Skip to content# ay-framework > Agent swarm framework for Claude Code. Plan, build, review, test, ship — with human-in-the-loop gates. **Status:** Released | **Language:** Shell | **License:** MIT **Repo:** https://github.com/liwala/ay-framework ## The Problem Ad-hoc prompting drifts. Agents skip planning, mix concerns, and lose context across sessions. Without a structured cycle and clear gates, you end up reviewing half-baked code instead of approving a plan you trust. ## What It Does - Autonomous /go cycle: observe, lock, plan, build, self-review, test, learn, unlock - Two human gates only: approve the plan, review the code - Six cognitive modes — PLAN, BUILD, REVIEW, QA, SHIP, RETRO — never mixed - Scoped agents that own specific directories to prevent parallel-write conflicts - Auto-generated agents and tasks from your project structure (First Contact) - Tracking layer: BOARD, HANDOFFS, lock files, atomic commits per task - Six runtimes: Claude Code, Cursor, Windsurf, Codex, Trae, Kiro - Injects into existing CLAUDE.md without overwriting your config ## Quick Start ```bash npx ay-framework # or skip prompts: npx ay-framework --global --yes # Claude Code, global npx ay-framework --local --runtime=cursor # Cursor, this project # then in your agent: "I want to build [describe your project]" # triggers /go cycle ```