This project addresses the challenge of executing complex, multi-step AI-generated code plans autonomously, overcoming the limitations of interactive AI coding sessions such as context degradation and constant developer oversight. It targets developers and researchers leveraging AI coding assistants like Claude Code, offering a significant benefit by enabling hands-off execution of development tasks, ensuring AI model quality throughout, and saving considerable development time.

How It Works

ralphex orchestrates Claude Code to execute plans defined in Markdown files, breaking them into discrete tasks. Each task is executed in a fresh Claude Code session, preserving model sharpness and preventing errors caused by accumulated context. The process involves four phases: sequential task execution with validation commands and automatic commits; a comprehensive first code review by five parallel Claude Code agents (quality, implementation, testing, simplification, documentation); an optional independent review by Codex; and a final, focused review by two agents. This multi-phase approach ensures robust code quality and correctness before plan completion.

Quick Start & Requirements

Installation can be done via Go (go install), Homebrew (brew install), direct binary download, or Docker. A Git repository is mandatory. Users require the Claude Code CLI with configured credentials; fzf is recommended for interactive plan selection, and Codex CLI is optional for its review phase. Docker usage necessitates Docker installation. Plans are typically stored in docs/plans/ and follow a specific Markdown format.

Highlighted Details

• Zero setup required with sensible defaults.
• Autonomous task execution with automatic retries and commits.
• Multi-phase code review pipeline involving multiple AI agents and optional Codex integration.
• Highly customizable review agents and prompts for tailored workflows.
• Real-time monitoring via terminal streaming and an optional web dashboard (--serve).
• Docker support provides an isolated and safer execution environment.
• Interactive plan creation via CLI (--plan) or Claude Code slash commands.

Maintenance & Community

The provided README does not contain specific details regarding maintainers, community channels (like Discord or Slack), or project roadmap beyond the features described.

Licensing & Compatibility

The project is released under the MIT License, which is permissive and generally allows for commercial use and integration into closed-source projects without significant restrictions.

Limitations & Caveats

ralphex strictly requires a Git repository for its operation. Successful execution depends on having Claude Code CLI and its credentials configured; Codex is optional but enhances the review process. The use of --dangerously-skip-permissions by Claude Code necessitates careful consideration, making Docker isolation a strongly recommended safety measure. The effectiveness of the autonomous execution and review phases is contingent on the capabilities and configuration of the underlying AI models and user-defined prompts.