Codex Skill for Claude Code: The Official Codex Plugin

If you searched for a "codex skill for Claude Code," what you want is a plugin. OpenAI publishes an official one, codex-plugin-cc, that lets you use Codex in Claude Code for two jobs: reviewing your code and taking over a task you hand it.

You install it from inside Claude Code in four commands, and it adds eight /codex: slash commands plus a Codex subagent. It doesn't bring its own copy of Codex. It drives the Codex CLI already on your machine, with your existing sign-in and config.

The quick version:

Below: what you need first, the install steps, each command in practice, and the one setting that can burn through your usage.

Is it a Codex skill or a Codex plugin?

Both words show up in searches, so here's the distinction.

In Claude Code, a skill is a single set of instructions Claude can load. A plugin is a bundle that can hold skills, slash commands, subagents, hooks, and MCP servers together. Our Claude Code plugins guide covers that structure in detail.

OpenAI's Codex plugin uses most of those parts:

• Eight slash commands, one file each, under the codex: namespace
• One subagent, codex:codex-rescue, which shows up in /agents
• Hooks on session start, session end, and stop
• Three skills, all marked user-invocable: false

That last point answers the search directly. The plugin does contain skills, but they're internal helpers that tell Claude how to call the Codex runtime, how to present results, and how to prompt Codex. You never run them yourself. The commands are the interface.

The repo is open source under the Apache 2.0 license.

Requirements before you install

The plugin's README lists two:

1. A ChatGPT subscription or an OpenAI API key. The README notes this includes ChatGPT Free. Whatever you use, plugin runs count toward your Codex usage limits.
2. Node.js 18.18 or later.

You also need the Codex CLI itself, but the setup command can install it for you. If you haven't used Codex before, our OpenAI Codex guide explains the CLI, sign-in, and sandbox modes.

How to install the Codex plugin in Claude Code

Run these inside a Claude Code session, in order.

1. Add OpenAI's marketplace:

/plugin marketplace add openai/codex-plugin-cc

2. Install the plugin:

/plugin install codex@openai-codex

codex is the plugin and openai-codex is the marketplace name the repo registers. Per Claude Code's plugin docs, running /plugin install in a session opens the plugin's details in the /plugin panel first, so you can review it and choose a scope before it installs.

3. Reload plugins:

/reload-plugins

4. Check Codex is ready:

/codex:setup

/codex:setup checks whether Codex is installed and signed in. If Codex is missing and npm is available, it offers to install it. You can also install it yourself with npm install -g @openai/codex.

If Codex is installed but not signed in, the README's fix is !codex login. The ! runs the command from inside your Claude Code session, and codex login accepts either a ChatGPT account or an API key.

You'll know it worked when the /codex: commands appear and codex:codex-rescue is listed in /agents.

Codex reviews: /codex:review and /codex:adversarial-review

These two commands are the main reason people install the plugin. Claude wrote the code; a different model reads it cold. Both are read-only. Neither one changes files.

/codex:review

A standard review of your current work. The README says it gives the same quality of review as running /review inside Codex directly.

/codex:review
/codex:review --base main
/codex:review --background

• With no flags, it reviews your uncommitted changes.
• --base <ref> reviews your branch against a base branch.
• --wait and --background choose whether you wait for the result.

It isn't steerable. You can't add focus text. Multi-file reviews can take a while, so the README recommends running them in the background.

/codex:adversarial-review

The steerable version. Instead of checking code details, it questions the approach: tradeoffs, hidden assumptions, failure modes, and whether something simpler would have been safer.

/codex:adversarial-review --base main challenge whether this was the right caching and retry design

It takes the same --base, --wait, and --background flags, plus free text after them. Use it before shipping anything where the design is the risk, such as auth, data migrations, rollback paths, or concurrency.

Delegating work: /codex:rescue and /codex:transfer

/codex:rescue

This hands a task to Codex through the codex:codex-rescue subagent. Unlike the review commands, Codex can change code here.

/codex:rescue investigate why the tests started failing
/codex:rescue fix the failing test with the smallest safe patch
/codex:rescue --resume apply the top fix from the last run
/codex:rescue --background investigate the regression

It supports --background, --wait, --resume, and --fresh. Leave out both --resume and --fresh and the plugin can offer to continue your latest rescue thread in that repo. You can also pick a model and reasoning effort with --model and --effort. Skip them, and Codex uses its own defaults.

