ralph

What problem does it solve? Long multi-step coding tasks lose focus when handled in one pass: items get skipped, quality checks are forgotten, and failures cause thrashing. This Skill enforces a disciplined loop where each iteration completes exactly one plan item, runs quality gates, and records progress. ## Core Features & Use Cases - Scoped Iterations: Picks the first unchecked item from plan.md and works on only that item per iteration, keeping context small and changes reviewable. - Quality Gates Every Pass: Runs repo-native checks (make format, make check, make test) after each item and commits the iteration via jj. - Explicit Stop Conditions: Halts when the plan is complete, when an item fails twice consecutively, or when a user decision is required. - Use Case: Point it at a failing test suite with the unit-tests mode; it creates a task directory, lists every failure as a checklist item, and fixes one per iteration until the suite is green. ## Quick Start Run /loop /ralph with the path to your plan file or task directory to start unattended iterations, for example by asking the agent to run ralph against .agents/plans/my-task__inprogress.