ashley222/Aurora-ASR-self-healing-CI-CD-AgentAurora-ASR-self-healing-CI-CD-Agent

Aurora ASR: The Self-Healing CI/CD Agent 🚀

Aurora ASR is a high-performance, OpenEnv-v4 compliant Automated Software Repair system. It leverages Reinforcement Learning and Large Language Models (LLMs) to autonomously detect and repair software bugs within isolated sandboxes.

🌟 Features

• —OpenEnv-v4 Compliant: Fully compatible with the Meta PyTorch Hackathon automated judging system ([OK] : Ready for multi-mode deployment).
• —Typed API: Uses strict Pydantic models for Observations, Actions, and Rewards.
• —3-Tier Task System: Includes Easy (Typo), Medium (Logic), and Hard (Edge Case) repair challenges.
• —Smart Fallback (Mock Mode): Automatically switches to a scripted Mock Agent if no API key is provided, allowing for offline demos and testing.
• —Structured Logging: Emits [START], [STEP], and [END] tags for real-time progress tracking and grading.
• —Dockerized Sandboxing: Safe execution of agent-generated code with resource limits.

🛠️ Installation

Using the Python Launcher for Windows (py):

# Install core dependencies
py -m pip install -r requirements.txt

# Install the OpenEnv core library
py -m pip install openenv-core

🎮 Running the Agent

Aurora uses inference.py as its primary execution entrypoint. It utilizes the OpenAI client to iterate through the 3 task tiers.

# Option A: Real LLM Mode (requires OpenAI API Key)
set OPENAI_API_KEY=sk-your-key-here
py inference.py

# Option B: Mock Demo Mode (no key required)
set OPENAI_API_KEY=mock
py inference.py

[!NOTE] Aurora will automatically detect if OPENAI_API_KEY is missing or invalid and fall back to the Mock Agent. To switch back to the real LLM, simply provide a valid API key in your environment variables.

🌳 Pydantic API Spec

Aurora adheres to the strict OpenEnv-v4 data schemas:

Observation (ASRObservation)

{
    "file_tree": "...",      # Directory structure
    "current_file": "...",   # Content with line numbers
    "test_results": {...},    # Pytest summary
    "reward": 0.0,           # Scalar reward
    "asr_reward": {...},      # Detailed Pydantic Reward Model
    "done": False            # Termination flag
}

Action (ASRAction)

{
    "command": "read_file",  # Options: read_file, write_file, run_pytest
    "params": {"path": "..."}
}

📊 Task Tiers

1. 1.Easy (`tasks/easy`): Fix a simple NameError (undefined variable).
2. 2.Medium (`tasks/medium`): Repair an off-by-one error in a factorial implementation.
3. 3.Hard (`tasks/hard`): Address a ZeroDivisionError in a complex data processor.

🛡️ Validation

Verify the repository structure and compliance using the OpenEnv CLI:

py -m openenv.cli validate .

🤝 Contributing

Contributions welcome! Please see ARCHITECTURE.md for deeper technical details.

📄 License

MIT License - See LICENSE file