OpenAI Releases Official Claude Code Plugin — What codex-plugin-cc Means

For / Key Points

Audience: Engineers using Claude Code or OpenAI Codex in production. Developers considering using both tools together.

Key Points:

• OpenAI's official plugin codex-plugin-cc enables Codex code review and task delegation directly from Claude Code
• Providing an official plugin for a competitor's tool is unprecedented in the AI coding tool market, marking a strategic turning point
• Practical concerns include Codex usage limits, background execution latency, and dual authentication management

What Happened

On March 30, 2026, OpenAI published the repository openai/codex-plugin-cc on GitHub¹. Once installed through Claude Code's plugin system, it allows calling OpenAI Codex directly from within a Claude Code session.

This is part of the Codex plugin system OpenAI announced in late March². The notable aspect is that the target platform is Anthropic's Claude Code. OpenAI shipped its own AI coding agent into a competitor's ecosystem.

So what exactly can it do?

What codex-plugin-cc Offers

The plugin provides two categories of functionality: review commands and delegation commands.

Review Commands

/codex:review runs a Codex code review against uncommitted changes or branch diffs. It supports --base main for branch comparison and --background for background execution. It's read-only and makes no code changes.

/codex:adversarial-review performs a more aggressive review. It's a "devil's advocate" style review that challenges design trade-offs, hidden assumptions, and failure modes. It can focus on specific risk areas such as authentication, data loss, and race conditions.

Delegation Commands

/codex:rescue hands off the task itself to Codex. It can investigate bugs, attempt fixes, or continue a previous Codex task. Model switching with --model gpt-5.4-mini or --model spark enables cost control based on task complexity.

Task management uses /codex:status (progress check), /codex:result (retrieve results), and /codex:cancel (cancellation).

Workflow Example

The typical flow is to write code in Claude Code, then run the following at a natural breakpoint:

# After implementing in Claude Code, delegate review to Codex
/codex:review --base main --background

# Continue other work while checking progress
/codex:status

# Retrieve review results and fix in Claude Code
/codex:result

Claude Code acts as the "implementer" while Codex serves as the "reviewer." Conversely, /codex:rescue lets you hand implementation to Codex while supervising from Claude Code.

Setup

Prerequisites: Node.js 18.18+ and a ChatGPT subscription (including Free tier) or an OpenAI API key¹.

# Add the marketplace
/plugin marketplace add openai/codex-plugin-cc

# Install the plugin
/plugin install codex@openai-codex

# Verify setup (can auto-install Codex CLI)
/codex:setup

If Codex CLI is not installed, /codex:setup detects this and offers to install it. Authentication uses !codex login with a ChatGPT account or API key.

Review Gate: An Experimental Feature

Basic setup is complete at this point. However, codex-plugin-cc includes one more experimental feature that can spike costs if misused.

Running /codex:setup --enable-review-gate enables a "stop hook" where Codex automatically reviews Claude Code's output. If Codex detects issues, Claude Code's output is blocked until fixes are applied.

However, the README includes a clear warning: loops between Claude Code and Codex can rapidly consume usage limits¹. This feature should only be enabled under human supervision.

How It Differs from DIY Approaches

Calling Codex from Claude Code isn't new. Third-party implementations existed by February 2026, including SKILL.md-based approaches using codex exec directly and the claude-review-loop plugin with Stop Hook auto-triggering³.

Here's how the architectures compare:

Aspect	codex-plugin-cc (Official)	DIY (direct codex exec)
Execution path	Via Codex App Server	Direct codex exec from bash
Session management	Abstracted by plugin	Self-implemented
Loop control	Not customizable	Free to set stop conditions and iterations
Setup cost	Single slash command	SKILL.md / Hook configuration required

The choice is straightforward: codex-plugin-cc to get started quickly, DIY for fine-grained control over review loop behavior. The two aren't mutually exclusive — use the official plugin for daily reviews and DIY for pre-release final audits.

DIY Approach Details

For a step-by-step guide to the codex exec approach, see "Automating Codex Review: Claude Code × Codex Review Loop Implementation Guide."

Why OpenAI Entered a Competitor's Ecosystem

The features are clear. The next question is why OpenAI would ship a plugin into a competitor's ecosystem in the first place.

