ralph

Enforce PRD story completion with reviewer-verified acceptance criteria.

Updated May 20, 2026
One-click install
npx skills add https://github.com/xdkp/oh-my-claudecode --skill ralph-xdkp
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: ralph
Source: https://github.com/xdkp/oh-my-claudecode/tree/main/skills/ralph
Command: npx skills add https://github.com/xdkp/oh-my-claudecode --skill ralph-xdkp

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

Ralph prevents incomplete or falsely “finished” work by enforcing a PRD-driven, story-by-story loop that only declares completion after explicit reviewer verification.

Core Features & Use Cases

  • PRD-driven persistence loop: Breaks work into user stories backed by verifiable acceptance criteria in prd.json and keeps iterating until all stories pass.
  • Evidence-based verification: Re-checks each story’s acceptance criteria and requires reviewer sign-off (architect/critic/codex).
  • Automatic retry and progress tracking: Persists iteration progress in progress.txt and continues across retries until completion is truly met.
  • Mandatory post-approval cleanup: Runs an ai-slop-cleaner pass on only the session’s changed files, then re-runs regression checks.

Quick Start

Ask your Claude Code agent to run ralph on your task, optionally adding a verifier choice like --critic=architect.

Frequently Asked Questions about ralph

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I enforce PRD acceptance criteria verification before marking coding tasks complete?▼

PRD acceptance criteria verification requires a story-by-story loop that re-checks deterministic evidence and mandates tiered reviewer approval before a task is declared done. This prevents silent partial implementations by explicitly validating each story against its defined constraints.

What is the best way to prevent silent partial implementations in multi-iteration coding tasks?▼

Preventing silent partial implementations in multi-iteration coding requires persisting progress across retries and only finishing after structured story verification. A PRD-driven loop tracks iteration progress continuously until all stories pass explicit acceptance checks.

How do I set up PRD-driven task verification for high-assurance coding workflows?▼

PRD-driven task verification requires initializing and refining a prd.json file containing user stories with verifiable acceptance criteria. You select stories by priority, execute them, and enforce deterministic evidence checks before allowing reviewer sign-off.

Can I use automated cleanup passes on only the files changed during a coding session?▼

Automated cleanup passes can be bounded to only the session's changed files by running a targeted deslop pass after reviewer approval. This ensures the cleanup process does not alter previously approved code while removing newly introduced slop.

Why does task verification require regression re-verification after a cleanup pass?▼

Regression re-verification after a cleanup pass ensures that removing slop or unused code from changed files does not break previously passing acceptance criteria. This post-deslop check guarantees the final codebase still meets all PRD story requirements.

Does ralph work with tiered reviewer approval for task completion?▼

Tiered reviewer approval for task completion is supported by allowing you to specify a verifier choice like an architect or critic. The system requires explicit sign-off from the designated reviewer before advancing stories or declaring the task finished.