luckeyfaraday/codex-router — explained in plain English
Analysis updated 2026-05-18
Offload a large mechanical refactor across many files to Codex while Claude keeps working on other tasks.
Delegate bulk lint or test failure fixes to a background Codex worker and have Claude review the diff.
Generate boilerplate code from a clear spec using Codex, supervised and verified by Claude.
| luckeyfaraday/codex-router | abhayparolkar/zboy | dmaynard51/ultima1-ios | |
|---|---|---|---|
| Stars | 6 | 6 | 6 |
| Language | Shell | Shell | Shell |
| Setup difficulty | moderate | hard | moderate |
| Complexity | 3/5 | 4/5 | 3/5 |
| Audience | developer | developer | general |
Figures from each repo's GitHub metadata at analysis time.
Requires both a Claude Code setup and a separate OpenAI Codex CLI login with its own plan or API key.
codex-router is a plugin for Claude Code that lets a Claude coding agent hand off large, well defined coding tasks to OpenAI's Codex tool, then check the work when it comes back. The idea is that Claude plans and reviews, while Codex does the repetitive grinding: sweeping refactors, bulk migrations, fixing many similar test failures, or writing boilerplate from a clear spec. When you tell Claude to offload a task, either by typing a command or just describing the intent in plain language, Claude first writes a self contained brief describing the goal, the files involved, numbered requirements, and what counts as done. It then picks one of three GPT-5.6 models depending on how hard the task looks: one for difficult work, one as a balanced everyday default, and one for narrow, repetitive, high throughput jobs. Codex runs in the background using this brief, without blocking your terminal, so Claude can keep working on other things while it waits. Once Codex finishes, Claude reads the actual code changes it produced, not just a summary, runs the checks that define a finished task, and either fixes small gaps itself or sends Codex back for another round with corrections, picking up the same session instead of starting over. Every run keeps a record of the exact prompt sent, the full event log, and Codex's final message, so you can trace what happened. By default Codex runs sandboxed to your project folder with no network access and cannot commit or push changes on its own, so you and Claude review the result before anything is finalized. To use it you need both Claude Code and the OpenAI Codex command line tool installed and logged in, plus an active plan or API key for each. The project installs as a Claude Code plugin or by copying a skill folder into your configuration, and it is released under the MIT license.
A Claude Code plugin that lets Claude delegate big mechanical coding tasks to OpenAI Codex in the background, then reviews and verifies the result.
Mainly Shell. The stack also includes Shell, Claude Code, OpenAI Codex CLI.
Free to use, modify, and distribute for any purpose, including commercial use, as long as the copyright notice is kept.
Setup difficulty is rated moderate, with roughly 30min to a first successful run.
Mainly developer.
This repo across BitVibe Labs
double-check against the repo, no cap.