You don't have to type the command, either. Telling Claude something like "ask Codex to redesign the database connection" works too, because Claude can route the request to the rescue subagent.

/codex:transfer

This one moves the conversation itself. It creates a persistent Codex thread from your current Claude Code session and prints a codex resume <session-id> command you can run to keep going in Codex.

A plugin hook supplies the current transcript path automatically. Two limits from the README: the source transcript must live under ~/.claude/projects, and older Codex versions without session import need upgrading first.

Managing background jobs

Long reviews and rescues are best run in the background. Three commands manage them:

Each one takes an optional job ID, such as /codex:result task-abc123. With the session ID from /codex:result, you can reopen that run in Codex with codex resume and continue the work there.

The README's suggested first run puts all of this together:

/codex:review --background
/codex:status
/codex:result

The review gate, and why it can drain your usage

/codex:setup also controls an optional review gate:

/codex:setup --enable-review-gate
/codex:setup --disable-review-gate

When it's on, the plugin uses a Claude Code Stop hook to run a targeted Codex review every time Claude finishes a response. If Codex finds issues, the stop is blocked and Claude has to address them first.

That's an automatic second reviewer on every turn. It's also the riskiest setting in the plugin. The README warns that the gate can create a long-running Claude and Codex loop and may drain usage limits quickly. Turn it on only for sessions you're actively watching.

How the plugin uses your existing Codex setup

The plugin wraps the Codex app server and calls the global codex binary on your machine. In practice:

• Same install. It runs the Codex CLI you'd use directly.
• Same sign-in. If you're already logged in to Codex, the plugin works immediately.
• Same config. It reads ~/.codex/config.toml, plus a project-level .codex/config.toml when that project is trusted. Set a default model or reasoning effort there and the plugin picks it up.
• Same machine. Codex works on the same repository checkout and local environment as Claude Code.

That design means the plugin adds no new account, key, or runtime to manage. It also means whatever limits apply to your Codex account apply here too. If you're deciding which agent should be your main one in the first place, our Claude Code vs Codex comparison covers where they still differ.

Where Taku fits

The plugin is a neat fit if you already live in a terminal with two coding agents installed. Plenty of people never got either one running. If the setup step is where you keep stalling, Taku mirrors a working AI setup into a desktop workspace and runs it, so you can use a power user's configuration without reproducing their environment first. Taku is in Beta, and the Mac app is available now.

FAQ

Is there an official Codex skill for Claude Code?

OpenAI ships it as a plugin, not a standalone skill: codex-plugin-cc on GitHub. It bundles slash commands, a subagent, hooks, and three internal skills. You install the plugin and use its /codex: commands; the skills run behind the scenes.

How do I install the Codex plugin in Claude Code?

In a Claude Code session, run /plugin marketplace add openai/codex-plugin-cc, then /plugin install codex@openai-codex, then /reload-plugins. Finish with /codex:setup to confirm Codex is installed and signed in.

Do I need a separate Codex account to use Codex in Claude Code?

No separate account for the plugin. It uses your local Codex CLI sign-in. If you've never used Codex, sign in with a ChatGPT account or an OpenAI API key, for example with !codex login from inside Claude Code.

Does the Codex plugin change my code?

The review commands don't. /codex:review and /codex:adversarial-review are read-only. /codex:rescue is the delegation command, and that one can investigate and apply fixes.

Does using the plugin cost extra?

The plugin itself is open source. Its runs count toward your Codex usage limits under whatever ChatGPT plan or API key you sign in with, so check OpenAI's own Codex pricing for your account.

Can I continue a Codex task outside Claude Code?

Yes. /codex:result shows a Codex session ID when available, and codex resume reopens that run in Codex. /codex:transfer goes further and turns your whole Claude Code conversation into a Codex thread.

Key points

• "Codex skill for Claude Code" means OpenAI's official codex-plugin-cc plugin; its three skills are internal.
• Requirements: a ChatGPT subscription or OpenAI API key, and Node.js 18.18 or later.
• Install with /plugin marketplace add openai/codex-plugin-cc, /plugin install codex@openai-codex, /reload-plugins, then /codex:setup.
• /codex:review and /codex:adversarial-review are read-only; /codex:rescue delegates real work; /codex:transfer moves the session.
• The review gate runs Codex on every stop. Useful, but it can drain usage fast.