| name: CodeGuardian |
| description: A code review environment that tests AI agents on detecting security issues, assessing risks, and suggesting fixes |
| version: 1.0.0 |
|
|
| spaces: |
| - name: code_review |
| description: Code review and security analysis tasks |
| type: gymnasium |
| |
| tasks: |
| - id: easy_off_by_one |
| name: "Off-by-One Error" |
| difficulty: easy |
| description: "Detect off-by-one error in array summation" |
| space: code_review |
| |
| - id: medium_sql_injection |
| name: "SQL Injection" |
| difficulty: medium |
| description: "Identify SQL injection vulnerability in user query" |
| space: code_review |
| |
| - id: hard_async_race |
| name: "Async Race Condition" |
| difficulty: hard |
| description: "Find missing await and race condition in async code" |
| space: code_review |
|
|
| reward: |
| type: continuous |
| min: 0.0 |
| max: 1.0 |
| description: "Score based on correct issue detection, risk assessment, and fix suggestions" |
|
|
| environment: |
| class: CodeGuardianEnv |
| module: env.environment |
| observation_space: |
| type: dict |
| fields: |
| code: str |
| intent: str |
| constraints: list |
| task_id: str |
| difficulty: str |
| step_index: int |
| action_space: |
| type: dict |
| fields: |
| action_type: str |
| detected_issues: list |
| risk_level: str |
| suggested_fix: str |
|
|
| metadata: |
| author: CodeGuardian Team |
| tags: ["code-review", "security", "ai-testing"] |
| license: MIT |
|
|
