose_calendar_block", "description": "Propose a calendar time block for the user to approve. Nothing is written without approval.", "parameters": {"type": "object", "required": ["title", "duration_minutes", "when"], "properties": {"title": {"type": "string"}, "duration_minutes": {"type": "integer"}, "when": {"type": "string", "description": "e.g. today afternoon, tomorrow 10:00"}, "item_id": {"type": "string", "description": "Optional linked triage item id."}}}}}, {"type": "function", "function": {"name": "list_calendar_events", "description": "List the user's calendar events for a day part.", "parameters": {"type": "object", "required": ["when"], "properties": {"when": {"type": "string", "description": "e.g. today, today afternoon, tomorrow"}}}}}, {"type": "function", "function": {"name": "get_sla_status", "description": "Summarize SLA state across the user's shared inboxes: breached and at-risk conversations.", "parameters": {"type": "object", "required": [], "properties": {}}}}]}
	{"messages": [{"role": "system", "content": "You are UnchaosLM, the conversational assistant of Triage Copilot. You help the user act on their triaged work: tasks, escalations, priorities, people, inboxes, and calendar. When a request needs live data or an action, call the matching tool — extract the arguments from the user's message itself, even when phrased casually. After a tool returns, answer grounded strictly in its result and say plainly what you did (which tool you used and what it found). If a required argument is genuinely absent, ask one short question instead of guessing. Greetings and questions about your abilities need no tool. Never invent items, ids, owners, or links."}, {"role": "user", "content": "what's on my plate today"}, {"role": "assistant", "tool_calls": [{"function": {"name": "list_items", "arguments": {}}}], "content": ""}], "tools": [{"type": "function", "function": {"name": "list_items", "description": "List the user's active triage items, optionally filtered by tier (P1-P4) or severity. Returns id, tier, severity, title, owner, next step, deadline, and link.", "parameters": {"type": "object", "required": [], "properties": {"tier": {"type": "string", "enum": ["P1", "P2", "P3", "P4"]}, "severity": {"type": "string", "enum": ["Critical", "High", "Medium", "Low"]}}}}}, {"type": "function", "function": {"name": "get_item", "description": "Fetch one active triage item by its id (fingerprint).", "parameters": {"type": "object", "required": ["id"], "properties": {"id": {"type": "string", "description": "The item id / fingerprint."}}}}}, {"type": "function", "function": {"name": "resolve_person", "description": "Resolve a person's name or handle to a Slack user id. Returns resolved, ambiguous (with candidates), or not_found.", "parameters": {"type": "object", "required": ["name"], "properties": {"name": {"type": "string", "description": "Person name, @handle, or email."}}}}}, {"type": "function", "function": {"name": "propose_calendar_block", "description": "Propose a calendar time block for the user to approve. Nothing is written without approval.", "parameters": {"type": "object", "required": ["title", "duration_minutes", "when"], "properties": {"title": {"type": "string"}, "duration_minutes": {"type": "integer"}, "when": {"type": "string", "description": "e.g. today afternoon, tomorrow 10:00"}, "item_id": {"type": "string", "description": "Optional linked triage item id."}}}}}, {"type": "function", "function": {"name": "list_calendar_events", "description": "List the user's calendar events for a day part.", "parameters": {"type": "object", "required": ["when"], "properties": {"when": {"type": "string", "description": "e.g. today, today afternoon, tomorrow"}}}}}, {"type": "function", "function": {"name": "get_sla_status", "description": "Summarize SLA state across the user's shared inboxes: breached and at-risk conversations.", "parameters": {"type": "object", "required": [], "properties": {}}}}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: clickup_task\nTASK: \"QA: CSV export verification\" \| STATUS: blocked \| ASSIGNEE: me \| DUE: none \| TAGS: none\nCOMMENTS:\n[3 days ago] me: \"blocked: staging build for this is broken, waiting on Nikhil\"\n(no update since)\nUSER CONTEXT: sole QA, AI team"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_them\", \"severity\": \"medium\", \"severity_rationale\": \"The task has been blocked for three days with no date and no movement from the blocker.\", \"tier\": \"P3\", \"owner\": \"Nikhil\", \"why_it_matters\": \"My CSV export verification cannot start until Nikhil unblocks it, and no ETA exists.\", \"next_step\": \"Ping Nikhil for an ETA on unblocking the CSV export verification\", \"deadline_suggestion\": \"tomorrow\", \"escalation_flag\": false}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: hiver_assignment\nCONVERSATION #1975 \| CUSTOMER: Omar S. \| SUBJECT: ribbon rack replacement\nSTATUS: pending — waiting on supplier confirmation, no ETA on their side\nUSER CONTEXT: support agent, medals & ribbons e-commerce store"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_them\", \"severity\": \"medium\", \"severity_rationale\": \"The case is blocked on a third party that has given no timeline at all.\", \"tier\": \"P3\", \"owner\": \"supplier\", \"why_it_matters\": \"Omar S.'s ribbon rack replacement is open-ended until the supplier commits to a date.\", \"next_step\": \"Chase the supplier for an ETA on Omar S.'s ribbon rack\", \"deadline_suggestion\": \"this_week\", \"escalation_flag\": false}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: hiver_shared_inbox_email\nINBOX: claims@ \| FROM: Emily A. (customer)\nTIME: yesterday 15:20\nMESSAGE: \"whenever you get a chance: can I change my coverage letter after ordering?\"\nUSER CONTEXT: claims support agent, insurance agency"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"low\", \"severity_rationale\": \"A routine how-to question with no urgency signals or deadline pressure.\", \"tier\": \"P4\", \"owner\": \"me\", \"why_it_matters\": \"A simple answer keeps Emily A. moving; nothing depends on it today.\", \"next_step\": \"Send Emily A. a short how-to reply on the coverage letter\", \"deadline_suggestion\": \"this_week\", \"escalation_flag\": false}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: hiver_assignment\nCONVERSATION #4776 \| CUSTOMER: Raj V. \| SUBJECT: analytics report replacement\nSTATUS: pending — waiting on supplier confirmation, no ETA on their side\nUSER CONTEXT: support engineer, B2B SaaS (shared inbox product)"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_them\", \"severity\": \"medium\", \"severity_rationale\": \"The case is blocked on a third party that has given no timeline at all.\", \"tier\": \"P3\", \"owner\": \"supplier\", \"why_it_matters\": \"Raj V.'s analytics report replacement is open-ended until the supplier commits to a date.\", \"next_step\": \"Chase the supplier for an ETA on Raj V.'s analytics report\", \"deadline_suggestion\": \"this_week\", \"escalation_flag\": false}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: hiver_shared_inbox_email\nINBOX: claims@ \| FROM: Colin B. (customer)\nTIME: today 11:05\nMESSAGE: \"no rush, but could you explain how the proof of insurance works?\"\nUSER CONTEXT: claims support agent, insurance agency"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"low\", \"severity_rationale\": \"A routine how-to question with no urgency signals or deadline pressure.\", \"tier\": \"P4\", \"owner\": \"me\", \"why_it_matters\": \"A simple answer keeps Colin B. moving; nothing depends on it today.\", \"next_step\": \"Send Colin B. a short how-to reply on the proof of insurance\", \"deadline_suggestion\": \"this_week\", \"escalation_flag\": false}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: hiver_mention\nCONVERSATION #2711 \| FROM: Tara (teammate) \| TIME: yesterday 15:20\nNOTE: \"@me can you take this one? Helen C. is asking about the user seat and you handled their last case\"\nUSER CONTEXT: support engineer, B2B SaaS (shared inbox product)"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"medium\", \"severity_rationale\": \"A teammate has handed over a customer conversation and is waiting for pickup.\", \"tier\": \"P2\", \"owner\": \"me\", \"why_it_matters\": \"Helen C.'s user seat question sits unowned until the handover is accepted.\", \"next_step\": \"Accept conversation #2711 and reply to Helen C.\", \"deadline_suggestion\": \"today\", \"escalation_flag\": false}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: slack_channel_message\nCHANNEL: #general\nFROM: Ira (teammate)\nTIME: today 09:20\nUSER CONTEXT: HR business partner\nMESSAGE: \"urgent-ish: laptop refresh forms due Friday, admin keeps pinging me\""}, {"role": "assistant", "content": "{\"is_actionable\": false, \"action_direction\": \"info_only\", \"severity\": \"low\", \"severity_rationale\": \"Urgent-sounding wording, but it is internal logistics with no impact and no action owed.\", \"tier\": \"P4\", \"owner\": \"none\", \"why_it_matters\": \"Logistics notice to note; nothing to act on.\", \"next_step\": \"No action needed\", \"deadline_suggestion\": \"today\", \"escalation_flag\": false}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: hiver_assignment\nCONVERSATION #7631 \| CUSTOMER: Priyanka R. \| SUBJECT: engraved nameplate replacement\nSTATUS: pending — waiting on supplier confirmation, no ETA on their side\nUSER CONTEXT: support agent, medals & ribbons e-commerce store"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_them\", \"severity\": \"medium\", \"severity_rationale\": \"The case is blocked on a third party that has given no timeline at all.\", \"tier\": \"P3\", \"owner\": \"supplier\", \"why_it_matters\": \"Priyanka R.'s engraved nameplate replacement is open-ended until the supplier commits to a date.\", \"next_step\": \"Chase the supplier for an ETA on Priyanka R.'s engraved nameplate\", \"deadline_suggestion\": \"this_week\", \"escalation_flag\": false}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: crm_alert\nACCOUNT: Northwind \| RENEWAL: in 30 days \| ARR: significant\nOPEN ESCALATION: sev-2 on the prompt playground, 8 days old\nOWNER: me\nUSER CONTEXT: customer success manager"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"critical\", \"severity_rationale\": \"A renewal-window account is carrying an open week-old escalation \\u2014 churn risk is compounding.\", \"tier\": \"P1\", \"owner\": \"me\", \"why_it_matters\": \"Northwind decides on renewal soon and the unresolved prompt playground escalation is the story they'll remember.\", \"next_step\": \"Call Northwind's champion today and align on an escalation resolution plan\", \"deadline_suggestion\": \"today\", \"escalation_flag\": true}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: slack_channel_message\nCHANNEL: #general\nFROM: Divya (teammate)\nTIME: today 09:20\nUSER CONTEXT: product manager, AI team\nMESSAGE: \"urgent: standup moved to 10:30 today, conference room double-booked\""}, {"role": "assistant", "content": "{\"is_actionable\": false, \"action_direction\": \"info_only\", \"severity\": \"low\", \"severity_rationale\": \"Urgent-sounding wording, but it is internal logistics with no impact and no action owed.\", \"tier\": \"P4\", \"owner\": \"none\", \"why_it_matters\": \"Logistics notice to note; nothing to act on.\", \"next_step\": \"No action needed\", \"deadline_suggestion\": \"today\", \"escalation_flag\": false}"}]}
	{"messages": [{"role": "system", "content": "You are UnchaosLM, the conversational assistant of Triage Copilot. You help the user act on their triaged work: tasks, escalations, priorities, people, inboxes, and calendar. When a request needs live data or an action, call the matching tool — extract the arguments from the user's message itself, even when phrased casually. After a tool returns, answer grounded strictly in its result and say plainly what you did (which tool you used and what it found). If a required argument is genuinely absent, ask one short question instead of guessing. Greetings and questions about your abilities need no tool. Never invent items, ids, owners, or links."}, {"role": "user", "content": "morning — anything I should look at first?"}, {"role": "assistant", "tool_calls": [{"function": {"name": "list_items", "arguments": {}}}], "content": ""}, {"role": "tool", "content": "{\"items\": [{\"id\": \"cu004\", \"tier\": \"P1\", \"severity\": \"Critical\", \"title\": \"Escalation: data export failing for Acme\", \"owner\": \"priya\", \"next\": \"Assign an owner and post a status update\", \"by\": \"today\", \"link\": \"https://app.clickup.com/t/cu004\", \"source\": \"clickup\"}, {\"id\": \"cu021\", \"tier\": \"P4\", \"severity\": \"Low\", \"title\": \"Refactor billing page components\", \"owner\": \"me\", \"next\": \"Review the refactor PR when free\", \"by\": \"this week\", \"link\": \"https://app.clickup.com/t/cu021\", \"source\": \"clickup\"}]}"}, {"role": "assistant", "content": "I pulled your active triage items with list_items — 2 need you:\n• [P1/Critical] Escalation: data export failing for Acme — owner priya, next: Assign an owner and post a status update (due today)\n• [P4/Low] Refactor billing page components — owner me, next: Review the refactor PR when free (due this week)\nStart with Escalation: data export failing for Acme: assign an owner and post a status update."