Ralph — автономный цикл разработки для Claude Code

★ 9.6K

Ralph for Claude Code is an implementation of an autonomous development-loop technique for Claude Code: the agent iteratively improves your project until completion, with intelligent exit detection and rate limiting. Reach for it when you want to run a continuous autonomous loop: Claude Code cycles through "do → check → continue" iterations on a task until it finishes, while built-in safeguards prevent infinite loops and API overuse. It installs once as a global command available in any directory. It is based on Geoffrey Huntley's Ralph technique. Its focus is an autonomous-loop wrapper over Claude Code with completion detection and rate limiting, not the agent itself or a multi-agent orchestration environment: the value is reliably running Claude Code in a long autonomous mode on a task. In job it is close to snarktank/ralph; choose by implementation details. It suits people who want autonomous long runs of Claude Code with loop protection.