Skip to content
Test version · build of 28.10.26 · prototypes · direction note

Tested on 28.10.26

Plug Codex into Claude Code

Claude Code builds, Codex reviews and challenges. Both in the same terminal, in 20 minutes, and you check every handoff.

When you finishYou type /codex:review in Claude Code, Codex reviews your code in that same terminal, and you know when to trust it and when to push back.

Let’s go
What you need (4)
  • Claude Code installed and running on your machine: Get started with Claude Code.
  • A ChatGPT account, even a free one, or an OpenAI API key. Every review counts toward your Codex usage limits.
  • Node.js 18.18 or newer. Check with node -v.
  • A project folder tracked by Git: the review works on your changes, not on an empty folder.
  • Level 3 · So, can you actually?
  • 20 min
  • Free: No
  • 12 steps
My mascot about to plug a Claude plug into a Codex socket

Chapter 01 / 07 · 2 min

Understand the bridge

You stay in Claude Code. When you want a second look at the same code, you ask Codex, OpenAI's assistant, without switching windows or copying anything.

This isn't a tool of mine: it's an official OpenAI plugin. It plugs into the Codex installed on your machine and uses your own login. No second account to create.

Chapter 02 / 07 · 5 min

Install the bridge

Four commands, typed in Claude Code: add the catalog, install the plugin, reload, check.

  1. Step 2.1: Add OpenAI's catalog (command 1).
  2. Step 2.2: Install the Codex plugin (command 2).
  3. Step 2.3: Reload the plugins (command 3).
  4. Step 2.4: Check that everything is ready (command 4).
In the terminal
/plugin marketplace add openai/codex-plugin-cc
In the terminal
/plugin install codex@openai-codex
In the terminal
/reload-plugins
In the terminal
/codex:setup

/codex:setup tells you whether Codex is installed and signed in. If Codex is missing and npm is there, it offers to install it.

Plan BIf you get stuck

Codex missing or not signed in? Install it yourself with npm install -g @openai/codex, then type !codex login in Claude Code (the ! runs the command in the terminal). A command fails? Compare with the plugin's page: openai/codex-plugin-cc.

Chapter 03 / 07 · 4 min

Run a first review

Start small. The basic review is read-only: Codex reads your changes in progress and tells you what it thinks. It changes nothing.

  1. Step 3.1: Start the review in the background (command 1). On several files, it can take a while.
  2. Step 3.2: Check how far along it is (command 2).
  3. Step 3.3: Show the result when it's done (command 3).
In the terminal
/codex:review --background
In the terminal
/codex:status
In the terminal
/codex:result
$ /codex:review --background1$ /codex:status$ /codex:result
  1. The command that starts the review, in the background.

Want to compare your whole branch to main instead of just your changes in progress? Add --base main.

Weak

Opening two windows and copying code from one chat to the other.

Better

Letting Codex read the same folder as Claude, without copying anything.

Chapter 04 / 07 · 3 min

Ask it to push back

The normal review looks for mistakes. The "adversarial" review does something else: it questions your choices. It also takes extra text so you can aim it.

In the terminal
/codex:adversarial-review --background look for data loss risks and challenge my approach

It doesn't fix anything either. You decide what stays. It's useful right before shipping, when you can't see your own code anymore.

Chapter 05 / 07 · 4 min

Hand it a task, then check

Here Codex no longer just reads: it can hunt a bug and try a fix. That's the /codex:rescue command.

In the terminal
/codex:rescue --background find out why the tests fail and propose the smallest possible fix
  1. Step 5.1: Follow it with /codex:status, then read the result with /codex:result.
  2. Step 5.2: Look at what changed in your files before accepting anything.
  3. Step 5.3: Run your tests yourself. Not the ones Codex says it passed.

To pick up where it stopped: /codex:rescue --resume. To stop a running task: /codex:cancel.

Chapter 06 / 07 · 3 min

Hand over the whole conversation

Claude is close to your quota, or you'd rather finish in Codex? Transfer the session. Codex gets the whole story, not just your last sentence.

  1. Step 6.1: Type /codex:transfer in the Claude Code session you're in.
  2. Step 6.2: Copy the codex resume command it shows and run it in a terminal.
In the terminal
/codex:transfer
In the terminal
codex resume <session-id>
Plan BIf you get stuck

No thread created? Update Codex: older versions can't import a Claude session.

Chapter 07 / 07 · 2 min

The automatic safeguard: only turn it on knowingly

The plugin has an option: every time Claude finishes a reply, Codex reviews it, and if it finds a problem, Claude has to deal with it before stopping.

In the terminal
/codex:setup --enable-review-gate
In the terminal
/codex:setup --disable-review-gate

The cheat sheet

The whole guide in 9 lines

Copy it into your notes.

Cheat sheet · Plug Codex into Claude Code
1. ChatGPT account or API key, Node.js 18.18 minimum, folder tracked by Git.
2. /plugin marketplace add openai/codex-plugin-cc, then /plugin install codex@openai-codex, then /reload-plugins.
3. /codex:setup checks that Codex is installed and signed in (otherwise !codex login).
4. /codex:review --background, then /codex:status, then /codex:result.
5. /codex:adversarial-review challenges your choices before shipping.
6. /codex:rescue hands over a task: read the diff and rerun your tests yourself.
7. /codex:transfer sends the whole conversation to Codex.
8. The automatic safeguard loops and burns usage: only turn it on if you are watching.
9. Checked on 10/28/2026 at github.com/openai/codex-plugin-cc (v1.0.6).
You did them allSee all the guides