An AI-powered SDLC pipeline that takes a software project from idea to deployment-ready code. Ten phases — setup, discovery, PRD, tasking, specification, implementation, code review, integration, deployment prep, and release — each orchestrated by specialized agents with full state management, rollback, and memory.
# Runtime install (~23MB, pipeline essentials only)
curl -fsSL https://raw.githubusercontent.com/turbobeest/atomic-claude/python/install.sh | bash
# Full install (includes tests, docs, dev tooling)
curl -fsSL .../install.sh | bash -s -- --fullcd atomic-claude
python main.py run 0 # Start with Phase 0 (Setup)
python main.py status # Check progress
python main.py run 1 # Continue to Phase 1You point it at a project idea. It walks you through requirements gathering, generates a PRD, breaks it into tasks, writes specifications, implements code, reviews it, integrates components, and prepares deployment artifacts. Each phase produces outputs that feed the next.
LLM providers fall back automatically: Claude Code CLI → Anthropic API → AWS Bedrock → Ollama (local).
221 agents across 19 categories handle specialized work — backend, frontend, security, DevOps, data, and more.
State is atomic. Every task checkpoints. Backtrack to any point with python main.py backtrack <phase> [task].
Copy .env.example to .env and set your provider credentials. See the User Guide for details.
- User Guide — Using the pipeline
- Developer Guide — Extending and contributing
- API Reference — Core interfaces
- CLAUDE.md — Project conventions and structure
MIT