Stang

A 3-layer agentic architecture for reliable, deterministic task execution.

This project follows a strict 3-layer separation of concerns (see CLAUDE.md for full details):

Stang/
├── directives/       # Markdown SOPs (living documents)
├── execution/        # Python utility scripts
├── frontend/         # Next.js app (when applicable)
├── backend/          # FastAPI backend (when applicable)
├── .tmp/             # Intermediate processing files (never committed)
├── .env              # Secrets (never committed — see .env.example)
├── .env.example      # Template for required environment variables
├── .gitignore
├── CLAUDE.md         # Agent instructions & operating principles
└── README.md

1. Clone the repo
2. Copy .env.example to .env and fill in your values
3. Install Python dependencies (per script — see execution/requirements.txt if present)
4. Install frontend dependencies (if using the web app):

   cd frontend
   npm install
   npm run dev

• .tmp/ files are always regenerable — can be deleted freely
• Directives are updated as the agent learns (API limits, edge cases, etc.)
• Deliverables live in cloud services (Google Sheets, Firebase, etc.)
• Business logic lives in execution/ scripts, not in the agent