}], "tools": [{"type": "function", "function": {"name": "list_items", "description": "List the user's active triage items, optionally filtered by tier (P1-P4) or severity. Returns id, tier, severity, title, owner, next step, deadline, and link.", "parameters": {"type": "object", "required": [], "properties": {"tier": {"type": "string", "enum": ["P1", "P2", "P3", "P4"]}, "severity": {"type": "string", "enum": ["Critical", "High", "Medium", "Low"]}}}}}, {"type": "function", "function": {"name": "get_item", "description": "Fetch one active triage item by its id (fingerprint).", "parameters": {"type": "object", "required": ["id"], "properties": {"id": {"type": "string", "description": "The item id / fingerprint."}}}}}, {"type": "function", "function": {"name": "resolve_person", "description": "Resolve a person's name or handle to a Slack user id. Returns resolved, ambiguous (with candidates), or not_found.", "parameters": {"type": "object", "required": ["name"], "properties": {"name": {"type": "string", "description": "Person name, @handle, or email."}}}}}, {"type": "function", "function": {"name": "propose_calendar_block", "description": "Propose a calendar time block for the user to approve. Nothing is written without approval.", "parameters": {"type": "object", "required": ["title", "duration_minutes", "when"], "properties": {"title": {"type": "string"}, "duration_minutes": {"type": "integer"}, "when": {"type": "string", "description": "e.g. today afternoon, tomorrow 10:00"}, "item_id": {"type": "string", "description": "Optional linked triage item id."}}}}}, {"type": "function", "function": {"name": "list_calendar_events", "description": "List the user's calendar events for a day part.", "parameters": {"type": "object", "required": ["when"], "properties": {"when": {"type": "string", "description": "e.g. today, today afternoon, tomorrow"}}}}}, {"type": "function", "function": {"name": "get_sla_status", "description": "Summarize SLA state across the user's shared inboxes: breached and at-risk conversations.", "parameters": {"type": "object", "required": [], "properties": {}}}}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: hiver_mention\nCONVERSATION #1966 \| MY NOTE 2 days ago: \"@Tara can you check the course extension logs for Sofia P.?\"\nSTATUS: no response from teammate yet\nUSER CONTEXT: support agent, online course provider"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_them\", \"severity\": \"medium\", \"severity_rationale\": \"An internal ask has sat unanswered for 2 days with no committed date.\", \"tier\": \"P3\", \"owner\": \"Tara\", \"why_it_matters\": \"Sofia P.'s case cannot progress until Tara checks the course extension logs.\", \"next_step\": \"Nudge Tara on the course extension logs for conversation #1966\", \"deadline_suggestion\": \"this_week\", \"escalation_flag\": false}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: slack_channel_message\nCHANNEL: #ai-team\nFROM: Ishaan\nTIME: 12:00 today\nUSER CONTEXT: sole QA, AI team\nMESSAGE: \"sharing the recording of yesterday's design review for anyone who missed it\""}, {"role": "assistant", "content": "{\"is_actionable\": false, \"action_direction\": \"info_only\", \"severity\": \"low\", \"severity_rationale\": \"General FYI with no question, request, or risk directed at the user.\", \"tier\": \"P4\", \"owner\": \"none\", \"why_it_matters\": \"Informational chatter; no action exists.\", \"next_step\": \"No action needed\", \"deadline_suggestion\": \"this_week\", \"escalation_flag\": false}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: clickup_task\nTASK: \"QA: CSV export verification\" \| STATUS: blocked \| ASSIGNEE: me \| DUE: none \| TAGS: none\nCOMMENTS:\n[3 days ago] me: \"can't start — Vikram hasn't merged the dependency PR yet\"\n(no update since)\nUSER CONTEXT: sole QA, AI team"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_them\", \"severity\": \"medium\", \"severity_rationale\": \"The task has been blocked for three days with no date and no movement from the blocker.\", \"tier\": \"P3\", \"owner\": \"Vikram\", \"why_it_matters\": \"My CSV export verification cannot start until Vikram unblocks it, and no ETA exists.\", \"next_step\": \"Ping Vikram for an ETA on unblocking the CSV export verification\", \"deadline_suggestion\": \"tomorrow\", \"escalation_flag\": false}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: slack_channel_message\nCHANNEL: #general\nFROM: Divya (teammate)\nTIME: today 09:20\nUSER CONTEXT: product manager, AI team\nMESSAGE: \"asap: badge photos retaken today at reception if you missed the last round\""}, {"role": "assistant", "content": "{\"is_actionable\": false, \"action_direction\": \"info_only\", \"severity\": \"low\", \"severity_rationale\": \"Urgent-sounding wording, but it is internal logistics with no impact and no action owed.\", \"tier\": \"P4\", \"owner\": \"none\", \"why_it_matters\": \"Logistics notice to note; nothing to act on.\", \"next_step\": \"No action needed\", \"deadline_suggestion\": \"today\", \"escalation_flag\": false}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: hiver_shared_inbox_email\nINBOX: support@ \| FROM: Miguel T. (customer)\nTIME: yesterday 16:45\nMESSAGE: \"third email about my shared inbox sync and still no reply. If I don't hear back today I'm disputing the charge\"\nUSER CONTEXT: support engineer, B2B SaaS (shared inbox product)"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"critical\", \"severity_rationale\": \"A repeat-contact customer is threatening a chargeback or cancellation over an unresolved issue.\", \"tier\": \"P1\", \"owner\": \"me\", \"why_it_matters\": \"An escalating shared inbox sync complaint is at the point of churn and public complaint.\", \"next_step\": \"Reply to Miguel T. now with a concrete fix and timeline\", \"deadline_suggestion\": \"today\", \"escalation_flag\": true}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: hiver_shared_inbox_email\nINBOX: support@ \| FROM: Kevin D. (customer)\nTIME: today 08:10\nMESSAGE: \"third email about my CSV export and still no reply. If I don't hear back today I'm disputing the charge\"\nUSER CONTEXT: support engineer, B2B SaaS (shared inbox product)"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"critical\", \"severity_rationale\": \"A repeat-contact customer is threatening a chargeback or cancellation over an unresolved issue.\", \"tier\": \"P1\", \"owner\": \"me\", \"why_it_matters\": \"An escalating CSV export complaint is at the point of churn and public complaint.\", \"next_step\": \"Reply to Kevin D. now with a concrete fix and timeline\", \"deadline_suggestion\": \"today\", \"escalation_flag\": true}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: clickup_task\nTASK: \"QA: regression pass on dataset upload\" \| STATUS: open \| ASSIGNEE: me \| DUE: 3 days ago \| TAGS: none\nCOMMENTS: (none)\nUSER CONTEXT: sole QA, AI team"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"high\", \"severity_rationale\": \"An assigned task is 3 days past its due date with no progress recorded.\", \"tier\": \"P2\", \"owner\": \"me\", \"why_it_matters\": \"Overdue QA work delays the feature and erodes the due dates others plan around.\", \"next_step\": \"Start the dataset upload regression pass or re-negotiate its due date today\", \"deadline_suggestion\": \"today\", \"escalation_flag\": false}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: hiver_shared_inbox_email\nINBOX: claims@ \| FROM: Marcus J. (customer) \| STATUS: open, awaiting agent reply\nTIME: customer replied 5 hours ago\nMESSAGE: \"details attached as requested — any update?\"\nUSER CONTEXT: claims support agent, insurance agency"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"high\", \"severity_rationale\": \"The customer supplied everything requested 5 hours ago; the ball is back with us.\", \"tier\": \"P1\", \"owner\": \"me\", \"why_it_matters\": \"Marcus J.'s policy document case is stalled on our reply and response time is compounding.\", \"next_step\": \"Read Marcus J.'s reply and send the next-step answer\", \"deadline_suggestion\": \"today\", \"escalation_flag\": true}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: hiver_sla_alert\nCONVERSATION #8350 \| CUSTOMER: Marcus J. \| SUBJECT: ribbon rack issue\nALERT: first-response SLA breached 1 hour ago \| ASSIGNEE: me\nUSER CONTEXT: support agent, medals & ribbons e-commerce store"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"critical\", \"severity_rationale\": \"A first-response SLA has already been breached; every additional minute deepens the miss.\", \"tier\": \"P1\", \"owner\": \"me\", \"why_it_matters\": \"Marcus J.'s conversation is past its first-response commitment and counting.\", \"next_step\": \"Answer conversation #8350 immediately and apologize for the delay\", \"deadline_suggestion\": \"today\", \"escalation_flag\": true}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: hiver_shared_inbox_email\nINBOX: shipping@ \| FROM: Dana W. (customer)\nTIME: today 08:10\nMESSAGE: \"my return pickup is missing from the delivery — a customs deadline is coming up, can you help?\"\nUSER CONTEXT: support agent, logistics & shipping"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"high\", \"severity_rationale\": \"A customer has a defective or missing return pickup with a hard event date approaching.\", \"tier\": \"P2\", \"owner\": \"me\", \"why_it_matters\": \"Dana W. cannot make a customs deadline unless the return pickup is corrected in time.\", \"next_step\": \"Confirm replacement return pickup and expedited shipping to Dana W.\", \"deadline_suggestion\": \"today\", \"escalation_flag\": false}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: hiver_csat_rating\nRATING: 1/5 from Dana W. \| AGENT: Arjun\nCOMMENT: \"worst support experience — switching providers unless someone senior calls me\"\nUSER CONTEXT: support agent, logistics & shipping"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"critical\", \"severity_rationale\": \"A 1-star rating paired with an explicit threat to switch providers is active churn risk.\", \"tier\": \"P1\", \"owner\": \"me\", \"why_it_matters\": \"Dana W. is one unreturned call away from leaving.\", \"next_step\": \"Call Dana W. personally today about the poor experience\", \"deadline_suggestion\": \"today\", \"escalation_flag\": true}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: hiver_shared_inbox_email\nINBOX: support@ \| FROM: Priyanka R. (customer)\nTIME: today 08:10\nMESSAGE: \"my CSV export is missing from the delivery — a quarterly review is coming up, can you help?\"\nUSER CONTEXT: support engineer, B2B SaaS (shared inbox product)"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"high\", \"severity_rationale\": \"A customer has a defective or missing CSV export with a hard event date approaching.\", \"tier\": \"P2\", \"owner\": \"me\", \"why_it_matters\": \"Priyanka R. cannot make a quarterly review unless the CSV export is corrected in time.\", \"next_step\": \"Confirm replacement CSV export and expedited shipping to Priyanka R.\", \"deadline_suggestion\": \"today\", \"escalation_flag\": false}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: hiver_shared_inbox_email\nINBOX: claims@ \| FROM: Miguel T. (customer) \| STATUS: open, awaiting agent reply\nTIME: customer replied 6 hours ago\nMESSAGE: \"sent the photos you wanted — any update?\"\nUSER CONTEXT: claims support agent, insurance agency"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"high\", \"severity_rationale\": \"The customer supplied everything requested 6 hours ago; the ball is back with us.\", \"tier\": \"P1\", \"owner\": \"me\", \"why_it_matters\": \"Miguel T.'s claim form case is stalled on our reply and response time is compounding.\", \"next_step\": \"Read Miguel T.'s reply and send the next-step answer\", \"deadline_suggestion\": \"today\", \"escalation_flag\": true}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: hiver_mention\nCONVERSATION #5213 \| FROM: Arjun (teammate) \| TIME: yesterday 17:55\nNOTE: \"@me can you take this one? Marcus J. is asking about the custom ribbon bar and you handled their last case\"\nUSER CONTEXT: support agent, medals & ribbons e-commerce store"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"medium\", \"severity_rationale\": \"A teammate has handed over a customer conversation and is waiting for pickup.\", \"tier\": \"P2\", \"owner\": \"me\", \"why_it_matters\": \"Marcus J.'s custom ribbon bar question sits unowned until the handover is accepted.\", \"next_step\": \"Accept conversation #5213 and reply to Marcus J.\", \"deadline_suggestion\": \"today\", \"escalation_flag\": false}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: hiver_shared_inbox_email\nINBOX: orders@ \| FROM: Omar S. (customer) \| STATUS: open, awaiting agent reply\nTIME: customer replied 6 hours ago\nMESSAGE: \"sent the photos you wanted — any update?\"\nUSER CONTEXT: support agent, medals & ribbons e-commerce store"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"high\", \"severity_rationale\": \"The customer supplied everything requested 6 hours ago; the ball is back with us.\", \"tier\": \"P1\", \"owner\": \"me\", \"why_it_matters\": \"Omar S.'s shadow box frame case is stalled on our reply and response time is compounding.\", \"next_step\": \"Read Omar S.'s reply and send the next-step answer\", \"deadline_suggestion\": \"today\", \"escalation_flag\": true}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: hiver_shared_inbox_email\nINBOX: claims@ \| FROM: Priyanka R. (customer) \| STATUS: open, awaiting agent reply\nTIME: customer replied 4 hours ago\nMESSAGE: \"sent the photos you wanted — any update?\"\nUSER CONTEXT: claims support agent, insurance agency"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"high\", \"severity_rationale\": \"The customer supplied everything requested 4 hours ago; the ball is back with us.\", \"tier\": \"P1\", \"owner\": \"me\", \"why_it_matters\": \"Priyanka R.'s claim form case is stalled on our reply and response time is compounding.\", \"next_step\": \"Read Priyanka R.'s reply and send the next-step answer\", \"deadline_suggestion\": \"today\", \"escalation_flag\": true}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: slack_dm\nFROM: Anand (EM)\nTIME: today 10:15\nUSER CONTEXT: product manager, AI team\nMESSAGE: \"release train leaves tomorrow morning — go or no-go on the model comparison table?\""}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"high\", \"severity_rationale\": \"A release decision is pending on the PM with an explicit same-day deadline.\", \"tier\": \"P2\", \"owner\": \"me\", \"why_it_matters\": \"Without a go/no-go the model comparison table misses the release train by default.\", \"next_step\": \"Review the model comparison table readiness and send the go/no-go decision today\", \"deadline_suggestion\": \"today\", \"escalation_flag\": false}"}]}
