A FastAPI web application that uses the Claude Agent SDK to run iterative AI agents on GitHub repositories, generating formal clean room specifications with provable properties, purity boundaries, and verification tooling.

• Python 3.12
• uv package manager
• An Anthropic API key exported as ANTHROPIC_API_KEY

uv sync --extra dev

uv run uvicorn clean_room.main:app --reload --port 8317

Open http://localhost:8317 to access the dashboard.

1. Click Add Repo and paste a GitHub URL
2. On the repo detail page, select a prompt and click Create Job
3. Watch the agent generate specs in real time via the SSE log stream
4. Find generated specs in ~/.clean-room/specs-monorepo/

uv run pytest tests/ -v                        # Run tests
uv run ruff check src/ tests/                   # Lint
uv run mypy src/ --ignore-missing-imports       # Type check

This project uses Jujutsu (jj) for version control, not git.

Full documentation is in the docs/ directory, organized using the Diataxis framework:

• Tutorial: Getting Started -- Step-by-step first run walkthrough
• How-to Guides
  • Manage Prompts
  • Manage Repos
  • Run Jobs
  • Run Tests
• Reference
  • API Endpoints
  • Configuration
  • Database Schema
  • Project Structure
• Explanation: Architecture -- Design decisions and trade-offs