Cette page est affichée en anglais. Une traduction française est en cours.
SkillsApr 28, 2026·4 min de lecture

Ralph Wiggum — Anthropic Official Autonomous Loop Plugin

Official Anthropic plugin that turns Claude Code into an autonomous loop. Adds /ralph-loop and /cancel-ralph for long-running self-improving task execution.

Introduction

Ralph Wiggum is the official Anthropic plugin that turns Claude Code into an autonomous, self-iterating development loop — the same pattern Boris Cherny (Claude Code's creator) uses for his longest-running tasks. Best for refactors, multi-file migrations, and "until-it-passes" TDD where you would otherwise babysit Claude over many turns.

The technique is named after Ralph Wiggum from The Simpsons — embodying persistent iteration despite setbacks. Under the hood it is a loop that re-feeds Claude its prompt + completion criteria until it emits a finish signal. Works with Claude Code 1.x and above. Setup time: under 1 minute.


How Ralph Wiggum Works

A Ralph loop is structurally a while true that re-runs Claude Code with the same prompt file each iteration, accumulating context and self-correcting until either:

  1. The model emits an explicit completion sentinel (the --completion-promise string), OR
  2. --max-iterations is reached.

Boris Cherny on his personal setup page (howborisusesclaudecode.com): "For very long running tasks, I either prompt Claude to verify its work with a background agent, use an agent-stop hook, or use the Ralph Wiggin plugin for autonomous looping."

Slash commands installed

Command What it does
/ralph-loop "<prompt>" [--max-iterations N] [--completion-promise "TEXT"] Start an autonomous loop in the current session
/cancel-ralph Stop the active loop

When to reach for it

  • Multi-file refactors that need many edit-test cycles
  • TDD: "make all tests pass, then output COMPLETE"
  • Long migrations (e.g. "convert all class components to hooks")
  • Avoid for open-ended exploration (Ralph keeps going — burns tokens)
  • Avoid when verification is hard to express as a single sentinel string

Cost discipline (must-read)

Ralph will happily burn through your token budget. Boris's safety pattern:

  1. Always set --max-iterations (never default unlimited).
  2. Make --completion-promise something Claude can only emit when all tests pass.
  3. Start with small tasks; observe iteration count before scaling.

Example: full TDD loop

/ralph-loop "
Implement a TypeScript function 'parseDuration(input: string): number' that
parses strings like '2h30m', '1d', '45s' into milliseconds.

Requirements:
1. Run npm test after each iteration
2. Fix all failing tests
3. When all tests pass AND npm run lint is clean, output:
   <promise>RALPH_COMPLETE</promise>
" --completion-promise "RALPH_COMPLETE" --max-iterations 30

FAQ

Q: What is Ralph Wiggum for Claude Code? A: An official Anthropic-maintained Claude Code plugin that adds /ralph-loop and /cancel-ralph slash commands, letting Claude run an autonomous self-improving loop until it hits a completion sentinel or max-iterations cap.

Q: Is Ralph Wiggum free? A: Yes, the plugin itself is free and open source under Anthropic's claude-plugins-official marketplace. You only pay for the Claude API tokens consumed during the loop.

Q: How do I install Ralph Wiggum? A: Run /plugins install ralph-wiggum inside any Claude Code session. No additional config or marketplace registration needed.

Q: How do I stop a runaway loop? A: Use /cancel-ralph from inside the same session, or set --max-iterations upfront so the loop self-terminates.

Q: Can I use Ralph for non-coding tasks? A: Yes — community variants like ralph-wiggum-marketer use the same loop pattern for copy iteration. The official plugin is task-agnostic; success depends on writing a verifiable completion-promise.


🙏

Source et remerciements

Built and maintained by Anthropic as part of the official claude-plugins-official marketplace. Inspired by Geoffrey Huntley's "Ralph" technique.

anthropics/claude-code · plugins/ralph-wiggum

Boris Cherny credits this plugin specifically for his "very long running tasks" workflow on howborisusesclaudecode.com.

Fil de discussion

Connectez-vous pour rejoindre la discussion.
Aucun commentaire pour l'instant. Soyez le premier à partager votre avis.

Actifs similaires