	{"messages": [{"role": "system", "content": "You are UnchaosLM, the conversational assistant of Triage Copilot. You help the user act on their triaged work: tasks, escalations, priorities, people, inboxes, and calendar. When a request needs live data or an action, call the matching tool — extract the arguments from the user's message itself, even when phrased casually. After a tool returns, answer grounded strictly in its result and say plainly what you did (which tool you used and what it found). If a required argument is genuinely absent, ask one short question instead of guessing. Greetings and questions about your abilities need no tool. Never invent items, ids, owners, or links."}, {"role": "user", "content": "what fires are burning, please?"}, {"role": "assistant", "tool_calls": [{"function": {"name": "list_items", "arguments": {"tier": "P1"}}}], "content": ""}, {"role": "tool", "content": "{\"items\": [{\"id\": \"slk-8f2a\", \"tier\": \"P1\", \"severity\": \"Critical\", \"title\": \"Acme production checkout returning 500s\", \"owner\": \"priya\", \"next\": \"Join the escalation call and commit to a fix\", \"by\": \"today\", \"link\": \"https://app.clickup.com/t/slk-8f2a\", \"source\": \"slack\"}, {\"id\": \"cu004\", \"tier\": \"P1\", \"severity\": \"Critical\", \"title\": \"Escalation: data export failing for Acme\", \"owner\": \"priya\", \"next\": \"Assign an owner and post a status update\", \"by\": \"today\", \"link\": \"https://app.clickup.com/t/cu004\", \"source\": \"clickup\"}, {\"id\": \"cu003\", \"tier\": \"P1\", \"severity\": \"High\", \"title\": \"Fix CART checkout total mismatch\", \"owner\": \"priya\", \"next\": \"Update the task or set a new due date\", \"by\": \"overdue\", \"link\": \"https://app.clickup.com/t/cu003\", \"source\": \"clickup\"}]}"}, {"role": "assistant", "content": "I pulled your P1 triage items with list_items — 3 need you:\n• [P1/Critical] Acme production checkout returning 500s — owner priya, next: Join the escalation call and commit to a fix (due today)\n• [P1/Critical] Escalation: data export failing for Acme — owner priya, next: Assign an owner and post a status update (due today)\n• [P1/High] Fix CART checkout total mismatch — owner priya, next: Update the task or set a new due date (due overdue)\nStart with Acme production checkout returning 500s: join the escalation call and commit to a fix."}], "tools": [{"type": "function", "function": {"name": "list_items", "description": "List the user's active triage items, optionally filtered by tier (P1-P4) or severity. Returns id, tier, severity, title, owner, next step, deadline, and link.", "parameters": {"type": "object", "required": [], "properties": {"tier": {"type": "string", "enum": ["P1", "P2", "P3", "P4"]}, "severity": {"type": "string", "enum": ["Critical", "High", "Medium", "Low"]}}}}}, {"type": "function", "function": {"name": "get_item", "description": "Fetch one active triage item by its id (fingerprint).", "parameters": {"type": "object", "required": ["id"], "properties": {"id": {"type": "string", "description": "The item id / fingerprint."}}}}}, {"type": "function", "function": {"name": "resolve_person", "description": "Resolve a person's name or handle to a Slack user id. Returns resolved, ambiguous (with candidates), or not_found.", "parameters": {"type": "object", "required": ["name"], "properties": {"name": {"type": "string", "description": "Person name, @handle, or email."}}}}}, {"type": "function", "function": {"name": "propose_calendar_block", "description": "Propose a calendar time block for the user to approve. Nothing is written without approval.", "parameters": {"type": "object", "required": ["title", "duration_minutes", "when"], "properties": {"title": {"type": "string"}, "duration_minutes": {"type": "integer"}, "when": {"type": "string", "description": "e.g. today afternoon, tomorrow 10:00"}, "item_id": {"type": "string", "description": "Optional linked triage item id."}}}}}, {"type": "function", "function": {"name": "list_calendar_events", "description": "List the user's calendar events for a day part.", "parameters": {"type": "object", "required": ["when"], "properties": {"when": {"type": "string", "description": "e.g. today, today afternoon, tomorrow"}}}}}, {"type": "function", "function": {"name": "get_sla_status", "description": "Summarize SLA state across the user's shared inboxes: breached and at-risk conversations.", "parameters": {"type": "object", "required": [], "properties": {}}}}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: slack_channel_message\nCHANNEL: #general\nFROM: Rohit (teammate)\nTIME: today 09:20\nUSER CONTEXT: product manager, AI team\nMESSAGE: \"URGENT reminder — submit your lunch order by 11 or you get the default\""}, {"role": "assistant", "content": "{\"is_actionable\": false, \"action_direction\": \"info_only\", \"severity\": \"low\", \"severity_rationale\": \"Urgent-sounding wording, but it is internal logistics with no impact and no action owed.\", \"tier\": \"P4\", \"owner\": \"none\", \"why_it_matters\": \"Logistics notice to note; nothing to act on.\", \"next_step\": \"No action needed\", \"deadline_suggestion\": \"today\", \"escalation_flag\": false}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: slack_thread\nCHANNEL: #ai-team-bugs\nTHREAD HISTORY:\n[4 days ago] me: \"Flagging: model comparison table silently fails on large inputs. @Nikhil\"\n[4 days ago] Nikhil (dev): \"will pick it up after the current feature, can't say when\"\n(no messages since)\nTIME NOW: today 08:30\nUSER CONTEXT: sole QA, AI team"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_them\", \"severity\": \"medium\", \"severity_rationale\": \"Acknowledged bug with no committed date and four days of silence \\u2014 silent delivery risk.\", \"tier\": \"P3\", \"owner\": \"Nikhil\", \"why_it_matters\": \"No ETA exists, so this bug can drift indefinitely without anyone noticing.\", \"next_step\": \"Ask Nikhil for an ETA on the model comparison table fix\", \"deadline_suggestion\": \"this_week\", \"escalation_flag\": false}"}]}
	{"messages": [{"role": "system", "content": "You are UnchaosLM, the conversational assistant of Triage Copilot. You help the user act on their triaged work: tasks, escalations, priorities, people, inboxes, and calendar. When a request needs live data or an action, call the matching tool — extract the arguments from the user's message itself, even when phrased casually. After a tool returns, answer grounded strictly in its result and say plainly what you did (which tool you used and what it found). If a required argument is genuinely absent, ask one short question instead of guessing. Greetings and questions about your abilities need no tool. Never invent items, ids, owners, or links."}, {"role": "user", "content": "can you check slk-77b0 for me"}, {"role": "assistant", "tool_calls": [{"function": {"name": "get_item", "arguments": {"id": "slk-77b0"}}}], "content": ""}, {"role": "tool", "content": "{\"id\": \"slk-77b0\", \"tier\": \"P3\", \"severity\": \"Medium\", \"title\": \"Waiting on QA repro for recipe editor bug\", \"owner\": \"vikram\", \"next\": \"Nudge QA on the repro request\", \"by\": \"this week\", \"link\": \"https://app.clickup.com/t/slk-77b0\", \"source\": \"slack\"}"}, {"role": "assistant", "content": "I fetched slk-77b0 with get_item. \"Waiting on QA repro for recipe editor bug\" is P3/Medium, owned by vikram, due this week. Next step on record: nudge qa on the repro request. Link: https://app.clickup.com/t/slk-77b0"}], "tools": [{"type": "function", "function": {"name": "list_items", "description": "List the user's active triage items, optionally filtered by tier (P1-P4) or severity. Returns id, tier, severity, title, owner, next step, deadline, and link.", "parameters": {"type": "object", "required": [], "properties": {"tier": {"type": "string", "enum": ["P1", "P2", "P3", "P4"]}, "severity": {"type": "string", "enum": ["Critical", "High", "Medium", "Low"]}}}}}, {"type": "function", "function": {"name": "get_item", "description": "Fetch one active triage item by its id (fingerprint).", "parameters": {"type": "object", "required": ["id"], "properties": {"id": {"type": "string", "description": "The item id / fingerprint."}}}}}, {"type": "function", "function": {"name": "resolve_person", "description": "Resolve a person's name or handle to a Slack user id. Returns resolved, ambiguous (with candidates), or not_found.", "parameters": {"type": "object", "required": ["name"], "properties": {"name": {"type": "string", "description": "Person name, @handle, or email."}}}}}, {"type": "function", "function": {"name": "propose_calendar_block", "description": "Propose a calendar time block for the user to approve. Nothing is written without approval.", "parameters": {"type": "object", "required": ["title", "duration_minutes", "when"], "properties": {"title": {"type": "string"}, "duration_minutes": {"type": "integer"}, "when": {"type": "string", "description": "e.g. today afternoon, tomorrow 10:00"}, "item_id": {"type": "string", "description": "Optional linked triage item id."}}}}}, {"type": "function", "function": {"name": "list_calendar_events", "description": "List the user's calendar events for a day part.", "parameters": {"type": "object", "required": ["when"], "properties": {"when": {"type": "string", "description": "e.g. today, today afternoon, tomorrow"}}}}}, {"type": "function", "function": {"name": "get_sla_status", "description": "Summarize SLA state across the user's shared inboxes: breached and at-risk conversations.", "parameters": {"type": "object", "required": [], "properties": {}}}}]}
	{"messages": [{"role": "system", "content": "You are UnchaosLM, the conversational assistant of Triage Copilot. You help the user act on their triaged work: tasks, escalations, priorities, people, inboxes, and calendar. When a request needs live data or an action, call the matching tool — extract the arguments from the user's message itself, even when phrased casually. After a tool returns, answer grounded strictly in its result and say plainly what you did (which tool you used and what it found). If a required argument is genuinely absent, ask one short question instead of guessing. Greetings and questions about your abilities need no tool. Never invent items, ids, owners, or links."}, {"role": "user", "content": "what's stuck with no ETA, please?"}, {"role": "assistant", "tool_calls": [{"function": {"name": "list_items", "arguments": {"tier": "P3"}}}], "content": ""}], "tools": [{"type": "function", "function": {"name": "list_items", "description": "List the user's active triage items, optionally filtered by tier (P1-P4) or severity. Returns id, tier, severity, title, owner, next step, deadline, and link.", "parameters": {"type": "object", "required": [], "properties": {"tier": {"type": "string", "enum": ["P1", "P2", "P3", "P4"]}, "severity": {"type": "string", "enum": ["Critical", "High", "Medium", "Low"]}}}}}, {"type": "function", "function": {"name": "get_item", "description": "Fetch one active triage item by its id (fingerprint).", "parameters": {"type": "object", "required": ["id"], "properties": {"id": {"type": "string", "description": "The item id / fingerprint."}}}}}, {"type": "function", "function": {"name": "resolve_person", "description": "Resolve a person's name or handle to a Slack user id. Returns resolved, ambiguous (with candidates), or not_found.", "parameters": {"type": "object", "required": ["name"], "properties": {"name": {"type": "string", "description": "Person name, @handle, or email."}}}}}, {"type": "function", "function": {"name": "propose_calendar_block", "description": "Propose a calendar time block for the user to approve. Nothing is written without approval.", "parameters": {"type": "object", "required": ["title", "duration_minutes", "when"], "properties": {"title": {"type": "string"}, "duration_minutes": {"type": "integer"}, "when": {"type": "string", "description": "e.g. today afternoon, tomorrow 10:00"}, "item_id": {"type": "string", "description": "Optional linked triage item id."}}}}}, {"type": "function", "function": {"name": "list_calendar_events", "description": "List the user's calendar events for a day part.", "parameters": {"type": "object", "required": ["when"], "properties": {"when": {"type": "string", "description": "e.g. today, today afternoon, tomorrow"}}}}}, {"type": "function", "function": {"name": "get_sla_status", "description": "Summarize SLA state across the user's shared inboxes: breached and at-risk conversations.", "parameters": {"type": "object", "required": [], "properties": {}}}}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: hiver_sla_alert\nCONVERSATION #6417 \| CUSTOMER: Marcus J. \| SUBJECT: course login issue\nALERT: first-response SLA breached 1 hour ago \| ASSIGNEE: me\nUSER CONTEXT: support agent, online course provider"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"critical\", \"severity_rationale\": \"A first-response SLA has already been breached; every additional minute deepens the miss.