Bridging the Divide: OpenAI Codex Plugin Lands in Claude Code

Switching between AI coding assistants is a pain. You lose your place, open another terminal, set up another environment, and break your flow. codex-plugin-cc fixes that by bringing Codex into Claude Code.

If you already use Claude Code as your main workspace, this is the part that matters. You can hand specific tasks to Codex without leaving your session.

What problem this solves

Claude Code is good at holding project context across files and larger changes. But sometimes I want a different kind of help. A skeptical review. A separate agent for a refactor. Something isolated, while I keep moving on the main task.

Before this plugin, that meant juggling separate tools, separate auth, separate output, and separate context. It wastes time. It also muddies the main session and makes it harder to stay focused.

codex-plugin-cc closes that gap. It wraps your local Codex CLI and sends requests from Claude Code to Codex and its GPT models, including GPT-4o and GPT-5.x variants. So you can keep Claude Code for project-wide work, and use Codex for targeted reviews, deeper analysis, or delegated tasks.

How it works

The install is simple:

/plugin marketplace add openai/codex-plugin-cc
/plugin install codex@openai-codex
/reload-plugins

Then run:

/codex:setup

That setup command checks whether the Codex CLI is installed. If npm is available, it can offer to install it for you. It also reuses your existing Codex CLI auth and config, which I like a lot. No second setup spiral. No duplicate configuration. The plugin just calls your system-wide Codex CLI and its app server.

After that, you get these slash commands:

• /codex:review [path/to/files] --base <ref> [--background | --wait] Run a code review on files or directories, optionally against a Git base like main. With --background, the review runs separately while you keep working. You can check progress with /codex:status and fetch the report with /codex:result.

• /codex:adversarial-review [path/to/files] --base <ref> [--background | --wait] [focus text] This one is the interesting command. It questions your implementation, design choices, and failure modes. If you want your assumptions pushed hard before shipping, this is the command.

• /codex:rescue <prompt> Send a thorny task to a Codex subagent and keep it in a persistent thread. Good fit for debugging or refactors that need their own space.

• /codex:transfer <prompt> Create a persistent Codex thread from the current Claude Code session. Useful when you want to hand off a longer task and keep the main session cleaner.

• /codex:status Check active Codex tasks.

• /codex:result Fetch the output when the task is done.

A concrete workflow

Say I’m building a new auth module in Claude Code and want someone to tear it apart before I push.

1. Setup, if you haven’t already

/plugin marketplace add openai/codex-plugin-cc
/plugin install codex@openai-codex
/reload-plugins
/codex:setup

2. Start an adversarial review in the background

/codex:adversarial-review src/auth/ --base main --background "Focus on security vulnerabilities and potential race conditions."

The task runs, and I keep working.

3. Check status

/codex:status

4. Get the result

/codex:result

At that point, Codex gives back a critique of the implementation, including error handling and alternative cryptographic primitives.

5. Hand off a heavier refactor with rescue

/codex:rescue "Refactor the `tokenValidation` function in `src/auth/utils.ts` to be more modular and testable, providing a clear plan before modifying."

That creates a dedicated Codex subagent for the refactor while I keep using Claude Code for the rest of the codebase.

When to use it

Use codex-plugin-cc when you want a different AI perspective or need to offload work without leaving Claude Code.

• Deep, skeptical reviews /codex:adversarial-review pressure-tests code and design choices harder than a standard review.

• Persistent background tasks /codex:rescue and /codex:transfer give longer tasks their own Codex thread, which helps keep the main Claude Code session from getting cluttered.

• Direct access to Codex features If you already use Codex CLI and like its subagent behavior or local sandbox execution, this gives you access from inside Claude Code.

When not to use it

If the task is small and Claude Code can handle it directly, adding Codex is extra overhead. For quick questions or simple edits, Claude Code alone is usually the faster path.

What I keep coming back to is the flow. This plugin removes a bunch of annoying tool switching, and that matters more than it sounds.

Until next time, happy coding 👨‍💻
– Patricio Marroquin 💜