iTranslated by AI

The content below is an AI-generated translation. This is an experimental feature, and may contain errors. View original article
🔌

OpenAI Launches Official Codex Plugin for Claude Code: How to Use It and Decoding the Strategy Behind Coexistence

に公開
AI
DevTools
OpenAI
Claude Code
Codex
tech

What is the official OpenAI codex-plugin-cc?

codex-plugin-cc is an official plugin for Claude Code, released by OpenAI in March 2026. It allows you to invoke Codex directly from within Claude Code using slash commands such as /codex:review or /codex:rescue.

The repository is openai/codex-plugin-cc. It functions as a lightweight wrapper that does not include its own Codex runtime, but instead calls the locally installed @openai/codex binary and the Codex App Server. The design also reuses existing authentication credentials from ~/.codex.

Crucially, because it is a wrapper, all configurations previously set up in Codex are inherited. Custom skills registered in [[skills.config]] of ~/.codex/config.toml, subagents defined in [profiles.<name>], custom instructions in ~/.codex/AGENTS.md, and MCP servers from mcp_servers all remain effective even when calling Codex via Claude Code. You can reuse your existing Codex operational assets within the Claude Code environment.

Which slash commands are added?

Once the plugin is enabled, the following commands are added to your Claude Code session. Their roles are clear when categorized:

Command Role
/codex:review Request a standard code review from Codex for uncommitted or branch diffs
/codex:adversarial-review Challenge-based review that questions design decisions and assumptions (supports optional focus text)
/codex:rescue Delegate bug investigation or fix tasks to Codex
/codex:status / /codex:result / /codex:cancel Check, retrieve results from, or cancel background jobs
/codex:setup Check Codex installation/auth status and enable review gates

All commands support --background. This is practically useful as you can offload long-running reviews or investigations while continuing other work in Claude Code.

How do I set it up?

It takes 3 steps to complete. It works with Node.js 18.18 or higher and a ChatGPT subscription (including Free) or an OpenAI API key.

Step 1: Add to marketplace and install

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

Step 2: Verify the Codex binary

/codex:setup

If Codex is not installed and you are in an environment where npm is available, it will suggest an installation. For manual installation, follow these steps:

npm install -g @openai/codex
!codex login   # Authenticate with your ChatGPT account or API key

Step 3: Verify operation

For the first time, it is easiest to test it using a background review.

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

Since /codex:result includes the Codex session ID, you can reopen the same conversation in the Codex UI using codex resume <session-id>. The workflow is designed to allow you to carry over work started in Claude Code into the Codex UI.

Optional feature: Review gates

The review gate feature, which can be enabled with /codex:setup --enable-review-gate, inserts an automatic Codex review at the exact moment Claude Code is about to finish its response. If an issue is found, it blocks the completion and prompts Claude to make corrections.

Technically, this replaces the Stop hook in Claude Code (a user-defined script that runs just before a response is completed), effectively tasking Codex as an "automatic checker just before finishing."

Where is it most effective?

The greatest benefit is "when you want to offload a second opinion or a heavy task to a separate context without contaminating the main agent's context." You can keep your Claude Code session alive while having Codex evaluate your work from a different perspective.

Specific use cases include:

  • Pre-deployment review: Use /codex:review --base main to check branch diffs.
  • Adversarial design review: Run /codex:adversarial-review --background Challenge the design of the cache and retry logic using focus text.
  • Task delegation: Execute /codex:rescue --background Investigate the root cause of the failing test and continue editing other files in Claude Code.
  • Model selection: Tailor your usage with options like /codex:rescue --model gpt-5.4-mini --effort medium ... to offload to a lighter model.

While /codex:review is for "bug hunting," /codex:adversarial-review is a challenge-based review that questions "if this design was actually good." You can use focus text to specify high-risk design areas, such as rollback safety, race conditions, or authorization omissions.

Why did OpenAI release an extension for a competitor's Claude Code?

In conclusion, it is to expand Codex's touchpoints to "where users already are." As Claude Code gains traction among developers, it became clear that a strategy of "frontal assault" as a standalone product alone could not capture all segments.

There is data to support this:

  • By early 2026, Anthropic reached an ARR of approximately $30 billion, with Claude Code estimated to be a core contributor at $2.5 billion ARR.
  • Claude Code is reported to account for approximately 4% of public GitHub commits.
  • Meanwhile, Codex surpassed 3 million weekly active users in April 2026, a 5x growth in three months.
  • In the same month, OpenAI launched the $100/month "ChatGPT Pro" middle-tier plan, structured with a Codex usage quota 5x that of Plus to target the Claude Max user segment.

In other words, the plugin release is not a gesture of goodwill; it is a strategic move based on the calculation that "if you are confident in the quality of Codex, you will win in total by being called from a competitor's frontend." The fact that /codex:result returns a Codex session ID and is designed with a pathway to bring work back into the main Codex application is a classic indicator of this.

References

For more in-depth coverage of the strategic background covered in this article and real-world operational details of combining Claude Code and Codex, please refer to the ZenChAIne article.

Discussion