\", \"tier\": \"P1\", \"owner\": \"me\", \"why_it_matters\": \"Marcus J.'s conversation is past its first-response commitment and counting.\", \"next_step\": \"Answer conversation #6417 immediately and apologize for the delay\", \"deadline_suggestion\": \"today\", \"escalation_flag\": true}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: hiver_mention\nCONVERSATION #9251 \| FROM: Dev (teammate) \| TIME: today 11:05\nNOTE: \"@me can you take this one? Linda M. is asking about the CE credit record and you handled their last case\"\nUSER CONTEXT: support agent, online course provider"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"medium\", \"severity_rationale\": \"A teammate has handed over a customer conversation and is waiting for pickup.\", \"tier\": \"P2\", \"owner\": \"me\", \"why_it_matters\": \"Linda M.'s CE credit record question sits unowned until the handover is accepted.\", \"next_step\": \"Accept conversation #9251 and reply to Linda M.\", \"deadline_suggestion\": \"today\", \"escalation_flag\": false}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: hiver_escalation_tag\nCONVERSATION #6795 \| TAGGED: Escalation by Dev \| ACCOUNT: Summit Insurance Group\nREASON: repeat outage reports on policy document from Summit Insurance Group, third conversation this week\nTIME: 2 days ago 09:15\nUSER CONTEXT: claims support agent, insurance agency"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"critical\", \"severity_rationale\": \"A teammate has formally tagged the conversation as an escalation with customer impact.\", \"tier\": \"P1\", \"owner\": \"me\", \"why_it_matters\": \"Summit Insurance Group's escalation is live and visible; delay compounds the damage.\", \"next_step\": \"Own conversation #6795 and respond to the escalation now\", \"deadline_suggestion\": \"today\", \"escalation_flag\": true}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: slack_dm\nFROM: Priya (Support Engineer)\nTIME: 11:40 today\nUSER CONTEXT: sole QA, AI team\nMESSAGE: \"P0 from support: customer completely stuck on the eval-runs page. free for 15 min?\""}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"critical\", \"severity_rationale\": \"Support engineer reports a customer actively blocked and needs immediate repro help.\", \"tier\": \"P1\", \"owner\": \"me\", \"why_it_matters\": \"A customer is blocked right now; the repro is the gate to the fix.\", \"next_step\": \"Huddle with Priya now to reproduce the eval-runs page issue\", \"deadline_suggestion\": \"today\", \"escalation_flag\": true}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: slack_dm\nFROM: Kavya (EM)\nTIME: today 10:15\nUSER CONTEXT: product manager, AI team\nMESSAGE: \"need your ship decision on the SSO login today, the changelog is being written now\""}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"high\", \"severity_rationale\": \"A release decision is pending on the PM with an explicit same-day deadline.\", \"tier\": \"P2\", \"owner\": \"me\", \"why_it_matters\": \"Without a go/no-go the SSO login misses the release train by default.\", \"next_step\": \"Review the SSO login readiness and send the go/no-go decision today\", \"deadline_suggestion\": \"today\", \"escalation_flag\": false}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: hiver_shared_inbox_email\nINBOX: claims@ \| FROM: Aisha K. (customer)\nTIME: today 13:30\nMESSAGE: \"third email about my coverage letter and still no reply. If I don't hear back today I'm disputing the charge\"\nUSER CONTEXT: claims support agent, insurance agency"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"critical\", \"severity_rationale\": \"A repeat-contact customer is threatening a chargeback or cancellation over an unresolved issue.\", \"tier\": \"P1\", \"owner\": \"me\", \"why_it_matters\": \"An escalating coverage letter complaint is at the point of churn and public complaint.\", \"next_step\": \"Reply to Aisha K. now with a concrete fix and timeline\", \"deadline_suggestion\": \"today\", \"escalation_flag\": true}"}]}
	{"messages": [{"role": "system", "content": "You are UnchaosLM, the conversational assistant of Triage Copilot. You help the user act on their triaged work: tasks, escalations, priorities, people, inboxes, and calendar. When a request needs live data or an action, call the matching tool — extract the arguments from the user's message itself, even when phrased casually. After a tool returns, answer grounded strictly in its result and say plainly what you did (which tool you used and what it found). If a required argument is genuinely absent, ask one short question instead of guessing. Greetings and questions about your abilities need no tool. Never invent items, ids, owners, or links."}, {"role": "user", "content": "what needs me right now, please?"}, {"role": "assistant", "tool_calls": [{"function": {"name": "list_items", "arguments": {}}}], "content": ""}], "tools": [{"type": "function", "function": {"name": "list_items", "description": "List the user's active triage items, optionally filtered by tier (P1-P4) or severity. Returns id, tier, severity, title, owner, next step, deadline, and link.", "parameters": {"type": "object", "required": [], "properties": {"tier": {"type": "string", "enum": ["P1", "P2", "P3", "P4"]}, "severity": {"type": "string", "enum": ["Critical", "High", "Medium", "Low"]}}}}}, {"type": "function", "function": {"name": "get_item", "description": "Fetch one active triage item by its id (fingerprint).", "parameters": {"type": "object", "required": ["id"], "properties": {"id": {"type": "string", "description": "The item id / fingerprint."}}}}}, {"type": "function", "function": {"name": "resolve_person", "description": "Resolve a person's name or handle to a Slack user id. Returns resolved, ambiguous (with candidates), or not_found.", "parameters": {"type": "object", "required": ["name"], "properties": {"name": {"type": "string", "description": "Person name, @handle, or email."}}}}}, {"type": "function", "function": {"name": "propose_calendar_block", "description": "Propose a calendar time block for the user to approve. Nothing is written without approval.", "parameters": {"type": "object", "required": ["title", "duration_minutes", "when"], "properties": {"title": {"type": "string"}, "duration_minutes": {"type": "integer"}, "when": {"type": "string", "description": "e.g. today afternoon, tomorrow 10:00"}, "item_id": {"type": "string", "description": "Optional linked triage item id."}}}}}, {"type": "function", "function": {"name": "list_calendar_events", "description": "List the user's calendar events for a day part.", "parameters": {"type": "object", "required": ["when"], "properties": {"when": {"type": "string", "description": "e.g. today, today afternoon, tomorrow"}}}}}, {"type": "function", "function": {"name": "get_sla_status", "description": "Summarize SLA state across the user's shared inboxes: breached and at-risk conversations.", "parameters": {"type": "object", "required": [], "properties": {}}}}]}
	{"messages": [{"role": "system", "content": "You are UnchaosLM, the conversational assistant of Triage Copilot. You help the user act on their triaged work: tasks, escalations, priorities, people, inboxes, and calendar. When a request needs live data or an action, call the matching tool — extract the arguments from the user's message itself, even when phrased casually. After a tool returns, answer grounded strictly in its result and say plainly what you did (which tool you used and what it found). If a required argument is genuinely absent, ask one short question instead of guessing. Greetings and questions about your abilities need no tool. Never invent items, ids, owners, or links."}, {"role": "user", "content": "what's on my plate today"}, {"role": "assistant", "tool_calls": [{"function": {"name": "list_items", "arguments": {}}}], "content": ""}, {"role": "tool", "content": "{\"items\": [{\"id\": \"slk-8f2a\", \"tier\": \"P1\", \"severity\": \"Critical\", \"title\": \"Acme production checkout returning 500s\", \"owner\": \"priya\", \"next\": \"Join the escalation call and commit to a fix\", \"by\": \"today\", \"link\": \"https://app.clickup.com/t/slk-8f2a\", \"source\": \"slack\"}, {\"id\": \"cu004\", \"tier\": \"P1\", \"severity\": \"Critical\", \"title\": \"Escalation: data export failing for Acme\", \"owner\": \"priya\", \"next\": \"Assign an owner and post a status update\", \"by\": \"today\", \"link\": \"https://app.clickup.com/t/cu004\", \"source\": \"clickup\"}, {\"id\": \"slk-3c91\", \"tier\": \"P2\", \"severity\": \"High\", \"title\": \"Pricing page copy decision\", \"owner\": \"meera\", \"next\": \"Make the go/no-go call and post it\", \"by\": \"today\", \"link\": \"https://app.clickup.com/t/slk-3c91\", \"source\": \"slack\"}]}"}, {"role": "assistant", "content": "I pulled your active triage items with list_items — 3 need you:\n• [P1/Critical] Acme production checkout returning 500s — owner priya, next: Join the escalation call and commit to a fix (due today)\n• [P1/Critical] Escalation: data export failing for Acme — owner priya, next: Assign an owner and post a status update (due today)\n• [P2/High] Pricing page copy decision — owner meera, next: Make the go/no-go call and post it (due today)\nStart with Acme production checkout returning 500s: join the escalation call and commit to a fix."}], "tools": [{"type": "function", "function": {"name": "list_items", "description": "List the user's active triage items, optionally filtered by tier (P1-P4) or severity. Returns id, tier, severity, title, owner, next step, deadline, and link.", "parameters": {"type": "object", "required": [], "properties": {"tier": {"type": "string", "enum": ["P1", "P2", "P3", "P4"]}, "severity": {"type": "string", "enum": ["Critical", "High", "Medium", "Low"]}}}}}, {"type": "function", "function": {"name": "get_item", "description": "Fetch one active triage item by its id (fingerprint).", "parameters": {"type": "object", "required": ["id"], "properties": {"id": {"type": "string", "description": "The item id / fingerprint."}}}}}, {"type": "function", "function": {"name": "resolve_person", "description": "Resolve a person's name or handle to a Slack user id. Returns resolved, ambiguous (with candidates), or not_found.", "parameters": {"type": "object", "required": ["name"], "properties": {"name": {"type": "string", "description": "Person name, @handle, or email."}}}}}, {"type": "function", "function": {"name": "propose_calendar_block", "description": "Propose a calendar time block for the user to approve. Nothing is written without approval.", "parameters": {"type": "object", "required": ["title", "duration_minutes", "when"], "properties": {"title": {"type": "string"}, "duration_minutes": {"type": "integer"}, "when": {"type": "string", "description": "e.g. today afternoon, tomorrow 10:00"}, "item_id": {"type": "string", "description": "Optional linked triage item id."}}}}}, {"type": "function", "function": {"name": "list_calendar_events", "description": "List the user's calendar events for a day part.", "parameters": {"type": "object", "required": ["when"], "properties": {"when": {"type": "string", "description": "e.g. today, today afternoon, tomorrow"}}}}}, {"type": "function", "function": {"name": "get_sla_status", "description": "Summarize SLA state across the user's shared inboxes: breached and at-risk conversations.", "parameters": {"type": "object", "required": [], "properties": {}}}}]}
	{"messages": [{"role": "system", "content": "You are UnchaosLM, the conversational assistant of Triage Copilot. You help the user act on their triaged work: tasks, escalations, priorities, people, inboxes, and calendar. When a request needs live data or an action, call the matching tool — extract the arguments from the user's message itself, even when phrased casually. After a tool returns, answer grounded strictly in its result and say plainly what you did (which tool you used and what it found). If a required argument is genuinely absent, ask one short question instead of guessing. Greetings and questions about your abilities need no tool. Never invent items, ids, owners, or links."}, {"role": "user", "content": "can you check cu011 for me"}, {"role": "assistant", "tool_calls": [{"function": {"name": "get_item", "arguments": {"id": "cu011"}}}], "content": ""}, {"role": "tool", "content": "{\"id\": \"cu011\", \"tier\": \"P2\", \"severity\": \"Medium\", \"title\": \"Finalize pricing experiment plan\", \"owner\": \"meera\", \"next\": \"Get the analytics baseline\", \"by\": \"tomorrow\", \"link\": \"https://app.clickup.com/t/cu011\", \"source\": \"clickup\"}"}, {"role": "assistant", "content": "I fetched cu011 with get_item. \"Finalize pricing experiment plan\" is P2/Medium, owned by meera, due tomorrow. Next step on record: get the analytics baseline. Link: https://app.clickup.com/t/cu011"}], "tools": [{"type": "function", "function": {"name": "list_items", "description": "List the user's active triage items, optionally filtered by tier (P1-P4) or severity. Returns id, tier, severity, title, owner, next step, deadline, and link.", "parameters": {"type": "object", "required": [], "properties": {"tier": {"type": "string", "enum": ["P1", "P2", "P3", "P4"]}, "severity": {"type": "string", "enum": ["Critical", "High", "Medium", "Low"]}}}}}, {"type": "function", "function": {"name": "get_item", "description": "Fetch one active triage item by its id (fingerprint).", "parameters": {"type": "object", "required": ["id"], "properties": {"id": {"type": "string", "description": "The item id / fingerprint."