GPT inside Claude Code: Claude builds, GPT reviews
- GitHub stars
- 34K+
- modes: review, challenge, task
- 3
- official plugin
- OpenAI
Claude says "done", and then it doesn't work. A model is bad at catching its own mistakes, and a different model catches them. OpenAI's official plugin runs GPT right inside Claude Code: Claude builds, GPT reviews, and it can take tasks off Claude's plate.
What it is
The Codex plugin for Claude Code, made by OpenAI. It adds commands to review your latest changes, stress-test a decision, or hand a task to GPT. It runs on your ChatGPT account, and the free plan works. About 35K stars on GitHub.
Steps
Step 1. Open your project
Open Claude Code in your project folder.
Step 2. Paste the prompt
Paste the prompt below. Claude checks what's needed and installs anything missing, Codex and the plugin itself.
Step 3. Reload and sign in
When Claude is done, type /reload-plugins yourself, then /codex:setup. If Codex needs a sign-in, Claude starts it for you: a browser window opens, and you sign in to ChatGPT yourself.
Step 4. Review after any task
After any task, type /codex:review and GPT goes through what Claude did.
Step 5. Challenge or hand off
To make GPT challenge one specific decision, type /codex:adversarial-review followed by what to check. To hand GPT a whole task, type /codex:rescue and the task.
Paste into Claude Code
Install OpenAI's official plugin for Claude Code (github.com/openai/codex-plugin-cc). First read its README, check what it needs (Node.js 18.18 or newer) and install anything missing for my system. Install Codex and the plugin itself: add the openai/codex-plugin-cc marketplace and install the codex@openai-codex plugin. If any part can only be typed by me, tell me exactly what to type. At the end, tell me what to type myself next: /reload-plugins, then /codex:setup, and if a sign-in is needed, start the ChatGPT sign-in and walk me through it. I'm not a developer: do everything you can yourself.
Want this working for your business?
- AI agents for your business
- Automations for the work you repeat
- 1:1 consulting calls
Good to know
Reviews don't change your code
Review only reads. You or Claude make the fixes after reading the notes.
When to hand a task to GPT
If a task is long or Claude keeps going in circles, send it with /codex:rescue. It runs on your ChatGPT limits, not Claude's.
Background reviews
Add --background to a review, then check it with /codex:status and /codex:result.
The honest read
- Free ChatGPT has limits, and frequent reviews will hit them sooner.
- The second model makes mistakes too. Read its notes, don't just accept them.
- Claude can guide the ChatGPT sign-in, but you type your own login.
Want this built for your business?
I build AI agents and automations for businesses, and do 1:1 consulting calls. Message me on Instagram with what your business does, and I'll tell you what can be automated.
Message me on Instagram