OpenAI Codex Plugin for Claude Code: Review & Delegate
openai/codex-plugin-ccThe OpenAI Codex plugin for Claude Code is an official OpenAI plugin that lets you call Codex from inside Claude Code. It adds a set of `/codex:*` slash commands so you can get a Codex code review of your diff, or hand a task off to Codex, without switching to a separate terminal.
Reach for this if you work inside Claude Code and want a second opinion from Codex on your changes, or to pass a stuck task to a different model without leaving the session. Skip it if you don't have a ChatGPT or OpenAI API sign-in, or you'd rather just run Codex directly — this is a convenience wrapper on top of the Codex CLI, not a new engine.
The problem it solves
If your daily driver is Claude Code but you also want Codex's take on a change, you normally have to jump to another terminal, re-explain the context, and manage two tools by hand. That context-switch is enough friction that most people just skip the second review. This plugin exists to keep the Codex review and hand-off inside the Claude Code workflow you already have.
What is it?
The OpenAI Codex plugin for Claude Code is a Claude Code plugin (JavaScript, Apache-2.0) published by OpenAI that wires Codex into Claude Code through slash commands. `/codex:review` runs a read-only Codex review of your current changes; `/codex:adversarial-review` runs a steerable review that challenges your design; `/codex:rescue` delegates a task to Codex via a `codex:codex-rescue` subagent. It calls the Codex app server through the global `codex` binary already installed on your machine, so it uses your existing Codex sign-in and config rather than a separate runtime.
Why it's getting attention
It's an official OpenAI project sitting at roughly 27.5k stars, and it connects the two coding tools a lot of developers already juggle — Claude Code and Codex — instead of asking them to pick one. The pitch is narrow and practical: review your diff with a second model, or delegate work to Codex, from the prompt you're already in.
Key features
- ✓`/codex:review` — a read-only Codex review of your uncommitted changes, or your branch against a base ref with `--base <ref>`
- ✓`/codex:adversarial-review` — a steerable review that questions the design, tradeoffs, and hidden assumptions, and accepts custom focus text
- ✓`/codex:rescue` — delegates a task (investigate a bug, try a fix) to Codex via the `codex:codex-rescue` subagent, with `--model`, `--effort`, `--resume`, and `--fresh`
- ✓`/codex:transfer` — turns the current Claude Code session into a persistent Codex thread and prints a `codex resume <session-id>` command to continue it in Codex
- ✓Background jobs: `--background` plus `/codex:status`, `/codex:result`, and `/codex:cancel` to run long reviews without blocking
- ✓Optional review gate via a `Stop` hook that runs a Codex review before Claude finishes and blocks if it finds issues
- ✓Reuses your local Codex CLI, authentication, and `config.toml`, so no separate Codex account is needed if you're already signed in
Best use cases
- •Getting a second-model code review of your diff before you open a pull request
- •Pressure-testing a design decision with an adversarial review that challenges the approach, not just the syntax
- •Handing a stuck bug or refactor to Codex to investigate in the background while you keep working
- •Continuing a debugging conversation you started in Claude Code directly inside Codex via `/codex:transfer`
How to install / try
Install from inside Claude Code. Add the marketplace with `/plugin marketplace add openai/codex-plugin-cc`, install with `/plugin install codex@openai-codex`, then run `/reload-plugins`. Finish with `/codex:setup`, which checks whether Codex is ready and can offer to install it (via `npm install -g @openai/codex`) if npm is available. Requirements per the README: Node.js 18.18 or later, and either a ChatGPT subscription (Free included) or an OpenAI API key; usage counts against your Codex limits. If Codex isn't logged in, run `!codex login`.
How to use
Once installed, drive it with slash commands. `/codex:review` reviews your current work; `/codex:review --base main` reviews your branch against main; add `--background` for long multi-file reviews and check on them with `/codex:status` and `/codex:result`. Use `/codex:adversarial-review` when you want the review to challenge the direction. Delegate work with `/codex:rescue investigate why the tests started failing`. Model and reasoning-effort defaults can be set in your user- or project-level `.codex/config.toml`.
Strengths
- ✓Keeps the Codex review and delegation inside Claude Code, so you skip the terminal-switching and re-explaining
- ✓Reuses your existing local Codex install, sign-in, and config — no second account or runtime if you already use Codex
- ✓Background jobs mean a slow multi-file review doesn't freeze your session
- ✓Official OpenAI project under Apache-2.0, so you can read exactly what the commands do
Limitations & risks
- △It's a convenience wrapper, not a new model — everything routes through the Codex CLI, so you get nothing you couldn't get by running Codex directly
- △Hard dependency on Codex: you need the `codex` binary installed and a ChatGPT or API sign-in, and every run counts against your Codex usage limits
- △Requires Node.js 18.18 or later
- △The README warns the optional review gate can spin up a long Claude/Codex loop that drains usage limits, so it says to only enable it while actively watching the session
- △`/codex:transfer` only works on sessions stored under `~/.claude/projects` and needs a recent Codex version that supports session import
Alternatives
Who should try it — and who should skip
Developers whose main workflow is Claude Code and who already use, or are willing to sign into, Codex. If you want a second-model review of your diff or to delegate a task without leaving Claude Code, it removes real friction. If you don't have a Codex sign-in, or you already run Codex directly and don't mind the extra terminal, there's little reason to add it.
Frequently asked questions
It's an official OpenAI plugin (JavaScript, Apache-2.0) that adds `/codex:*` slash commands to Claude Code so you can run Codex code reviews and delegate tasks to Codex without leaving your Claude Code session.
No, if you're already signed into Codex on the machine — the plugin uses your local Codex CLI authentication. If you've only used Claude Code, you'll need to sign into Codex with a ChatGPT account or an API key first.
No. The README says it delegates through your local Codex CLI and Codex app server on the same machine, using the same install, authentication, and repository checkout you'd use running Codex directly.
`/codex:review` is a read-only review of your current changes with no custom focus. `/codex:adversarial-review` is steerable: it challenges the chosen design and tradeoffs and accepts extra focus text, for example to pressure-test auth, rollback, or race conditions.
The plugin itself is open source under Apache-2.0. Running it needs Codex access, which the README notes is available with a ChatGPT subscription (Free included) or an OpenAI API key, and usage contributes to your Codex limits.