agentleFS
pfeffpfeff/claude-skills/skills/ralph-wiggum/SKILL.md

ralph-wiggum

Skill3 starsChanged 3 months agoMITView on GitHub

Scaffold a project for the Ralph Wiggum autonomous control loop. Use when setting up a new project for autonomous iteration, creating loop configuration files, or preparing specs for AI-driven development. Supports single-repo and multi-repo workspaces.

7 other projects use a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

2 rules. 0 are common, 2 are distinctive, and 0 are very specific to this project.

  • 0

    Common

    Most of the 7 edited versions we read keep these.

  • 2

    Distinctive

    This author's own choices. Worth a look.

  • 0

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    When invoked, run operations sequentially:

    line 50
  2. Distinctive:

    Success: All configuration files created.

    line 66

“Common” is measured: kept by at least 6 in 10 of the 7 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Ralph Wiggum Project Scaffolding

Scaffolds the files needed to run a Ralph Wiggum autonomous control loop on a project. Supports both single-repo projects and multi-repo workspaces.

Overview

The Ralph Wiggum pattern is a while true bash loop that repeatedly feeds the same prompt to Claude Code. Between iterations, files and git history persist while context resets fresh. The agent sees its own previous work and iteratively improves until backpressure gates (tests, lint, build) all pass.

Core philosophy: Backpressure over direction—engineer environments where wrong outputs get rejected automatically.

Multi-Repo Support

Ralph can operate on workspaces containing multiple git repositories. In multi-repo mode:

  • A workspace manifest (.ralph/workspace.md) lists all repos with paths and gate configs
  • Each repo has its own .ralph/gates.md with repo-specific gate commands
  • Tasks in PLAN.md are annotated with [repo-name] prefix to target specific repos
  • Branches, commits, and PRs are managed per-repo
  • The loop runs from the workspace root directory

Operations

1. Gather Requirements

When: Starting scaffolding, need project info Implementation: Load operations/gather-requirements.md Quick summary: Detects single-repo or multi-repo workspace, collects language and gate commands (per-repo in multi-repo mode), and optional external service dependencies via AskUserQuestion.

2. Scaffold Project

When: Requirements gathered, ready to create files Implementation: Load operations/scaffold-project.md Quick summary: Checks for existing files, creates from templates with user-provided values. In multi-repo mode, creates workspace manifest and per-repo gate files. Shows next steps.

Prerequisites

  • Current directory is a git repository, or contains subdirectories that are git repos (for multi-repo mode)
  • Write and AskUserQuestion tools available

End-to-End Flow

When invoked, run operations sequentially:

Input

  • Working directory (scanned for git repos to detect workspace mode)
  • User responses to AskUserQuestion prompts

Sequence

1. gather-requirements → workspace_type, gate commands, optional services
2. scaffold-project(gathered_values) → created files list

Output

  • Success: All configuration files created. User shown file list and next steps (write specs, run plan, run build).
  • Partial: Some files skipped (user chose to preserve existing). User shown which files were created vs skipped.

Quick Reference

Invoke skill: User asks to "scaffold ralph wiggum", "set up autonomous loop", or "prepare project for ralph"

Template location: skills/ralph-wiggum/templates/

Three-phase workflow:

  1. Spec - Human writes requirements in specs/*.md
  2. Plan - Agent generates PLAN.md from specs
  3. Build - Agent implements one task per iteration

Backpressure gates: Tests, lint, typecheck, and build must all pass before a task is complete.

Workspace modes:

  • Single-repo: .ralph/gates.md at project root (original behavior)
  • Multi-repo: .ralph/workspace.md at workspace root + per-repo .ralph/gates.md

Progressive Disclosure

Load only what you need:

  • operations/gather-requirements.md - Interactive requirements collection (detects workspace type)
  • operations/scaffold-project.md - File creation from templates (supports both modes)
  • references/environment.md - Pre-flight checks and prerequisites
  • templates/ - File generation templates (gates, services, specs, prompts, workspace)
  • scripts/ - Host-side runtime: run-container.sh, loop.sh, lib.sh, setup-network.sh, devcontainer config

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.