ralph-loop
Self-improving iteration loop: the same task comes back at every end of turn, so the agent keeps refining its own work until an honest completion promise or an iteration cap.
Reference
| Field | Value |
|---|---|
| Name | ralph-loop |
| Version | 1.0.0 |
| Category | Productivity |
| Licence | Not declared |
| Source | catalog/navin/builtin_plugins/ralph-loop/plugin.json |
| README | catalog/navin/builtin_plugins/ralph-loop/README.md |
| Digest | sha256:f922296c286c4f3c12d8d7a30485d8672b45ac45043dcf277c57269bdceabf81 |
| plugin.json | agents, category, commands, description, name, skills, steps, version |
Commands
| Name | Arguments | Description |
|---|---|---|
/ralph-loop | "PROMPT" [--max-iterations N] [--completion-promise "TEXT"] | Start a Ralph loop - the same task comes back at every end of turn until a promise or an iteration cap |
/cancel-ralph | - | Cancel the active Ralph loop of this project |
/ralph-help | - | Explain the Ralph loop technique and the /ralph-loop and /cancel-ralph commands |
Agents
The plugin declares no agent.
Skills
The plugin declares no skill.
Hooks, MCP servers and settings
plugin.json declares no hook, no MCP server and no setting: the plugin is made of the components above only.
Workflow
| Step | Use |
|---|---|
| learn the technique and how to write the prompt | /ralph-help |
| start the loop with a cap and a completion promise | /ralph-loop |
| stop the loop early | /cancel-ralph |
Enable it in a project
Activating ralph-loop creates 4 activations in the project: the plugin itself and 3 commands. Removing the plugin activation removes them all.
curl -X POST <control-plane-url>/v1/projects/<project>/navin-library/activations \
-H "Authorization: Bearer <token>" -H "Content-Type: application/json" \
-d '{"kind": "plugin", "name": "ralph-loop"}'
<project>: id or slug of the project. <token>: API key with the editor role on the project.
client.navin_library.activate("<project>", "plugin", "ralph-loop")