Skill All skills & plugins / Claude Code official plugins
ralph-loop
Interactive self-referential AI loops for iterative development, implementing the Ralph Wiggum technique. Claude works on the same task repeatedly, seeing its previous work, until completion.
- Type
- Plugin by Anthropic
- Catalog
- Claude Code official plugins
- Source
- anthropics/claude-plugins-official/plugins/ralph-loop (9 files scanned)
- Install and details
- Agent Skills Index entry
- License
- Apache-2.0 (LICENSE file)
- Signals
- Pre-approved shellHooksBundled scripts
Pre-approved tools
Shell (scoped). Tools listed in allowed-tools (or an agent's tools) can be used without a permission prompt while the skill is active, depending on your agent and settings.
| Kind | Tools |
|---|---|
| shell, scoped | Bash(${CLAUDE_PLUGIN_ROOT}/scripts/setup-ralph-loop.sh:*)Bash(test -f .claude/ralph-loop.local.md:*)Bash(rm .claude/ralph-loop.local.md) |
| read/search | Read(.claude/ralph-loop.local.md) |
Declared in commands/ralph-loop.md, commands/cancel-ralph.md.
Hooks
Hooks run automatically on these agent events, without you invoking the skill: Stop.
| Event | Matcher | Runs |
|---|---|---|
| Stop | all | bash "${CLAUDE_PLUGIN_ROOT}/hooks/stop-hook.sh" |
MCP servers and apps
None.
Bundled code
2 code files (Shell):
Network calls
No HTTP or socket call pattern found in the code.
Starts other programs
No subprocess, exec or eval pattern found.
Credentials referenced
No key, token or secret environment variable found in code or MCP config.
Static scan of commit 41f619d on 2026-10-27. Signals are patterns found in the files, not a verdict: a network call can be the whole point of a skill. Read the linked lines before you decide. How the scan works.