Intriguing Move: OpenAI Turns Codex into a Plugin and Embeds It Directly into Claude Code

Just moments ago, OpenAI transformed its own Codex into a plugin for Claude Code.

You read that right!

A competitor's tool has voluntarily entered rival territory, embedding itself as a plugin within the opponent's ecosystem.

Image

Vaibhav (VB) Srivastav, a developer at OpenAI, announced that Codex is now officially available as a Claude Code plugin, installable with a single command:

/plugin marketplace add openai/codex-plugin-cc

The internet immediately erupted, with comment sections filled with exclamations of "brazen."

Codex Claude Code plugin article header image
Codex Claude Code plugin article header image

01 How to Use It

After installation, you will find three new commands in Claude Code, all starting with /codex::

/codex:review: A standard read-only code review, letting Codex walk through your code.

/codex:adversarial-review: An adversarial review designed specifically to pick apart and challenge your implementation logic.

/codex:rescue: Hands the task directly to Codex, effectively switching to a different agent to take over.

The first two are for review, while the last is for execution. The logic is clear.

Illustration of the three main Codex plugin commands
Illustration of the three main Codex plugin commands

Installation is simple but requires a ChatGPT subscription (including the free tier) or an OpenAI API key, along with Node.js version 18.18 or higher. If you haven't installed the Codex CLI locally yet, run:

npm install -g @openai/codex

Then log in:

!codex login

Once configured, a quick /codex:setup gets you started.

Screenshot of Codex demonstration video within Claude Code
Screenshot of Codex demonstration video within Claude Code

02 Why Do This?

VB explained the use cases in an accompanying long-form post.

The core value lies in: obtaining a secondary review from a different agent without ever leaving Claude Code.

Recommended usage is straightforward:

For daily code changes, run /codex:review.

For high-risk operations like database migrations, permission changes, or infrastructure scripts, use /codex:adversarial-review to let Codex specifically challenge your assumptions.

Stuck on a thread? Or perhaps you just want to try a different agent? Use /codex:rescue to hand it off directly.

Longer tasks can even run in the background:

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

03 Everything Runs Locally

Technically, this plugin leverages your local Codex CLI and Codex app server.

This means it utilizes your existing local authentication, configuration, environment variables, and MCP settings. It does not spin up a separate runtime.

As VB put it:

"This is not a standalone runtime. It is Codex itself, simply invoked from within Claude Code."

Therefore, it should feel quite lightweight, essentially just shouting across the room for the neighbor, Codex, to come lend a hand within Claude Code.

There is also an optional feature called "review gate":

/codex:setup --enable-review-gate

When enabled, Claude Code will wait for Codex to complete its review before finishing its own process.

While this sounds great, VB also warned that it could lead to a loop between Claude and Codex, causing token consumption to skyrocket.

04 Penetrating Enemy Territory

The move itself is even more thought-provoking than the plugin's functionality.

OpenAI turning its product into a plugin for Anthropic's flagship tool is akin to... Coca-Cola approaching Pepsi and saying, "Hey, save a spot for me in your vending machine, would you?"

When competitors run inside each other's tools, the moat is clearly no longer at the IDE layer.

OpenAI has evidently realized one thing: Go where the users are.

Illustration of competitors embedding strategies
Illustration of competitors embedding strategies

Claude Code is currently one of the most active terminal tools in the AI programming space. Instead of waiting for users to switch to their own platform, OpenAI decided to deliver Codex directly to their doorstep.

The logic behind this mirrors Microsoft's decision to bring Office to the iPad back in the day. You can talk about competition all you want, but your product must follow where the user's workflow goes.

You might ask: If Codex is truly that good, why run it inside a competitor's product?

But whether a tool is good is one thing; whether users can actually access it is another.

Many developers (myself included) have already built their daily workflows around Claude Code. No matter how good Codex is, if it requires switching environments to use, the friction cost alone is enough to deter a large portion of users.

To be honest, I even created my own /codex commands to use Codex within CC previously...

By making it a plugin, friction is reduced to nearly zero, making it much more convenient for users.

05 What's Next?

The plugin ecosystem for Claude Code is just getting started. With this move, OpenAI has arguably helped Anthropic validate one thing: The value of the plugin marketplace may exceed that of the IDE itself.

When competitors are willing to integrate their capabilities as plugins, the ecosystem flywheel for the Claude Code platform begins to turn.

And for developers like us, this is good news.

You no longer have to pick sides.

In the same terminal, Claude acts as the main force, while Codex handles reviews, each performing its own duty.

The competition between tools has finally evolved into collaboration between tools.

And the ones happiest about this, of course, are you sitting in front of the terminal!

◇ ◆ ◇

Relevant Links:

Plugin Repository: https://github.com/openai/codex-plugin-cc

Original Tweet: https://x.com/reach_vb/status/2038671858862583967

Original Article: https://x.com/reach_vb/article/2038670509768839458


分享網址
AINews·AI 新聞聚合平台
© 2026 AINews. All rights reserved.