ralph-loop

Orchestrate autonomous Claude subprocess iterations with threshold-based gating and version control.

1|Updated Mar 7, 2026

Search Tags:#automation #workflow #claude #orchestration #iteration #prompt-engineering #version-control

Ready to run

npx skills add https://github.com/AG9898/PigeonCoop --skill ralph-loop-ag9898

💬 Prompt for ChatGPT / Claude Chat▼

Please help me install this Agent Skill.
Skill: ralph-loop
Source: https://github.com/AG9898/PigeonCoop/tree/main/.claude/skills/ralph-loop
Command: npx skills add https://github.com/AG9898/PigeonCoop --skill ralph-loop-ag9898

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

Automates repeated Claude CLI executions to perform a skill or prompt, eliminating manual looping and enabling consistent iteration, gating, and version-control actions.

Core Features & Use Cases

Orchestrates fresh claude subprocesses per iteration to run a skill or prompt.
Tracks progress using thresholds: iterations or completed tasks, and obtains baseline for tasks-based thresholds.
Extracts RALPH-SUMMARY from sub-agent output to gate decisions without leaking context.
Maintains safe version control: commits and pushes only on successful cycles.

Quick Start

Invoke the loop with a target task and threshold, e.g., /ralph-loop start-task iterations:5.

Frequently Asked Questions about ralph-loop

High-intent search queries and answers about installing and using this skill.

FAQPage Schema

How do I automate Claude CLI iterations for prompt engineering workflows?▼

Automate Claude CLI iterations by orchestrating fresh subprocesses per cycle to execute a specified skill or prompt. This eliminates manual looping and enforces consistent iteration, gating, and version-control actions for prompt engineering workflows.

What is the best way to run autonomous AI evaluation cycles with safe version control?▼

Run autonomous AI evaluation cycles by orchestrating Claude subprocesses governed by task or iteration thresholds. The workflow enforces phase-based control and only commits and pushes code to version control upon successful cycle completion.

How does sub-agent context extraction work during automated workflow orchestration?▼

Sub-agent context extraction works by parsing a specific summary block from the subprocess output. This gates iteration decisions based on the extracted summary without leaking the full sub-agent context into the next automated workflow orchestration cycle.

Can I use task completion thresholds instead of fixed iteration counts to control autonomous agents?▼

Yes, you can control autonomous agents using task completion thresholds instead of fixed iteration counts. The orchestrator obtains a baseline and tracks progress based on the number of successfully completed tasks until the specified threshold is reached.

Do I need to manually commit changes when running automated Claude subprocess experiments?▼

No, you do not need to manually commit changes when running automated Claude subprocess experiments. The orchestration loop maintains safe version control by automatically committing and pushing changes only after a successful cycle is verified.

When should I not use autonomous iteration orchestration for AI workflows?▼

You should not use autonomous iteration orchestration when your AI workflows require continuous shared context across cycles, as the orchestrator spawns fresh subprocesses per iteration and gates decisions using only extracted summaries to prevent context leakage.