ralph-loop: a Claude Code plugin by Anthropic
Official
For developers with a well-defined task and a checkable finish line who want Claude to keep iterating alone. Skip it for vague goals or taste calls.
Before you install: By default there is no iteration limit, so a loop runs until the promise phrase appears or you cancel it unless you set --max-iterations.
- Score
- 76of 100
- Rank
- #15of 25
- Updated
- 2026
Install
/plugin install ralph-loop@claude-plugins-official
Type it in Claude Code. View the source
What does ralph-loop do?
ralph-loop is a Claude Code plugin that keeps Claude working on one task across many passes, a method called the Ralph Wiggum technique. You start it with /ralph-loop, a prompt, an optional --max-iterations limit and a --completion-promise phrase. A Stop hook then blocks Claude from ending the session and feeds the same prompt back each time. Because files and git history persist, each pass sees the earlier work, such as failing tests, and builds on it. The loop ends when Claude outputs the exact promise phrase, reaches the iteration limit, or you run /cancel-ralph.
When should you use ralph-loop?
Use it when
- Getting a test suite to pass through repeated attempts
- A task has a clear, checkable definition of done
- You can leave Claude running with an iteration cap
Skip it when
- Success depends on taste or human judgment
- The goal is vague or open-ended
What it needs and what it can touch
What it can do
- Installs a Stop hook that blocks the session from ending
- Runs a setup shell script
- Reads and edits project files on each pass
From reading its instructions, not from running it.
Why it scores 76
Strongest on upkeep (100 of 100), weakest on adoption (61).
Details
- Stars of the shared plugin repository
- 46,718
- Last repository change
- License
- Not stated
- Data confidence
- Medium
Did it do the job for you?
ralph-loop