ralph

To keep AI building until verification passes, use a PRD-driven persistence loop that iterates build and verifier checks until stories pass or exceed the blocked-attempt threshold. Progress is tracked and retried deterministically across the session.

Verifier-gated progress advances stories only when verifier-aligned evidence succeeds, reducing the chance of premature completion claims. The loop increments attempt counters and hard-stops after repeated blocking to prevent infinite iterations.

Tracking multi-story implementation tasks requires initializing state into per-session PRD JSON. This maintains persistent story status, allowing iterative builds to progress sequentially only when current stories are proven by verifier evidence.

To stop an infinite build and test loop, blocked-iteration guardrails increment attempt counters and hard-stop after reaching a blocked-attempt limit. You can also trigger user-controlled deactivation via stop commands to halt the workflow.

PRD-driven persistence is needed for long-running implementation tasks where progress must be tracked and advanced only on evidence. It ensures the agent continues iterating builds until the verifier passes or the threshold is exceeded.

Yes, Ralph operates without external dependencies, using internal scripts to manage the PRD-to-build loop. It initializes per-session state and handles attempt counters autonomously within your existing environment.