ralph

Pipe a persistent prompt to an agent loop using feature artifacts.

1|Updated May 6, 2023

Search Tags:#automation #workflow #ai-agent #loop #ralph #prompt-management #feature-plan

Ready to run

npx skills add https://github.com/TobiasLaross/dotfiles --skill ralph-tobiaslaross

💬 Prompt for ChatGPT / Claude Chat▼

Please help me install this Agent Skill.
Skill: ralph
Source: https://github.com/TobiasLaross/dotfiles/tree/main/agentic/skills/ralph
Command: npx skills add https://github.com/TobiasLaross/dotfiles --skill ralph-tobiaslaross

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

This workflow sets up and launches a true Ralph Wiggum loop for a planned feature. It pipes the same prompt to the agent on every iteration and lets it see its previous work via the filesystem. It requires /feature-plan to have been run first and avoids pre-decomposed tasks.

Core Features & Use Cases

True Ralph loop orchestration: pipes a single prompt to the agent each iteration and surfaces its prior work from the filesystem.
Prerequisite-driven: works only after /feature-plan and uses feature artifacts (story.md, progress.md, design.md) to guide execution.
Tooling: uses Read, Write, Edit, Grep, Glob, Bash, Agent, and AskUserQuestion to read, modify, and interact with the feature state.
Use case: end-to-end feature exploration where the agent iteratively decides next steps based on its past contributions.

Quick Start

Initiate a Ralph loop for <feature-name> after you have run /feature-plan.

Frequently Asked Questions about ralph

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

FAQPage Schema

What is an AI agent loop for iterative feature development?▼

An AI agent loop for iterative feature development pipes a single persistent prompt to the agent on every iteration, letting it read its previous work from the filesystem to decide the next steps autonomously.

How do I start an automated workflow loop for a planned feature?▼

To start an automated workflow loop, you first run the feature-plan process to generate story.md, progress.md, and design.md artifacts, then launch the loop to read repository state and guide execution iteratively.

Can I use an AI agent loop without pre-decomposed tasks?▼

Yes, this AI agent loop specifically avoids pre-decomposed tasks by relying on the agent to iteratively decide next steps based on its past contributions surfaced from the filesystem.

Does the agent loop support pausing and resuming feature execution?▼

Yes, the agent loop ensures safe, resumable operation by using Read, Write, Edit, and Bash tools to track repository state and progress.md artifacts between iterations.

What are the limitations of using a single prompt loop for feature execution?▼

A key limitation is that the single prompt loop requires the feature-plan prerequisite to have been run first, and it avoids pre-decomposed tasks, making it unsuitable for rigid, step-by-step task execution.