Autonomous Agentic Loop for AI Coding Agents This project provides a CLI tool that wraps AI coding agents (Claude Code, Codex, OpenCode) in a self-correcting, autonomous loop. It's designed for developers seeking to automate complex coding tasks, enabling AI agents to iteratively refine code based on previous outputs and test results, ultimately completing tasks with minimal human oversight.

How It Works

Open Ralph Wiggum implements the "Ralph Wiggum technique," creating an agentic loop where an AI coding agent repeatedly receives the same prompt. Crucially, each iteration, the agent observes the evolving codebase and Git history from its previous actions. This persistent feedback loop facilitates self-correction and incremental progress, allowing the agent to refine its work until a defined completion promise is met. The tool supports switching between different AI coding agents via a command-line flag.

Quick Start & Requirements

• Primary install / run command: npm install -g @th0rgal/ralph-wiggum or bun add -g @th0rgal/ralph-wiggum.
• Non-default prerequisites and dependencies: Bun runtime, and at least one AI coding agent CLI installed (Claude Code, OpenAI Codex, or OpenCode).
• Links: No specific external quick-start or demo links are provided beyond the CLI usage examples.

Highlighted Details

• Multi-Agent Support: Seamlessly switch between Anthropic's Claude Code (--agent claude-code), OpenAI's Codex (--agent codex), and the default OpenCode (--agent opencode).
• Self-Correcting Loops: Agents iteratively fix their own mistakes by observing previous outputs, file changes, and test results.
• Autonomous Execution: Designed for unattended operation, allowing developers to "set it running and come back to completed code."
• Task Tracking: Supports a --tasks mode for breaking down complex projects into smaller, manageable, and tracked steps stored in .ralph/ralph-tasks.md.
• Live Monitoring & Control: A --status command provides real-time insights into active loops, iteration history, and potential struggle indicators. Mid-loop guidance can be injected using --add-context.

Maintenance & Community

The provided README does not detail specific contributors, sponsorships, or community channels (e.g., Discord, Slack).

Licensing & Compatibility

• License type: MIT License.
• Compatibility notes: The MIT license generally permits commercial use and integration with closed-source projects without significant restrictions.

Limitations & Caveats

This project is strictly a CLI-only tool and requires specific AI agent CLIs to be installed and configured. It is not suitable for tasks demanding human judgment, one-shot operations, or those with unclear success criteria. Users should set max-iterations to prevent runaway loops, and troubleshooting may involve managing potential conflicts with AI agent plugins.