Spaces:
Sleeping
Sleeping
| name: PRReviewEnv | |
| version: "1.0.0" | |
| description: > | |
| OpenEnv environment for evaluating AI agents on pull request code review. | |
| Agents read git diffs, identify bugs and security issues, and make | |
| approve/reject decisions. | |
| tasks: | |
| - id: easy | |
| description: Find obvious bugs in short single-file diffs | |
| max_steps: 5 | |
| success_threshold: 0.7 | |
| - id: medium | |
| description: Find logic bugs and security issues in multi-file diffs | |
| max_steps: 10 | |
| success_threshold: 0.6 | |
| - id: hard | |
| description: Full review of large diffs with subtle cross-file bugs | |
| max_steps: 15 | |
| success_threshold: 0.5 | |
| observation_space: | |
| pr_title: string | |
| pr_description: string | |
| diff: string | |
| file_tree: list[string] | |
| comments_so_far: list[object] | |
| step_count: integer | |
| done: boolean | |
| action_space: | |
| action_type: "enum[comment, approve, request_changes]" | |
| file: "string (optional)" | |
| line: "integer (optional)" | |
| body: string | |
| reward_range: [0.0, 1.0] |