The Ralph Twins
Both ralph-snarktank (19.6k★) and ralph-claude-code (9.2k★) implement the same Geoffrey Huntley pattern: an XML completion contract where Claude embeds <promise>TASK_COMPLETE</promise> in output as a machine-detectable done signal. But their engineering philosophies could not be more different. ralph-snarktank is 200 lines of bash — minimal, widely-forked, dormant. ralph-claude-code is a multi-module production system with 566 passing tests, session-handoff, and active maintenance. Neither mentions the other in its README. The additional ralph variants (ralphy-openspec, pi-ralph-orch, smart-ralph) independently discovered and adapted the same stop-hook loop.
Run an AI coding tool in a fresh-context loop, one user story at a time, until all stories in prd.json pass.
Production-grade Claude Code autonomous loop with dual-condition exit gate, rate limiting, circuit breaker, session management, and live monitoring.
Combines OpenSpec spec governance with Ralph Loop iterative execution and a SQLite run ledger to provide crash-recoverable, budget-tracked AI coding sessions.
Add YAML-configurable hat-based multi-phase orchestration loops to the pi coding agent, with event-driven hat transitions and guard rails.
Turn a vague feature idea into structured specs and then execute them task-by-task with fresh subagent context, without losing state across sessions.