ralph

PRD task completion across multiple engineering iterations is ensured by running a PRD-driven persistence loop that iterates over user stories until all acceptance criteria pass reviewer verification. This prevents premature 'looks done' outcomes by tracking session-scoped state and enforcing structured story validation.

Preventing 'looks done' outcomes requires enforcing reviewer-verified acceptance criteria for each user story in a PRD. By requiring mandatory reviewer approval and an automated deslop cleanup pass, the task only finishes when all stories are truly satisfied and re-validated.

PRD-backed verification for multi-iteration feature implementation works by scaffolding a prd.json file, selecting the next failing story, and running evidence-based verification. It persists state across retries using session-scoped directories, ensuring work resumes correctly until reviewer sign-off.

PRD-driven task completion requires initializing and refining a prd.json file for story tracking. State persistence across retries is handled automatically by saving session data into a structured .omc/state/sessions/{sessionId}/ directory structure.

You can disable the automated cleanup pass during reviewer-verified regression checks by passing the --no-deslop flag. Without this flag, an unconditional deslop pass runs post-approval, followed by a mandatory regression re-verification of the work.

Use a persistence loop for engineering verification when building service features across multiple iterations, such as implementing an API endpoint with auth and tests. It ensures every acceptance criterion is met, verified by a reviewer, and re-validated until the PRD is fully satisfied.