PluginInstalls cleanly
Ask Codex to review Claude's work
Codex plugin for Claude Code
OpenAI's plugin lets you ask Codex for a second opinion on your code from inside Claude Code, or hand it a bug to look into while you keep working.
- Works with
- Claude Code
- Good for
- Build an app, Web developers
What it does
This plugin comes from OpenAI. It lets Claude Code users call Codex, OpenAI's coding assistant, without leaving Claude Code. It uses the Codex program (the Codex CLI) already installed on your computer and your own Codex sign-in. The current version is 1.0.6. The last code change was in July 2026.
Read this first. Three things happen without a request from you:
- Hooks run all the time. A hook is a small script that runs automatically. This plugin adds hooks for every session start, every time Claude stops, and every session end. They run even if you never type a Codex command. When a session ends, the hook stops that session's running Codex jobs.
- Codex can change your files without asking.
/codex:rescuestarts Codex with write access to your project and with approval prompts turned off. Codex edits files as it sees fit. - Claude can call Codex by itself. The plugin's rescue helper tells Claude to hand larger tasks to Codex "proactively", without waiting for you to ask. Each hand-off uses your Codex limits and can change files.
If you want Codex only when you ask for it, do not install this plugin. Use Codex on its own instead.
The main use is a second opinion. /codex:review asks Codex to read your current changes and point out problems. It cannot change files. /codex:adversarial-review does the same but questions the overall approach, and you can tell it what to focus on. While these two commands run, Claude Code lets any node and git command run without asking you.
You can also hand Codex a job with /codex:rescue, for example to look into a failing test. Long jobs can run in the background; /codex:status shows progress and /codex:result shows the answer. /codex:transfer copies your Claude Code conversation into a Codex thread, so you can continue in Codex. Codex sends that conversation to OpenAI when you continue the thread.
The optional review gate (/codex:setup --enable-review-gate) makes the stop hook run a Codex review each time Claude stops. If the review finds problems, Claude must keep working. The README warns that this can start a long Claude and Codex loop and use up your limits quickly. Turn it on only while you watch the session.
You need Node.js 18.18 or later. You also need the Codex CLI signed in with a ChatGPT plan or an OpenAI API key. OpenAI's pricing page lists the Codex CLI for Plus ($20 a month) and higher plans, and for API keys. The Free and Go plans list Codex in the ChatGPT desktop app only. The README says Free works, but the pricing page does not list the CLI for Free.
The plugin itself sends no usage data. Everything goes through your Codex CLI to OpenAI. It saves its job list in Claude Code's plugin data folder, or in a temporary folder.
Who it is for
People who build with Claude Code, already have Codex on a Plus or higher plan (or an OpenAI API key), and want a second model to check the work.
Example prompts
- "/codex:review"
- "/codex:review --background"
- "/codex:adversarial-review --base main challenge whether this was the right caching and retry design"
- "/codex:rescue investigate why the tests started failing"
- "Ask Codex to redesign the database connection to be more resilient."
How to install
/plugin marketplace add openai/codex-plugin-cc/plugin install codex@openai-codex/reload-plugins/codex:setupHow to use it
Type /codex:review to get a read-only Codex review of your current changes, or /codex:rescue followed by a plain request, such as investigate why the tests started failing.
Check it works
- Run
/codex:setup. It tells you whether Codex is installed and signed in. - Run
/agentsand look for thecodex:codex-rescuesubagent. The README says you should also see the/codex:slash commands. - Try the README's first test:
/codex:review --background, then/codex:status, then/codex:result.
Common problems
- Codex is installed but not signed in —
/codex:setupreports it — run!codex logininside Claude Code, as the README shows. - Codex is missing —
/codex:setupreports it — let/codex:setupinstall it (it asks first), or runnpm install -g @openai/codexyourself (README). - "The '…' model requires a newer version of Codex. Please upgrade to the latest app or CLI and try again." — your Codex CLI is too old for the model — run
npm install -g @openai/codex@latest(issue 468). - A model you named does not work — the README's examples
gpt-5.4-miniandspark(gpt-5.3-codex-spark) are retired for ChatGPT sign-in — leave out--model, or remove themodelline from.codex/config.toml(Codex models page). - Your computer gets slow after many sessions — leftover Codex background processes keep running — close them in your task manager, and update the plugin when a fix comes out (issues 543 and 718).