}}}}}, {"type": "function", "function": {"name": "resolve_person", "description": "Resolve a person's name or handle to a Slack user id. Returns resolved, ambiguous (with candidates), or not_found.", "parameters": {"type": "object", "required": ["name"], "properties": {"name": {"type": "string", "description": "Person name, @handle, or email."}}}}}, {"type": "function", "function": {"name": "propose_calendar_block", "description": "Propose a calendar time block for the user to approve. Nothing is written without approval.", "parameters": {"type": "object", "required": ["title", "duration_minutes", "when"], "properties": {"title": {"type": "string"}, "duration_minutes": {"type": "integer"}, "when": {"type": "string", "description": "e.g. today afternoon, tomorrow 10:00"}, "item_id": {"type": "string", "description": "Optional linked triage item id."}}}}}, {"type": "function", "function": {"name": "list_calendar_events", "description": "List the user's calendar events for a day part.", "parameters": {"type": "object", "required": ["when"], "properties": {"when": {"type": "string", "description": "e.g. today, today afternoon, tomorrow"}}}}}, {"type": "function", "function": {"name": "get_sla_status", "description": "Summarize SLA state across the user's shared inboxes: breached and at-risk conversations.", "parameters": {"type": "object", "required": [], "properties": {}}}}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: slack_thread\nCHANNEL: #ai-team\nTHREAD HISTORY:\n[Yesterday 15:02] Rohit (dev): \"@qa did you get a chance to test the new onboarding flow?\"\n[Yesterday 15:40] me: \"tested this morning, two minor notes added to the task\"\n[Yesterday 15:41] Rohit: \"perfect, thanks!\"\nTIME NOW: today 08:30\nUSER CONTEXT: sole QA, AI team"}, {"role": "assistant", "content": "{\"is_actionable\": false, \"action_direction\": \"info_only\", \"severity\": \"low\", \"severity_rationale\": \"The question was asked and fully answered yesterday; nothing remains open.\", \"tier\": \"P4\", \"owner\": \"none\", \"why_it_matters\": \"The thread is closed; surfacing it would be noise.\", \"next_step\": \"No action needed\", \"deadline_suggestion\": \"this_week\", \"escalation_flag\": false}"}]}
	{"messages": [{"role": "system", "content": "You are UnchaosLM, the conversational assistant of Triage Copilot. You help the user act on their triaged work: tasks, escalations, priorities, people, inboxes, and calendar. When a request needs live data or an action, call the matching tool — extract the arguments from the user's message itself, even when phrased casually. After a tool returns, answer grounded strictly in its result and say plainly what you did (which tool you used and what it found). If a required argument is genuinely absent, ask one short question instead of guessing. Greetings and questions about your abilities need no tool. Never invent items, ids, owners, or links."}, {"role": "user", "content": "give me the details on cu017"}, {"role": "assistant", "tool_calls": [{"function": {"name": "get_item", "arguments": {"id": "cu017"}}}], "content": ""}, {"role": "tool", "content": "{\"id\": \"cu017\", \"tier\": \"P3\", \"severity\": \"Medium\", \"title\": \"Vendor security questionnaire\", \"owner\": \"dev\", \"next\": \"Chase the vendor for an ETA\", \"by\": \"this week\", \"link\": \"https://app.clickup.com/t/cu017\", \"source\": \"clickup\"}"}, {"role": "assistant", "content": "I fetched cu017 with get_item. \"Vendor security questionnaire\" is P3/Medium, owned by dev, due this week. Next step on record: chase the vendor for an eta. Link: https://app.clickup.com/t/cu017"}], "tools": [{"type": "function", "function": {"name": "list_items", "description": "List the user's active triage items, optionally filtered by tier (P1-P4) or severity. Returns id, tier, severity, title, owner, next step, deadline, and link.", "parameters": {"type": "object", "required": [], "properties": {"tier": {"type": "string", "enum": ["P1", "P2", "P3", "P4"]}, "severity": {"type": "string", "enum": ["Critical", "High", "Medium", "Low"]}}}}}, {"type": "function", "function": {"name": "get_item", "description": "Fetch one active triage item by its id (fingerprint).", "parameters": {"type": "object", "required": ["id"], "properties": {"id": {"type": "string", "description": "The item id / fingerprint."}}}}}, {"type": "function", "function": {"name": "resolve_person", "description": "Resolve a person's name or handle to a Slack user id. Returns resolved, ambiguous (with candidates), or not_found.", "parameters": {"type": "object", "required": ["name"], "properties": {"name": {"type": "string", "description": "Person name, @handle, or email."}}}}}, {"type": "function", "function": {"name": "propose_calendar_block", "description": "Propose a calendar time block for the user to approve. Nothing is written without approval.", "parameters": {"type": "object", "required": ["title", "duration_minutes", "when"], "properties": {"title": {"type": "string"}, "duration_minutes": {"type": "integer"}, "when": {"type": "string", "description": "e.g. today afternoon, tomorrow 10:00"}, "item_id": {"type": "string", "description": "Optional linked triage item id."}}}}}, {"type": "function", "function": {"name": "list_calendar_events", "description": "List the user's calendar events for a day part.", "parameters": {"type": "object", "required": ["when"], "properties": {"when": {"type": "string", "description": "e.g. today, today afternoon, tomorrow"}}}}}, {"type": "function", "function": {"name": "get_sla_status", "description": "Summarize SLA state across the user's shared inboxes: breached and at-risk conversations.", "parameters": {"type": "object", "required": [], "properties": {}}}}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: hiver_assignment\nCONVERSATION #7113 \| CUSTOMER: Emily A. \| SUBJECT: delivery slot problem\nSTATUS: open, assigned to me 2 days ago, no reply sent yet\nUSER CONTEXT: support agent, logistics & shipping"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"high\", \"severity_rationale\": \"An assigned conversation has waited 2 days without a single reply.\", \"tier\": \"P2\", \"owner\": \"me\", \"why_it_matters\": \"Emily A. has had silence for 2 days on an open delivery slot problem.\", \"next_step\": \"Send Emily A. a first reply on the delivery slot problem\", \"deadline_suggestion\": \"today\", \"escalation_flag\": false}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: hiver_assignment\nCONVERSATION #2669 \| CUSTOMER: Helen C. \| SUBJECT: proctored exam slot replacement\nSTATUS: pending — waiting on supplier confirmation, no ETA on their side\nUSER CONTEXT: support agent, online course provider"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_them\", \"severity\": \"medium\", \"severity_rationale\": \"The case is blocked on a third party that has given no timeline at all.\", \"tier\": \"P3\", \"owner\": \"supplier\", \"why_it_matters\": \"Helen C.'s proctored exam slot replacement is open-ended until the supplier commits to a date.\", \"next_step\": \"Chase the supplier for an ETA on Helen C.'s proctored exam slot\", \"deadline_suggestion\": \"this_week\", \"escalation_flag\": false}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: hiver_escalation_tag\nCONVERSATION #9374 \| TAGGED: Escalation by Neha \| ACCOUNT: ClearWater Labs\nREASON: ClearWater Labs says the module 4 quiz failure is blocking payroll for their staff\nTIME: yesterday 15:20\nUSER CONTEXT: support agent, online course provider"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"critical\", \"severity_rationale\": \"A teammate has formally tagged the conversation as an escalation with customer impact.\", \"tier\": \"P1\", \"owner\": \"me\", \"why_it_matters\": \"ClearWater Labs's escalation is live and visible; delay compounds the damage.\", \"next_step\": \"Own conversation #9374 and respond to the escalation now\", \"deadline_suggestion\": \"today\", \"escalation_flag\": true}"}]}
	{"messages": [{"role": "system", "content": "You are UnchaosLM, the conversational assistant of Triage Copilot. You help the user act on their triaged work: tasks, escalations, priorities, people, inboxes, and calendar. When a request needs live data or an action, call the matching tool — extract the arguments from the user's message itself, even when phrased casually. After a tool returns, answer grounded strictly in its result and say plainly what you did (which tool you used and what it found). If a required argument is genuinely absent, ask one short question instead of guessing. Greetings and questions about your abilities need no tool. Never invent items, ids, owners, or links."}, {"role": "user", "content": "what are my tasks for today, please?"}, {"role": "assistant", "tool_calls": [{"function": {"name": "list_items", "arguments": {}}}], "content": ""}], "tools": [{"type": "function", "function": {"name": "list_items", "description": "List the user's active triage items, optionally filtered by tier (P1-P4) or severity. Returns id, tier, severity, title, owner, next step, deadline, and link.", "parameters": {"type": "object", "required": [], "properties": {"tier": {"type": "string", "enum": ["P1", "P2", "P3", "P4"]}, "severity": {"type": "string", "enum": ["Critical", "High", "Medium", "Low"]}}}}}, {"type": "function", "function": {"name": "get_item", "description": "Fetch one active triage item by its id (fingerprint).", "parameters": {"type": "object", "required": ["id"], "properties": {"id": {"type": "string", "description": "The item id / fingerprint."}}}}}, {"type": "function", "function": {"name": "resolve_person", "description": "Resolve a person's name or handle to a Slack user id. Returns resolved, ambiguous (with candidates), or not_found.", "parameters": {"type": "object", "required": ["name"], "properties": {"name": {"type": "string", "description": "Person name, @handle, or email."}}}}}, {"type": "function", "function": {"name": "propose_calendar_block", "description": "Propose a calendar time block for the user to approve. Nothing is written without approval.", "parameters": {"type": "object", "required": ["title", "duration_minutes", "when"], "properties": {"title": {"type": "string"}, "duration_minutes": {"type": "integer"}, "when": {"type": "string", "description": "e.g. today afternoon, tomorrow 10:00"}, "item_id": {"type": "string", "description": "Optional linked triage item id."}}}}}, {"type": "function", "function": {"name": "list_calendar_events", "description": "List the user's calendar events for a day part.", "parameters": {"type": "object", "required": ["when"], "properties": {"when": {"type": "string", "description": "e.g. today, today afternoon, tomorrow"}}}}}, {"type": "function", "function": {"name": "get_sla_status", "description": "Summarize SLA state across the user's shared inboxes: breached and at-risk conversations.", "parameters": {"type": "object", "required": [], "properties": {}}}}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: slack_channel_message\nCHANNEL: #ai-team (no escalation keywords configured)\nFROM: Priya (Solutions)\nTIME: 10:05 today\nUSER CONTEXT: sole QA, AI team\nMESSAGE: \"@qa customer demo in 2 hours and the dataset upload is broken on their instance — can someone look?\""}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"critical\", \"severity_rationale\": \"Customer-facing breakage with near-term business impact, even though no escalation keyword appears.\", \"tier\": \"P1\", \"owner\": \"me\", \"why_it_matters\": \"A customer-visible failure damages the account directly if not confirmed and routed fast.\", \"next_step\": \"Reproduce the dataset upload failure on the customer instance and report findings to Priya\", \"deadline_suggestion\": \"today\", \"escalation_flag\": true}"}]}
	{"messages": [{"role": "system", "content": "You are UnchaosLM, the conversational assistant of Triage Copilot. You help the user act on their triaged work: tasks, escalations, priorities, people, inboxes, and calendar. When a request needs live data or an action, call the matching tool — extract the arguments from the user's message itself, even when phrased casually. After a tool returns, answer grounded strictly in its result and say plainly what you did (which tool you used and what it found). If a required argument is genuinely absent, ask one short question instead of guessing. Greetings and questions about your abilities need no tool. Never invent items, ids, owners, or links."}, {"role": "user", "content": "are the support queues okay?"}, {"role": "assistant", "tool_calls": [{"function": {"name": "get_sla_status", "arguments": {}}}], "content": ""}, {"role": "tool", "content": "{\"breached\": [{\"conversation\": 4006, \"customer\": \"Dana W.\", \"over_by\": \"35m\"}], \"at_risk\": [{\"conversation\": 5006, \"customer\": \"Grace N.\", \"due_in\": \"25m\"}]}"}, {"role": "assistant", "content": "I ran get_sla_status — 1 conversation(s) already breached (Dana W. over by 35m) and 1 at risk. Answer the breached ones first."}], "tools": [{"type": "function", "function": {"name": "list_items", "description": "List the user's active triage items, optionally filtered by tier (P1-P4) or severity. Returns id, tier, severity, title, owner, next step, deadline, and link.", "parameters": {"type": "object", "required": [], "properties": {"tier": {"type": "string", "enum": ["P1", "P2", "P3", "P4"]}, "severity": {"type": "string", "enum": ["Critical", "High", "Medium", "Low"]}}}}}, {"type": "function", "function": {"name": "get_item", "description": "Fetch one active triage item by its id (fingerprint).", "parameters": {"type": "object", "required": ["id"], "properties": {"id": {"type": "string", "description": "The item id / fingerprint."