ralph

What problem does it solve? Complex coding tasks often get declared done prematurely: partial implementations pass as complete, tests get skipped, and edge cases are forgotten. Ralph prevents this by looping on a task until it is genuinely complete, requiring fresh verification evidence and explicit architect sign-off before allowing completion. ## Core Features & Use Cases - Persistence Loop: Keeps iterating on a task across retries, tracking state via omx state commands and resuming from where the previous iteration stopped. - Parallel Delegation: Fires independent subagent tasks simultaneously with explicit agent_type and reasoning_effort levels (low, medium, xhigh) matched to task complexity. - Mandatory Verification: Requires fresh test/build/lint output plus an architect subagent review, a deslop cleanup pass on changed files, and post-cleanup regression re-verification before completion. - Use Case: A user says "ralph: migrate the auth module to OAuth2 and don't stop until all tests pass." Ralph plans the work, delegates implementation and test lanes in parallel, runs the suite, gets architect approval, cleans up AI slop, re-runs regression, and only then reports completion. ## Quick Start Ask the agent to ralph the task of refactoring the payment module and keep iterating until all tests pass and an architect review approves the changes.