{"data":{"kind":"file","path":"README.md","version_id":"hrbudhmgig7wh4zz719lmzst","entry":{"name":"README.md","path":"README.md","is_directory":false,"size":1876,"modified_at":"2026-09-14T11:07:32.277000","content_hash":"d3a5b02cc17a333245a797326b284ea2e8addebcdaae334636b7dfe928bccd0e"},"entries":[],"content":"# Software Debugging Evaluation Environment (v1.0.0)\n\nA high-quality benchmark and reinforcement learning environment built on Prime Intellect's `verifiers` framework. This environment evaluates an LLM agent's capability to diagnose software defects, pinpoint root causes, explain reproduction steps, and generate correct bug fixes across real-world bug categories (Off-By-One errors, Concurrency Race Conditions, Resource Leaks, Type/Null Handling, and Boundary Logic Flaws).\n\n## Key Features\n\n- **Multi-signal verification**: Evaluates root-cause analysis quality, explanation accuracy, executable patch validity, syntax correctness, and security/safety.\n- **Hardened true metrics**: Replaces naive keyword matching with multi-dimensional AST inspection, semantic pattern verification, and diff correctness checks.\n- **Diverse defect taxonomy**:\n  - `Off-By-One / Boundary Indexing`\n  - `Concurrency / Race Condition & Thread Safety`\n  - `Resource Management / Memory & File Descriptor Leaks`\n  - `Null / Optional / Type Unwrapping Errors`\n  - `State Invalidation & Cache Stale Reads`\n- **Full compatibility with Prime Intellect**: Designed for both zero-shot/few-shot benchmark evaluation and RL training loops (`prime eval` & `prime train`).\n\n## Structure\n\n```\nenvironments/software-debugging-eval/\n├── software_debugging_eval.py   # Core environment logic & load_environment entry point\n├── pyproject.toml               # Package metadata and dependencies\n└── README.md                    # Environment documentation\n```\n\n## Quick Start\n\n### Installation & Inspection\n```bash\n# Run local verification\npython3 -c \"from software_debugging_eval import load_environment; env = load_environment(); print(env)\"\n```\n\n### Running Evaluation\n```bash\nprime eval run philips/software-debugging-eval -m meta-llama/Llama-3.2-3B-Instruct\n```\n\n## License\nMIT License.\n","encoding":"utf-8","truncated":false,"total_bytes":1876},"status":null}