ralph_skill

Overview

This skill sets up and runs the Ralph Wiggum loop: an autonomous AI coding pattern that iterates on PRD-driven features with clean-slate iterations and CI quality gates. It scaffolds the loop, helps you author a testable PRD, and runs the iteration script while enforcing CI green and a clear stop condition.

How this skill works

The skill creates a local loop script and a PRD file, adapts test commands to your project (detecting package manager and TypeScript), and guides interactive PRD creation with specific user stories and acceptance criteria. When run, it performs pre-flight checks, summarizes progress and incomplete stories, asks for confirmation, and executes the loop script that runs isolated iterations, updates progress logs, and stops when all stories pass or a max iteration limit is reached.

When to use it

• Working on multi-step features that benefit from repeated autonomous iterations.
• You need continuous enforcement of tests and type checks across automated commits.
• Converting high-level requirements into small, testable user stories.
• Running long-running autonomous coding tasks that must avoid state leakage.
• Enabling an AI agent to prioritize one feature at a time and track progress.

Best practices

• Write PRD stories that are specific, scoped, and completable in one iteration.
• Provide 3–5 clear, testable acceptance criteria per story to make CI decisive.
• Keep one active story per loop run to prevent scope creep and branching complexity.
• Ensure your project has working tests and type checks before starting the loop.
• Set a reasonable max-iteration safety limit to prevent infinite loops.
• Append concise, timestamped progress notes each iteration for auditability.

Example use cases

• Add a new API endpoint with 3 precise acceptance tests and let the loop implement until tests pass.
• Implement a single front-end component with defined behavior and CI checks in each commit.
• Refactor a module incrementally while ensuring tests and types remain green after every change.
• Prototype a feature by iterating on small user stories and stopping when acceptance criteria are satisfied.
• Automate long-running feature work where the AI must repeatedly reset context to avoid carry-over errors.