}}}}}, {"type": "function", "function": {"name": "resolve_person", "description": "Resolve a person's name or handle to a Slack user id. Returns resolved, ambiguous (with candidates), or not_found.", "parameters": {"type": "object", "required": ["name"], "properties": {"name": {"type": "string", "description": "Person name, @handle, or email."}}}}}, {"type": "function", "function": {"name": "propose_calendar_block", "description": "Propose a calendar time block for the user to approve. Nothing is written without approval.", "parameters": {"type": "object", "required": ["title", "duration_minutes", "when"], "properties": {"title": {"type": "string"}, "duration_minutes": {"type": "integer"}, "when": {"type": "string", "description": "e.g. today afternoon, tomorrow 10:00"}, "item_id": {"type": "string", "description": "Optional linked triage item id."}}}}}, {"type": "function", "function": {"name": "list_calendar_events", "description": "List the user's calendar events for a day part.", "parameters": {"type": "object", "required": ["when"], "properties": {"when": {"type": "string", "description": "e.g. today, today afternoon, tomorrow"}}}}}, {"type": "function", "function": {"name": "get_sla_status", "description": "Summarize SLA state across the user's shared inboxes: breached and at-risk conversations.", "parameters": {"type": "object", "required": [], "properties": {}}}}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: slack_dm\nFROM: Neha (CSM)\nTIME: today 10:15\nUSER CONTEXT: product manager, AI team\nMESSAGE: \"can you help me handle Kite Systems? they say the fine-tune dashboard behavior is a dealbreaker for renewal\""}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"critical\", \"severity_rationale\": \"A named account is escalating with churn or launch risk attached \\u2014 direct customer impact.\", \"tier\": \"P1\", \"owner\": \"me\", \"why_it_matters\": \"Kite Systems is at risk and needs a product commitment, not just a support reply.\", \"next_step\": \"Join the Kite Systems escalation call and commit to a fine-tune dashboard plan\", \"deadline_suggestion\": \"today\", \"escalation_flag\": true}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: hiver_sla_alert\nCONVERSATION #9453 \| CUSTOMER: Kevin D. \| SUBJECT: custom ribbon bar issue\nALERT: resolution SLA breached 3 hours ago \| ASSIGNEE: me\nUSER CONTEXT: support agent, medals & ribbons e-commerce store"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"critical\", \"severity_rationale\": \"A resolution SLA has already been breached; every additional minute deepens the miss.\", \"tier\": \"P1\", \"owner\": \"me\", \"why_it_matters\": \"Kevin D.'s conversation is past its resolution commitment and counting.\", \"next_step\": \"Answer conversation #9453 immediately and apologize for the delay\", \"deadline_suggestion\": \"today\", \"escalation_flag\": true}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: slack_channel_message\nCHANNEL: #ai-team\nFROM: Nikhil (dev)\nTIME: today 11:30\nUSER CONTEXT: product manager, AI team\nMESSAGE: \"the PRD doesn't say what happens when the chat memory gets empty input — which behavior is intended?\""}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"medium\", \"severity_rationale\": \"Repeated clarifying questions mean the spec is ambiguous and developers are building on guesses.\", \"tier\": \"P2\", \"owner\": \"me\", \"why_it_matters\": \"An ambiguous chat memory spec turns into rework and delivery slip if not resolved now.\", \"next_step\": \"Answer the chat memory question in thread and update the PRD section\", \"deadline_suggestion\": \"today\", \"escalation_flag\": false}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: hiver_csat_rating\nRATING: 5/5 from Dana W. \| AGENT: Meera\nCOMMENT: \"really appreciated the quick turnaround\"\nUSER CONTEXT: claims support agent, insurance agency"}, {"role": "assistant", "content": "{\"is_actionable\": false, \"action_direction\": \"info_only\", \"severity\": \"low\", \"severity_rationale\": \"A positive rating with praise; no request or follow-up implied.\", \"tier\": \"P4\", \"owner\": \"none\", \"why_it_matters\": \"Nice signal for the team; nothing to act on.\", \"next_step\": \"No action needed\", \"deadline_suggestion\": \"this_week\", \"escalation_flag\": false}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: slack_channel_message\nCHANNEL: #ai-team\nFROM: Divya\nTIME: 12:00 today\nUSER CONTEXT: sole QA, AI team\nMESSAGE: \"reminder that sprint demo notes live in the shared doc\""}, {"role": "assistant", "content": "{\"is_actionable\": false, \"action_direction\": \"info_only\", \"severity\": \"low\", \"severity_rationale\": \"General FYI with no question, request, or risk directed at the user.\", \"tier\": \"P4\", \"owner\": \"none\", \"why_it_matters\": \"Informational chatter; no action exists.\", \"next_step\": \"No action needed\", \"deadline_suggestion\": \"this_week\", \"escalation_flag\": false}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: slack_dm\nFROM: Neha (CSM)\nTIME: today 10:15\nUSER CONTEXT: product manager, AI team\nMESSAGE: \"heads up, Kite Systems raised the eval-runs page issue to their leadership; we need a plan today\""}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"critical\", \"severity_rationale\": \"A named account is escalating with churn or launch risk attached \\u2014 direct customer impact.\", \"tier\": \"P1\", \"owner\": \"me\", \"why_it_matters\": \"Kite Systems is at risk and needs a product commitment, not just a support reply.\", \"next_step\": \"Join the Kite Systems escalation call and commit to a eval-runs page plan\", \"deadline_suggestion\": \"today\", \"escalation_flag\": true}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: linked_ticket\nENG TASK: \"SSO login bug for Orbita\" \| STATUS: open \| ASSIGNEE: Rohit (eng) \| LAST ENG UPDATE: 3 days ago\nCUSTOMER: Orbita has chased twice since\nUSER CONTEXT: support engineer, B2B SaaS"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_them\", \"severity\": \"high\", \"severity_rationale\": \"A customer-impacting bug has no engineering update in three days while the customer chases.\", \"tier\": \"P1\", \"owner\": \"Rohit\", \"why_it_matters\": \"Orbita is waiting on a fix and silence from us reads as neglect.\", \"next_step\": \"Push Rohit for an ETA and send Orbita an honest status update\", \"deadline_suggestion\": \"today\", \"escalation_flag\": true}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: slack_thread\nCHANNEL: #ai-team-bugs\nTHREAD HISTORY:\n[Mon 14:30] me: \"Flagging: prompt playground shows stale data after a re-run. Repro steps in the task. cc @Meera\"\n[Tue 11:20] Meera (dev): \"pushed to staging, should work now\"\nTIME NOW: Wed 08:30\nUSER CONTEXT: sole QA, AI team; I have not replied since Meera's message"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"medium\", \"severity_rationale\": \"The fix is claimed but unverified; the bug stays possibly-live until QA confirms it.\", \"tier\": \"P2\", \"owner\": \"me\", \"why_it_matters\": \"An unverified fix can ship broken; verification is the open action and it is mine.\", \"next_step\": \"Verify Meera's fix for the prompt playground bug and reply on the thread\", \"deadline_suggestion\": \"today\", \"escalation_flag\": false}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: slack_dm\nFROM: Meera (dev)\nTIME: today 10:15\nUSER CONTEXT: senior developer, AI team\nMESSAGE: \"please prioritize my PR — it's the CART-1734 customer fix and the window closes at 2\""}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"high\", \"severity_rationale\": \"A customer-impact fix is blocked solely on this reviewer's approval.\", \"tier\": \"P1\", \"owner\": \"me\", \"why_it_matters\": \"Every hour of review delay extends Kite Systems's outage exposure.\", \"next_step\": \"Review and approve the CART-1734 fix PR this morning\", \"deadline_suggestion\": \"today\", \"escalation_flag\": true}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: hiver_shared_inbox_email\nINBOX: admissions@ \| FROM: Emily A. (customer)\nTIME: 2 days ago 09:15\nMESSAGE: \"the module 4 quiz arrived wrong and I need the corrected one before an exam sitting next week\"\nUSER CONTEXT: support agent, online course provider"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"high\", \"severity_rationale\": \"A customer has a defective or missing module 4 quiz with a hard event date approaching.\", \"tier\": \"P2\", \"owner\": \"me\", \"why_it_matters\": \"Emily A. cannot make an exam sitting unless the module 4 quiz is corrected in time.\", \"next_step\": \"Confirm replacement module 4 quiz and expedited shipping to Emily A.\", \"deadline_suggestion\": \"today\", \"escalation_flag\": false}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: hiver_shared_inbox_email\nINBOX: admissions@ \| FROM: Emily A. (customer)\nTIME: 2 days ago 09:15\nMESSAGE: \"third email about my course login and still no reply. If I don't hear back today I'm disputing the charge\"\nUSER CONTEXT: support agent, online course provider"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"critical\", \"severity_rationale\": \"A repeat-contact customer is threatening a chargeback or cancellation over an unresolved issue.\", \"tier\": \"P1\", \"owner\": \"me\", \"why_it_matters\": \"An escalating course login complaint is at the point of churn and public complaint.\", \"next_step\": \"Reply to Emily A. now with a concrete fix and timeline\", \"deadline_suggestion\": \"today\", \"escalation_flag\": true}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: hiver_shared_inbox_email\nINBOX: claims@ \| FROM: Raj V. (customer)\nTIME: today 10:40\nMESSAGE: \"just confirming I received the claim form — no action needed\"\nUSER CONTEXT: claims support agent, insurance agency"}, {"role": "assistant", "content": "{\"is_actionable\": false, \"action_direction\": \"info_only\", \"severity\": \"low\", \"severity_rationale\": \"A courtesy confirmation with explicit no-action language and no open request.\", \"tier\": \"P4\", \"owner\": \"none\", \"why_it_matters\": \"Positive close-out note; nothing to act on.\", \"next_step\": \"No action needed\", \"deadline_suggestion\": \"this_week\", \"escalation_flag\": false}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: slack_channel_message\nCHANNEL: #general\nFROM: Ayaan (teammate)\nTIME: today 09:20\nUSER CONTEXT: support engineer, B2B SaaS\nMESSAGE: \"urgent: standup moved to 10:30 today, conference room double-booked\""}, {"role": "assistant", "content": "{\"is_actionable\": false, \"action_direction\": \"info_only\", \"severity\": \"low\", \"severity_rationale\": \"Urgent-sounding wording, but it is internal logistics with no impact and no action owed.\", \"tier\": \"P4\", \"owner\": \"none\", \"why_it_matters\": \"Logistics notice to note; nothing to act on.\", \"next_step\": \"No action needed\", \"deadline_suggestion\": \"today\", \"escalation_flag\": false}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: hiver_mention\nCONVERSATION #8054 \| FROM: Neha (teammate) \| TIME: today, 20 minutes ago\nNOTE: \"@me urgent — Helen C. says the beneficiary update failure is blocking their whole team, can you jump in?\"\nUSER CONTEXT: claims support agent, insurance agency"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"critical\", \"severity_rationale\": \"A teammate flags a customer-blocking failure and asks for immediate help.\", \"tier\": \"P1\", \"owner\": \"me\", \"why_it_matters\": \"Helen C.'s team is blocked on the beneficiary update failure right now.\", \"next_step\": \"Join conversation #8054 and triage the beneficiary update failure\", \"deadline_suggestion\": \"today\", \"escalation_flag\": true}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: hiver_shared_inbox_email\nINBOX: admissions@ \| FROM: Dana W. (customer)\nTIME: today 10:40\nMESSAGE: \"received a damaged certificate download; I need a replacement in time for a license renewal deadline\"\nUSER CONTEXT: support agent, online course provider"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"high\", \"severity_rationale\": \"A customer has a defective or missing certificate download with a hard event date approaching.\", \"tier\": \"P2\", \"owner\": \"me\", \"why_it_matters\": \"Dana W. cannot make a license renewal deadline unless the certificate download is corrected in time.\", \"next_step\": \"Confirm replacement certificate download and expedited shipping to Dana W.\", \"deadline_suggestion\": \"today\", \"escalation_flag\": false}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: hiver_shared_inbox_email\nINBOX: orders@ \| FROM: Jenny F. (customer)\nTIME: today 08:10\nMESSAGE: \"my shadow box frame is missing from the delivery — a memorial service is coming up, can you help?\"\nUSER CONTEXT: support agent, medals & ribbons e-commerce store"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"high\", \"severity_rationale\": \"A customer has a defective or missing shadow box frame with a hard event date approaching.\", \"tier\": \"P2\", \"owner\": \"me\", \"why_it_matters\": \"Jenny F. cannot make a memorial service unless the shadow box frame is corrected in time.\", \"next_step\": \"Confirm replacement shadow box frame and expedited shipping to Jenny F.\", \"deadline_suggestion\": \"today\", \"escalation_flag\": false}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: clickup_task\nTASK: \"chat memory regression\" \| STATUS: in-progress \| ASSIGNEE: Sana (dev) \| DUE: none \| TAGS: none\nCOMMENTS:\n[yesterday 16:10] Sana: \"@qa was this on staging or prod data? and which browser?\"\n(no reply from me)\nUSER CONTEXT: sole QA, AI team"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"medium\", \"severity_rationale\": \"The assigned dev is blocked on details only I can provide about the repro.