- A Codex job in another window stopped — ending one Claude Code session can stop the shared Codex helper for other open sessions — keep one session open per project while Codex jobs run (issue 540).
- On Windows, long shell commands start to fail after a long session — the session-start hook adds lines to Claude Code's environment file each time and the file grows too long — start a new session (issue 528; our reading).
What to watch out for
- The plugin adds hooks that run at every session start, every stop and every session end, even if you never use a Codex command; they need Node.js 18.18 or later (hooks/hooks.json, README)
- You need the Codex CLI signed in with ChatGPT Plus or higher, or an OpenAI API key; OpenAI's pricing page lists the CLI only from Plus up, although the README says Free works. Every run counts against your Codex limits or API credits
- /codex:rescue runs Codex with write access and no approval prompts, so it can change files in your project; Claude may also hand a task to Codex without being asked (agents/codex-rescue.md, scripts/lib/codex.mjs)
- While a review runs it pre-approves any node and git command; /codex:setup pre-approves npm commands and can install Codex globally after it asks you (commands/*.md)
- The optional review gate makes the stop hook run a Codex review each time Claude stops; the README warns it can start a long loop and use up limits quickly
- /codex:transfer copies your Claude Code conversation into a Codex thread, which Codex sends to OpenAI when you continue that thread
- Open issues report background Codex processes left running after sessions end, and ending one session can stop another session's Codex job (issues 543, 540)
- The README's model examples no longer work with ChatGPT sign-in: gpt-5.4-mini retired on 2026-08-31 and gpt-5.3-codex-spark (spark) on 2026-09-14 (Codex models page); leave the model unset
Licence and author
- Licence
- Apache-2.0
- Author
- OpenAI
- Source code
- github.com/openai/codex-plugin-cc
How we tested it
We installed it on an empty test computer on 19 Oct 2026. Installed and started in a clean test system. Not yet used in a real AI session.
Not run: "/reload-plugins" (not a form the sandbox runs). Not run: "/codex:setup" (not a form the sandbox runs). No Codex sign-in in the sandbox, so no command or hook ran. Not yet used on real work.
Technical details of the test
- Date
- 19 Oct 2026
- Method
- Intake sandbox run on 2026-10-19 in a clean container with no account: cloned the repository and checked its skill, agent and plugin files; installed codex@openai-codex with the Claude Code command line.
Checked again on 19 Oct 2026
Alternatives
- PR Review Toolkit (Anthropic) — reviews done by Claude's own agents, no second account needed.
- CodeRabbit Skills and Plugins — a separate review service with its own account and limits.
- Running Codex on its own — open Codex in the same folder and type
/reviewthere.
FAQ
Do I need a separate Codex account? You need to be signed in to Codex on your computer, with a ChatGPT account or an API key (README).
Which ChatGPT plan do I need? OpenAI's pricing page lists the Codex CLI for Plus ($20 a month) and higher plans, and for API keys. The README says Free works, but the pricing page does not list the CLI for Free or Go.
Does it change my code?
The two review commands do not. /codex:rescue can, without asking you, and Claude can start it by itself.
Does it use my Codex settings?
Mostly. It reads the same config.toml as Codex, so your model and reasoning settings apply. But the plugin always sets the sandbox itself (read-only for reviews, write access for rescue), so sandbox_mode in your settings does not apply (issue 482).
Is it in Anthropic's plugin directory?
No. It installs from OpenAI's own marketplace, openai-codex.
Sources
- github.com/openai/codex-plugin-cc
- github.com/openai/codex-plugin-cc/blob/main/plugins/codex/hooks/hooks.json
- github.com/openai/codex-plugin-cc/blob/main/plugins/codex/agents/codex-rescue.md
- github.com/openai/codex-plugin-cc/blob/main/plugins/codex/scripts/lib/codex.mjs
- github.com/openai/codex-plugin-cc/issues/543
- github.com/openai/codex-plugin-cc/issues/540
- github.com/openai/codex-plugin-cc/issues/482
- developers.openai.com/codex/pricing
- developers.openai.com/codex/models