What problem does it solve?
Running a full AI-driven delivery cycle (planning, implementation, PR creation, code review, and corrections) across multiple context windows requires manual re-prompting and loses progress when sessions end. This Skill automates that entire loop in detached headless CLI sessions with persisted state, so long-running pipelines survive interruptions and resume exactly where they stopped.
Core Features & Use Cases
- Autonomous cyclic pipeline: Drives plan → implement (commit + PR) → review, then loops review → fix until the review verdict is clean or safety limits are hit.
- Resumable state tracking: Persists progress to a state file under ~/.prp/<key>/state/, so a halted or interrupted loop can be continued with --resume.
- Safety bounds: Refuses to open PRs from protected branches, caps review cycles and implementation iterations, and halts with state preserved on failures.
- Use Case: Ask the agent to run the full PRP loop for a new feature; it plans, implements, opens a PR, gets it reviewed, applies corrections, and stops only when the review is clean — all without further interaction.
Quick Start
Ask the agent to run the full PRP loop for your feature description, optionally with a base branch, a validation command, or --resume to continue a saved loop.