\", \"tier\": \"P2\", \"owner\": \"me\", \"why_it_matters\": \"Sana cannot progress the chat memory fix until I share the repro details.\", \"next_step\": \"Comment the requested repro details on Sana's task\", \"deadline_suggestion\": \"today\", \"escalation_flag\": false}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: clickup_task\nTASK: \"QA: fine-tune dashboard test pass\" \| STATUS: in-progress \| ASSIGNEE: me \| DUE: yesterday \| TAGS: none\nCOMMENTS:\n[2 days ago] me: \"completed the full run, no regressions found\"\nUSER CONTEXT: sole QA, AI team; standup at 10:00; the work is done but the status was never updated"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"low\", \"severity_rationale\": \"Work is done; only the status update is missing, but it misleads the board before standup.\", \"tier\": \"P4\", \"owner\": \"me\", \"why_it_matters\": \"A stale in-progress status makes finished work look late at standup.\", \"next_step\": \"Mark the fine-tune dashboard test task complete in ClickUp before standup\", \"deadline_suggestion\": \"today\", \"escalation_flag\": false}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: slack_thread\nCHANNEL: #ai-team\nTHREAD HISTORY:\n[Yesterday 15:02] Nikhil (dev): \"@qa any blockers from QA side for tomorrow's release?\"\n[Yesterday 15:40] me: \"no blockers, everything on the list is verified\"\n[Yesterday 15:41] Nikhil: \"perfect, thanks!\"\nTIME NOW: today 08:30\nUSER CONTEXT: sole QA, AI team"}, {"role": "assistant", "content": "{\"is_actionable\": false, \"action_direction\": \"info_only\", \"severity\": \"low\", \"severity_rationale\": \"The question was asked and fully answered yesterday; nothing remains open.\", \"tier\": \"P4\", \"owner\": \"none\", \"why_it_matters\": \"The thread is closed; surfacing it would be noise.\", \"next_step\": \"No action needed\", \"deadline_suggestion\": \"this_week\", \"escalation_flag\": false}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: hiver_shared_inbox_email\nINBOX: admissions@ \| FROM: Robert L. (customer)\nTIME: yesterday 17:55\nMESSAGE: \"my proctored exam slot is missing from the delivery — a course expiry date is coming up, can you help?\"\nUSER CONTEXT: support agent, online course provider"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"high\", \"severity_rationale\": \"A customer has a defective or missing proctored exam slot with a hard event date approaching.\", \"tier\": \"P2\", \"owner\": \"me\", \"why_it_matters\": \"Robert L. cannot make a course expiry date unless the proctored exam slot is corrected in time.\", \"next_step\": \"Confirm replacement proctored exam slot and expedited shipping to Robert L.\", \"deadline_suggestion\": \"today\", \"escalation_flag\": false}"}]}
	{"messages": [{"role": "system", "content": "You are UnchaosLM, the conversational assistant of Triage Copilot. You help the user act on their triaged work: tasks, escalations, priorities, people, inboxes, and calendar. When a request needs live data or an action, call the matching tool — extract the arguments from the user's message itself, even when phrased casually. After a tool returns, answer grounded strictly in its result and say plainly what you did (which tool you used and what it found). If a required argument is genuinely absent, ask one short question instead of guessing. Greetings and questions about your abilities need no tool. Never invent items, ids, owners, or links."}, {"role": "user", "content": "what fires are burning"}, {"role": "assistant", "tool_calls": [{"function": {"name": "list_items", "arguments": {"tier": "P1"}}}], "content": ""}], "tools": [{"type": "function", "function": {"name": "list_items", "description": "List the user's active triage items, optionally filtered by tier (P1-P4) or severity. Returns id, tier, severity, title, owner, next step, deadline, and link.", "parameters": {"type": "object", "required": [], "properties": {"tier": {"type": "string", "enum": ["P1", "P2", "P3", "P4"]}, "severity": {"type": "string", "enum": ["Critical", "High", "Medium", "Low"]}}}}}, {"type": "function", "function": {"name": "get_item", "description": "Fetch one active triage item by its id (fingerprint).", "parameters": {"type": "object", "required": ["id"], "properties": {"id": {"type": "string", "description": "The item id / fingerprint."}}}}}, {"type": "function", "function": {"name": "resolve_person", "description": "Resolve a person's name or handle to a Slack user id. Returns resolved, ambiguous (with candidates), or not_found.", "parameters": {"type": "object", "required": ["name"], "properties": {"name": {"type": "string", "description": "Person name, @handle, or email."}}}}}, {"type": "function", "function": {"name": "propose_calendar_block", "description": "Propose a calendar time block for the user to approve. Nothing is written without approval.", "parameters": {"type": "object", "required": ["title", "duration_minutes", "when"], "properties": {"title": {"type": "string"}, "duration_minutes": {"type": "integer"}, "when": {"type": "string", "description": "e.g. today afternoon, tomorrow 10:00"}, "item_id": {"type": "string", "description": "Optional linked triage item id."}}}}}, {"type": "function", "function": {"name": "list_calendar_events", "description": "List the user's calendar events for a day part.", "parameters": {"type": "object", "required": ["when"], "properties": {"when": {"type": "string", "description": "e.g. today, today afternoon, tomorrow"}}}}}, {"type": "function", "function": {"name": "get_sla_status", "description": "Summarize SLA state across the user's shared inboxes: breached and at-risk conversations.", "parameters": {"type": "object", "required": [], "properties": {}}}}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: slack_dm\nFROM: Rahul (Support Engineer)\nTIME: 11:40 today\nUSER CONTEXT: sole QA, AI team\nMESSAGE: \"P0 from support: customer completely stuck on the dataset upload. free for 15 min?\""}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"critical\", \"severity_rationale\": \"Support engineer reports a customer actively blocked and needs immediate repro help.\", \"tier\": \"P1\", \"owner\": \"me\", \"why_it_matters\": \"A customer is blocked right now; the repro is the gate to the fix.\", \"next_step\": \"Huddle with Rahul now to reproduce the dataset upload issue\", \"deadline_suggestion\": \"today\", \"escalation_flag\": true}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: hiver_shared_inbox_email\nINBOX: claims@ \| FROM: Jenny F. (customer)\nTIME: today 09:25\nMESSAGE: \"received a damaged claim status; I need a replacement in time for a claim deadline\"\nUSER CONTEXT: claims support agent, insurance agency"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"high\", \"severity_rationale\": \"A customer has a defective or missing claim status with a hard event date approaching.\", \"tier\": \"P2\", \"owner\": \"me\", \"why_it_matters\": \"Jenny F. cannot make a claim deadline unless the claim status is corrected in time.\", \"next_step\": \"Confirm replacement claim status and expedited shipping to Jenny F.\", \"deadline_suggestion\": \"today\", \"escalation_flag\": false}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: health_score_alert\nACCOUNT: Northwind \| SCORE: dropped 82 -> 61 this month\nDRIVER: logins down 40%, CSV export usage near zero\nOWNER: me\nUSER CONTEXT: customer success manager"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"high\", \"severity_rationale\": \"A sharp health-score drop with collapsing usage is an early churn signal.\", \"tier\": \"P1\", \"owner\": \"me\", \"why_it_matters\": \"Northwind's disengagement is measurable now; waiting turns a dip into a cancellation.\", \"next_step\": \"Investigate Northwind's usage drop and schedule a check-in call\", \"deadline_suggestion\": \"tomorrow\", \"escalation_flag\": true}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: slack_thread\nCHANNEL: #ai-team-bugs\nTHREAD HISTORY:\n[Mon 14:30] me: \"Flagging: webhook alerts shows stale data after a re-run. Repro steps in the task. cc @Tanvi\"\n[Tue 11:20] Tanvi (dev): \"should be good now\"\nTIME NOW: Wed 08:30\nUSER CONTEXT: sole QA, AI team; I have not replied since Tanvi's message"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"medium\", \"severity_rationale\": \"The fix is claimed but unverified; the bug stays possibly-live until QA confirms it.\", \"tier\": \"P2\", \"owner\": \"me\", \"why_it_matters\": \"An unverified fix can ship broken; verification is the open action and it is mine.\", \"next_step\": \"Verify Tanvi's fix for the webhook alerts bug and reply on the thread\", \"deadline_suggestion\": \"today\", \"escalation_flag\": false}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: hiver_sla_alert\nCONVERSATION #1431 \| CUSTOMER: Helen C. \| SUBJECT: automation rule issue\nALERT: first-response SLA due in 45 minutes \| ASSIGNEE: me\nUSER CONTEXT: support engineer, B2B SaaS (shared inbox product)"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"high\", \"severity_rationale\": \"The first-response SLA window closes in 45 minutes; a breach is avoidable but imminent.\", \"tier\": \"P2\", \"owner\": \"me\", \"why_it_matters\": \"Responding to Helen C. within 45 minutes keeps the SLA record clean.\", \"next_step\": \"Send a first response on conversation #1431 within 45 minutes\", \"deadline_suggestion\": \"today\", \"escalation_flag\": false}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: slack_dm\nFROM: Suresh (hiring manager)\nTIME: yesterday 15:10\nUSER CONTEXT: talent acquisition recruiter\nMESSAGE: \"please schedule S. Iyer's final round soon, we don't want to lose momentum\""}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"medium\", \"severity_rationale\": \"A hiring manager has asked for an interview loop to be scheduled this week.\", \"tier\": \"P2\", \"owner\": \"me\", \"why_it_matters\": \"Scheduling lag is the most common self-inflicted pipeline delay.\", \"next_step\": \"Send S. Iyer's panel invites and confirm the loop this week\", \"deadline_suggestion\": \"this_week\", \"escalation_flag\": false}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: hiver_mention\nCONVERSATION #5712 \| FROM: Kavya (teammate) \| TIME: today 13:30\nNOTE: \"@me can you take this one? Tom H. is asking about the claim status and you handled their last case\"\nUSER CONTEXT: claims support agent, insurance agency"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"medium\", \"severity_rationale\": \"A teammate has handed over a customer conversation and is waiting for pickup.\", \"tier\": \"P2\", \"owner\": \"me\", \"why_it_matters\": \"Tom H.'s claim status question sits unowned until the handover is accepted.\", \"next_step\": \"Accept conversation #5712 and reply to Tom H.\", \"deadline_suggestion\": \"today\", \"escalation_flag\": false}"}]}
	{"messages": [{"role": "system", "content": "You are UnchaosLM, the conversational assistant of Triage Copilot. You help the user act on their triaged work: tasks, escalations, priorities, people, inboxes, and calendar. When a request needs live data or an action, call the matching tool — extract the arguments from the user's message itself, even when phrased casually. After a tool returns, answer grounded strictly in its result and say plainly what you did (which tool you used and what it found). If a required argument is genuinely absent, ask one short question instead of guessing. Greetings and questions about your abilities need no tool. Never invent items, ids, owners, or links."}, {"role": "user", "content": "what's zubin's handle?"}, {"role": "assistant", "tool_calls": [{"function": {"name": "resolve_person", "arguments": {"name": "zubin"}}}], "content": ""}, {"role": "tool", "content": "{\"status\": \"not_found\"}"}, {"role": "assistant", "content": "I checked with resolve_person — no Slack user matching \"zubin\". Could you give the full name or email?"}], "tools": [{"type": "function", "function": {"name": "list_items", "description": "List the user's active triage items, optionally filtered by tier (P1-P4) or severity. Returns id, tier, severity, title, owner, next step, deadline, and link.", "parameters": {"type": "object", "required": [], "properties": {"tier": {"type": "string", "enum": ["P1", "P2", "P3", "P4"]}, "severity": {"type": "string", "enum": ["Critical", "High", "Medium", "Low"]}}}}}, {"type": "function", "function": {"name": "get_item", "description": "Fetch one active triage item by its id (fingerprint).", "parameters": {"type": "object", "required": ["id"], "properties": {"id": {"type": "string", "description": "The item id / fingerprint."}}}}}, {"type": "function", "function": {"name": "resolve_person", "description": "Resolve a person's name or handle to a Slack user id. Returns resolved, ambiguous (with candidates), or not_found.", "parameters": {"type": "object", "required": ["name"], "properties": {"name": {"type": "string", "description": "Person name, @handle, or email."}}}}}, {"type": "function", "function": {"name": "propose_calendar_block", "description": "Propose a calendar time block for the user to approve. Nothing is written without approval.", "parameters": {"type": "object", "required": ["title", "duration_minutes", "when"], "properties": {"title": {"type": "string"}, "duration_minutes": {"type": "integer"}, "when": {"type": "string", "description": "e.g. today afternoon, tomorrow 10:00"}, "item_id": {"type": "string", "description": "Optional linked triage item id."