CodeLeash

CodeLeash is an open source framework and toolkit for building full stack web applications with Claude Code while holding the agent to strict quality rules. Rather than instructing the agent through prompts and hoping it complies, CodeLeash enforces standards with code: hooks, scripts, custom checkers and feedback loops that the agent cannot ignore or bypass. Its author, Chris Adams, is explicit that it is not an agent orchestrator. It is for engineers who use a coding agent to write most of their code and want predictable results. Claude Code hooks force the agent through a red, green, refactor test driven development cycle, and a plan review hook assesses how well each plan accounts for testing. Unit tests carry a 10 millisecond timeout so the agent keeps tests fast. Custom linters check for unused code and routes and for conformance with the intended architecture layering, and a hook asks the agent to record surprises in a learnings file whenever it stops or compacts context. The reference stack is a monorepo with a FastAPI and Pydantic backend, a React, TypeScript, Vite and Tailwind frontend, Supabase for the database and authentication, and Prometheus, OpenTelemetry and Sentry for observability. CodeLeash is free and released under the MIT license on GitHub. There are no paid plans; running it requires a Claude Code subscription and the usual Supabase and Node.js tooling. The author does not recommend adopting CodeLeash as is. He presents it as a working demonstration of ideas that teams can adapt to their own repositories.