umputun/ralphex

What it solves

ralphex automates the execution of complex implementation plans using AI coding agents (Claude Code or codex). It eliminates the need for manual "babysitting" of interactive agents and prevents the quality degradation that occurs when a single long session fills up with too much context by executing each task in a fresh, minimal-context session.

How it works

ralphex operates as a CLI tool that reads a plan file (Markdown with checkboxes) from a git repository and executes it through four main phases:

1. Task Execution: It identifies incomplete tasks, sends them to the AI agent for implementation, runs validation commands (like tests), and commits the changes.
2. First Code Review: It launches five specialized agents (quality, implementation, testing, simplification, and documentation) in parallel to find and fix issues.
3. External Review: An optional phase using an external tool (like OpenAI Codex) to provide an independent cross-model review.
4. Second Code Review: A final pass using quality and implementation agents to catch any remaining critical issues.

It also supports a "finalize" step to rebase or squash commits and can run in isolated git worktrees to allow multiple plans to be executed in parallel.

Who it’s for

Developers who use AI coding agents for complex features and want to automate the end-to-end process from task execution to multi-phase code review without constant manual intervention.

Highlights

• Autonomous Execution: Automatically runs tasks, validates them, and commits results.
• Multi-Agent Review Pipeline: Uses a tiered review system with specialized agents to ensure code quality.
• Fresh Context Management: Starts new sessions for each task to maintain model performance.
• Flexible Executors: Supports both Claude Code and codex as the primary execution engine.
• Worktree Isolation: Allows running multiple plans in parallel via isolated git worktrees.
• Monitoring Tools: Includes a real-time streaming output and a web-based dashboard.