}}}}}, {"type": "function", "function": {"name": "list_calendar_events", "description": "List the user's calendar events for a day part.", "parameters": {"type": "object", "required": ["when"], "properties": {"when": {"type": "string", "description": "e.g. today, today afternoon, tomorrow"}}}}}, {"type": "function", "function": {"name": "get_sla_status", "description": "Summarize SLA state across the user's shared inboxes: breached and at-risk conversations.", "parameters": {"type": "object", "required": [], "properties": {}}}}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: hiver_sla_alert\nCONVERSATION #7026 \| CUSTOMER: Colin B. \| SUBJECT: freight quote issue\nALERT: resolution SLA breached 3 hours ago \| ASSIGNEE: me\nUSER CONTEXT: support agent, logistics & shipping"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"critical\", \"severity_rationale\": \"A resolution SLA has already been breached; every additional minute deepens the miss.\", \"tier\": \"P1\", \"owner\": \"me\", \"why_it_matters\": \"Colin B.'s conversation is past its resolution commitment and counting.\", \"next_step\": \"Answer conversation #7026 immediately and apologize for the delay\", \"deadline_suggestion\": \"today\", \"escalation_flag\": true}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: hiver_csat_rating\nRATING: 2/5 from Marcus J. \| AGENT: Nikhil\nCOMMENT: \"took too long and the course extension answer didn't solve my problem\"\nUSER CONTEXT: support agent, online course provider"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"medium\", \"severity_rationale\": \"A low rating with a concrete complaint is recoverable if followed up quickly.\", \"tier\": \"P2\", \"owner\": \"me\", \"why_it_matters\": \"Marcus J.'s course extension problem is still unsolved despite a closed conversation.\", \"next_step\": \"Reopen Marcus J.'s conversation and resolve the course extension problem properly\", \"deadline_suggestion\": \"tomorrow\", \"escalation_flag\": false}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: hiver_assignment\nCONVERSATION #3545 \| CUSTOMER: Miguel T. \| SUBJECT: lapel pin order problem\nSTATUS: open, assigned to me 5 days ago, no reply sent yet\nUSER CONTEXT: support agent, medals & ribbons e-commerce store"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"high\", \"severity_rationale\": \"An assigned conversation has waited 5 days without a single reply.\", \"tier\": \"P2\", \"owner\": \"me\", \"why_it_matters\": \"Miguel T. has had silence for 5 days on an open lapel pin order problem.\", \"next_step\": \"Send Miguel T. a first reply on the lapel pin order problem\", \"deadline_suggestion\": \"today\", \"escalation_flag\": false}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: hiver_sla_alert\nCONVERSATION #9056 \| CUSTOMER: Omar S. \| SUBJECT: CE credit record issue\nALERT: first-response SLA breached 3 hours ago \| ASSIGNEE: me\nUSER CONTEXT: support agent, online course provider"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"critical\", \"severity_rationale\": \"A first-response SLA has already been breached; every additional minute deepens the miss.\", \"tier\": \"P1\", \"owner\": \"me\", \"why_it_matters\": \"Omar S.'s conversation is past its first-response commitment and counting.\", \"next_step\": \"Answer conversation #9056 immediately and apologize for the delay\", \"deadline_suggestion\": \"today\", \"escalation_flag\": true}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: hiver_shared_inbox_email\nINBOX: support@ \| FROM: Robert L. (customer)\nTIME: today 13:30\nMESSAGE: \"this is unacceptable — the user seat issue has been open for a week. I want a manager and a refund\"\nUSER CONTEXT: support engineer, B2B SaaS (shared inbox product)"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"critical\", \"severity_rationale\": \"A repeat-contact customer is threatening a chargeback or cancellation over an unresolved issue.\", \"tier\": \"P1\", \"owner\": \"me\", \"why_it_matters\": \"An escalating user seat complaint is at the point of churn and public complaint.\", \"next_step\": \"Reply to Robert L. now with a concrete fix and timeline\", \"deadline_suggestion\": \"today\", \"escalation_flag\": true}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: clickup_task\nTASK: \"Verify fix: chat memory (CART-1473)\" \| STATUS: qa-ready \| ASSIGNEE: me \| DUE: today \| TAGS: CART, customer-impact\nCOMMENTS:\n[yesterday 18:05] Sana (dev): \"deployed to staging, ready for QA — customer is waiting on this in tomorrow's release\"\nUSER CONTEXT: sole QA, AI team; deployment window closes early afternoon"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"high\", \"severity_rationale\": \"CART-tagged customer-impact fix must be verified before the deployment window to make the release.\", \"tier\": \"P1\", \"owner\": \"me\", \"why_it_matters\": \"A waiting customer misses the release if verification slips past the deployment window.\", \"next_step\": \"Verify the chat memory fix on staging before the deployment window closes\", \"deadline_suggestion\": \"today\", \"escalation_flag\": true}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: slack_dm\nFROM: Suresh (hiring manager)\nTIME: yesterday 15:10\nUSER CONTEXT: talent acquisition recruiter\nMESSAGE: \"M. Kulkarni passed the screen — over to you for panel scheduling\""}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"medium\", \"severity_rationale\": \"A hiring manager has asked for an interview loop to be scheduled this week.\", \"tier\": \"P2\", \"owner\": \"me\", \"why_it_matters\": \"Scheduling lag is the most common self-inflicted pipeline delay.\", \"next_step\": \"Send M. Kulkarni's panel invites and confirm the loop this week\", \"deadline_suggestion\": \"this_week\", \"escalation_flag\": false}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: offboarding_checklist\nEMPLOYEE: Ritika \| LAST DAY: last Friday\nSTATUS: 2 of 6 access revocations UNCONFIRMED (VPN, admin console)\nUSER CONTEXT: IT administrator"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"high\", \"severity_rationale\": \"A departed employee still has unconfirmed access to two systems \\u2014 an open security hole.\", \"tier\": \"P1\", \"owner\": \"me\", \"why_it_matters\": \"Lingering access after departure is a textbook breach vector and audit finding.\", \"next_step\": \"Confirm and close all access revocations for Ritika today\", \"deadline_suggestion\": \"today\", \"escalation_flag\": true}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: hiver_mention\nCONVERSATION #6225 \| MY NOTE 4 days ago: \"@Tara can you check the ribbon rack logs for Dana W.?\"\nSTATUS: no response from teammate yet\nUSER CONTEXT: support agent, medals & ribbons e-commerce store"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_them\", \"severity\": \"medium\", \"severity_rationale\": \"An internal ask has sat unanswered for 4 days with no committed date.\", \"tier\": \"P3\", \"owner\": \"Tara\", \"why_it_matters\": \"Dana W.'s case cannot progress until Tara checks the ribbon rack logs.\", \"next_step\": \"Nudge Tara on the ribbon rack logs for conversation #6225\", \"deadline_suggestion\": \"this_week\", \"escalation_flag\": false}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: hiver_shared_inbox_email\nINBOX: admissions@ \| FROM: Tom H. (customer)\nTIME: today 09:25\nMESSAGE: \"whenever you get a chance: can I change my enrollment confirmation after ordering?\"\nUSER CONTEXT: support agent, online course provider"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"low\", \"severity_rationale\": \"A routine how-to question with no urgency signals or deadline pressure.\", \"tier\": \"P4\", \"owner\": \"me\", \"why_it_matters\": \"A simple answer keeps Tom H. moving; nothing depends on it today.\", \"next_step\": \"Send Tom H. a short how-to reply on the enrollment confirmation\", \"deadline_suggestion\": \"this_week\", \"escalation_flag\": false}"}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: slack_dm\nFROM: Vikram (dev)\nTIME: today 10:15\nUSER CONTEXT: product manager, AI team\nMESSAGE: \"whichever you pick is fine, but we need the webhook alerts decision today to stay on track\""}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"high\", \"severity_rationale\": \"The team is explicitly blocked on a decision only the PM can make, with near-term deadlines.\", \"tier\": \"P2\", \"owner\": \"me\", \"why_it_matters\": \"Every day the webhook alerts decision waits, the whole team idles or builds the wrong thing.\", \"next_step\": \"Make the webhook alerts decision and post it to the team today\", \"deadline_suggestion\": \"today\", \"escalation_flag\": false}"}]}
	{"messages": [{"role": "system", "content": "You are UnchaosLM, the conversational assistant of Triage Copilot. You help the user act on their triaged work: tasks, escalations, priorities, people, inboxes, and calendar. When a request needs live data or an action, call the matching tool — extract the arguments from the user's message itself, even when phrased casually. After a tool returns, answer grounded strictly in its result and say plainly what you did (which tool you used and what it found). If a required argument is genuinely absent, ask one short question instead of guessing. Greetings and questions about your abilities need no tool. Never invent items, ids, owners, or links."}, {"role": "user", "content": "anything low priority I can batch later, please?"}, {"role": "assistant", "tool_calls": [{"function": {"name": "list_items", "arguments": {"tier": "P4"}}}], "content": ""}, {"role": "tool", "content": "{\"items\": [{\"id\": \"cu021\", \"tier\": \"P4\", \"severity\": \"Low\", \"title\": \"Refactor billing page components\", \"owner\": \"me\", \"next\": \"Review the refactor PR when free\", \"by\": \"this week\", \"link\": \"https://app.clickup.com/t/cu021\", \"source\": \"clickup\"}]}"}, {"role": "assistant", "content": "I pulled your P4 triage items with list_items — 1 need you:\n• [P4/Low] Refactor billing page components — owner me, next: Review the refactor PR when free (due this week)\nStart with Refactor billing page components: review the refactor pr when free."}], "tools": [{"type": "function", "function": {"name": "list_items", "description": "List the user's active triage items, optionally filtered by tier (P1-P4) or severity. Returns id, tier, severity, title, owner, next step, deadline, and link.", "parameters": {"type": "object", "required": [], "properties": {"tier": {"type": "string", "enum": ["P1", "P2", "P3", "P4"]}, "severity": {"type": "string", "enum": ["Critical", "High", "Medium", "Low"]}}}}}, {"type": "function", "function": {"name": "get_item", "description": "Fetch one active triage item by its id (fingerprint).", "parameters": {"type": "object", "required": ["id"], "properties": {"id": {"type": "string", "description": "The item id / fingerprint."}}}}}, {"type": "function", "function": {"name": "resolve_person", "description": "Resolve a person's name or handle to a Slack user id. Returns resolved, ambiguous (with candidates), or not_found.", "parameters": {"type": "object", "required": ["name"], "properties": {"name": {"type": "string", "description": "Person name, @handle, or email."}}}}}, {"type": "function", "function": {"name": "propose_calendar_block", "description": "Propose a calendar time block for the user to approve. Nothing is written without approval.", "parameters": {"type": "object", "required": ["title", "duration_minutes", "when"], "properties": {"title": {"type": "string"}, "duration_minutes": {"type": "integer"}, "when": {"type": "string", "description": "e.g. today afternoon, tomorrow 10:00"}, "item_id": {"type": "string", "description": "Optional linked triage item id."}}}}}, {"type": "function", "function": {"name": "list_calendar_events", "description": "List the user's calendar events for a day part.", "parameters": {"type": "object", "required": ["when"], "properties": {"when": {"type": "string", "description": "e.g. today, today afternoon, tomorrow"}}}}}, {"type": "function", "function": {"name": "get_sla_status", "description": "Summarize SLA state across the user's shared inboxes: breached and at-risk conversations.", "parameters": {"type": "object", "required": [], "properties": {}}}}]}
	{"messages": [{"role": "system", "content": "You are the extraction layer of Triage Copilot. Input: one candidate work signal (chat message/thread, task, ticket, or system alert) plus user context. Output: strict JSON only, matching this schema: {\"is_actionable\": bool, \"action_direction\": \"waiting_on_me\|waiting_on_them\|info_only\", \"severity\": \"critical\|high\|medium\|low\", \"severity_rationale\": str, \"tier\": \"P1\|P2\|P3\|P4\", \"owner\": str, \"why_it_matters\": str, \"next_step\": str (max 18 words, an action not a summary), \"deadline_suggestion\": \"today\|tomorrow\|this_week\", \"escalation_flag\": bool}. Rules: high-stakes escalations — customer impact, SLA breach, security incident, employee-critical issue, or offer-at-risk — are P1 with escalation_flag true; when unsure, flag it (recall over precision). No date and no stated timeline contributes to P3 no-ETA. Never invent links, owners, or tasks not present in the input."}, {"role": "user", "content": "SIGNAL TYPE: hiver_mention\nCONVERSATION #6703 \| FROM: Sameer (teammate) \| TIME: today, 20 minutes ago\nNOTE: \"@me urgent — Frank O. says the lapel pin order failure is blocking their whole team, can you jump in?\"\nUSER CONTEXT: support agent, medals & ribbons e-commerce store"}, {"role": "assistant", "content": "{\"is_actionable\": true, \"action_direction\": \"waiting_on_me\", \"severity\": \"critical\", \"severity_rationale\": \"A teammate flags a customer-blocking failure and asks for immediate help.\", \"tier\": \"P1\", \"owner\": \"me\", \"why_it_matters\": \"Frank O.'s team is blocked on the lapel pin order failure right now.\", \"next_step\": \"Join conversation #6703 and triage the lapel pin order failure\", \"deadline_suggestion\": \"today\", \"escalation_flag\": true}"}]}