|
|
| name: CodeReview-Professional-Workflow
|
| version: 1.0.0
|
| description: |
|
| Multi‑turn code review environment for professional tasks.
|
| Agent must inspect, test, lint, query docs, and negotiate with a simulated author
|
| to fix injected bugs. Supports DPO training on full trajectories.
|
| author: yuvraj gupta
|
| license: MIT
|
|
|
|
|
|
|
|
|
| tasks:
|
| - id: easy
|
| description: "Fix missing null check in a dictionary lookup"
|
| - id: medium
|
| description: "Improve loop efficiency (replace range(len) with direct iteration)"
|
| - id: hard
|
| description: "Handle division by zero in average calculation"
|
| - id: harder
|
| description: "Fix race condition by adding a lock"
|
| - id: hardest
|
| description: "Resolve potential deadlock by standardising lock order"
|
|
|
|
|
|
|
|
|
| observation_space:
|
| type: object
|
| properties:
|
| code_snippet:
|
| type: string
|
| description: "Current code snippet (may contain injected bug)"
|
| last_tool_output:
|
| type: string
|
| description: "Raw output from last tool (test runner, linter, etc.)"
|
| author_response:
|
| type: string
|
| description: "Latest feedback from the simulated human developer"
|
| current_test_score:
|
| type: number
|
| description: "Proportion of tests passed (0.0–1.0)"
|
| current_lint_score:
|
| type: number
|
| description: "Normalised pylint score (0.0–1.0)"
|
| negotiation_score:
|
| type: number
|
| description: "Author's confidence minus pushback penalty"
|
| previous_test_score:
|
| type: number
|
| description: "Test score before the last action"
|
| previous_lint_score:
|
| type: number
|
| description: "Lint score before the last action"
|
| author_confidence:
|
| type: number
|
| description: "Internal belief of the author (0.0–1.0)"
|
| author_threshold:
|
| type: number
|
| description: "Confidence threshold for this personality"
|
| step:
|
| type: integer
|
| description: "Current step number"
|
| max_steps:
|
| type: integer
|
| description: "Maximum steps allowed in the episode"
|
| progress_ratio:
|
| type: number
|
| description: "step / max_steps"
|
| tests_run:
|
| type: boolean
|
| description: "Whether the agent has run tests at least once"
|
| linter_run:
|
| type: boolean
|
| description: "Whether the agent has run the linter at least once"
|
| docs_queried:
|
| type: boolean
|
| description: "Whether the agent has queried documentation"
|
| last_action_type:
|
| type: string
|
| description: "String name of the last executed action"
|
| action_history:
|
| type: array
|
| items:
|
| type: string
|
| description: "Last 5 action types"
|
| done:
|
| type: boolean
|
| description: "Whether the episode has finished"
|
| bug_description:
|
| type: string
|
| description: "Short description of the injected bug"
|
| comments_count:
|
| type: integer
|
| description: "Number of comments exchanged so far"
|
|
|
|
|
|
|
|
|
| action_space:
|
| type: object
|
| properties:
|
| action_type:
|
| type: string
|
| enum:
|
| - comment
|
| - skip
|
| - done
|
| - question
|
| - fix
|
| - execute
|
| - inspect
|
| - run_linter
|
| - run_tests
|
| - query_docs
|
| comment_text:
|
| type: string
|
| description: "Required for comment"
|
| question:
|
| type: string
|
| description: "Required for question"
|
| fix_code:
|
| type: string
|
| description: "Required for fix"
|
| query_topic:
|
| type: string
|
| description: "Required for query_docs"
|
|
|
|
|
|
|
|
|
| server:
|
| app: server.app:app
|
| port: 7860
|
|
|
