ralph

Automated PRD-driven execution loops iterate on user stories by persisting progress in state files, running parallel agent work, and enforcing reviewer sign-off until every story passes verified acceptance criteria. This prevents silent failures and scope reduction in complex development tasks.

Preventing premature completion requires persisting progress across iterations and enforcing a verification loop with explicit reviewer sign-off. State files track user stories, ensuring tasks do not end silently until all acceptance criteria pass and regression checks are re-verified.

Reviewer verification and regression re-verification are enforced by running parallel agent work, executing builds and tests, and requiring explicit reviewer sign-off. An optional cleanup pass can run before final regression checks to ensure no scope reduction occurs.

PRD state tracking supports multi-module refactors by auto-generating or reading a PRD file, delegating work to parallel agents, and tracking completion in a progress file. It ensures structured acceptance criteria are met before clean exit and state cleanup.

Structured acceptance criteria and persistent progress tracking are needed for multi-iteration development tasks requiring demonstrable completion, such as security fixes, critical integrations, or multi-module refactors where passing tests and reviewer sign-off are mandatory.

Multi-iteration task automation requires state cleanup and deslop regression passes to remove low-quality AI output and re-verify stability before marking completion. This guarantees the final state reflects genuinely passing tests and explicit reviewer approval.