First, market share realities. Claude Code reached an estimated $2.5 billion in annualized revenue by early 2026 and accounts for about 4% of all public GitHub commits (roughly 135,000 per day)⁴. It's establishing a de facto position in developer workflows.

Meanwhile, Codex has 2 million weekly active users, but its strengths lie in cloud sandbox autonomous execution and enterprise governance⁴. Its characteristics differ from Claude Code's local-execution-first approach.

This plugin represents a strategy of "creating a Codex touchpoint inside Claude Code" rather than waiting for Claude Code users to switch. Even if developers continue using Claude Code as their primary tool, Codex usage accrues through review and task delegation.

The revenue structure reinforces this strategy. Codex charges on a usage basis⁵. Even without acquiring the user outright, every plugin-initiated review generates revenue for OpenAI. Zero user acquisition cost with incremental billing — the "cross-platform plugin" play is extremely low-risk for OpenAI.

Operational Considerations

Four key points to consider before adoption:

• Dual authentication management — Claude Code requires an Anthropic account; Codex requires a ChatGPT account or OpenAI API key. Team adoption increases account management overhead
• Split usage limits — Codex usage counts against ChatGPT subscription limits⁵. Since this is separate from Claude Code limits, teams must monitor both
• Background execution latency — The README repeatedly recommends --background because multi-file reviews take time. Synchronous waiting blocks the Claude Code session
• Model selection cost impact — /codex:rescue lets Codex auto-select models by default. gpt-5.4 ($2.50/$15 per MTok) may be used unintentionally⁸. Explicitly specify --model gpt-5.4-mini or --model spark

Market Context — The Era of Cross-Platform Plugins

The previous section examined OpenAI's individual motivation. Here, we place codex-plugin-cc in the broader market context.

Third-party developers have already published Codex and Gemini plugins for Claude Code, normalizing the pattern of "calling any agent from any agent CLI"⁶.

OpenAI and Anthropic both have mature plugin systems, and Google is building out its own agent extension framework⁷. All three are moving toward bundling MCP servers, skills, and sub-agents. Interface compatibility doesn't exist yet, but the plugin abstraction layer has made inter-agent operability a practical reality.

From a developer's perspective, "choose one primary agent CLI and supplement missing capabilities with plugins" has become a viable multi-agent operational model.

Summary

Technically, codex-plugin-cc is just a wrapper around Codex CLI. But the fact that OpenAI shipped an official plugin into a competitor's ecosystem signals that the AI coding tool market has entered a phase where competition is about ecosystem penetration rather than platform lock-in.

In practice, the most natural workflow is using Claude Code as the primary tool while delegating "second-opinion code reviews" to Codex. Start with /codex:review --background after setting up dual authentication management and usage limit monitoring.

The next thing to watch is whether Anthropic or Google ship similar "reverse plugins." When cross-platform integration becomes bidirectional, the competitive axis for AI coding tools will fully shift from "model performance" to "integrated ecosystem experience." It's time to add "plugin ecosystem maturity" to your agent CLI selection criteria.

Post this article Copy link

Related Articles

• Automating Codex Review: Claude Code × Codex Review Loop Implementation Guide — A 3-stage implementation guide for the codex exec DIY approach covering SKILL.md, Stop Hooks, and pipeline setups

---

1. openai/codex-plugin-cc, GitHub, 2026-03-30. https://github.com/openai/codex-plugin-cc ↩↩↩

2. "OpenAI's Codex gets plugins," The New Stack, 2026-03-27. https://thenewstack.io/openais-codex-gets-plugins/ ↩

3. SmartScope, "Codex Reviewを自動化する：Claude Code × Codex レビューループ実装ガイド," 2026-03-06. https://smartscope.blog/blog/claude-code-codex-review-loop-automation-2026/ ↩

4. "OpenAI Codex Plugins Target Enterprises, Not Developers," Implicator.ai, 2026-03-29. https://www.implicator.ai/openai-wants-codex-to-be-a-platform-developers-already-made-claude-code-one/ ↩↩

5. Pricing – Codex, OpenAI Developers. https://developers.openai.com/codex/pricing ↩↩

6. claude-tools: A Plugin Marketplace for Claude Code, paddo.dev. https://paddo.dev/blog/claude-tools-plugin-marketplace/ ↩

7. "OpenAI introduces plugins for its Codex programming assistant," SiliconANGLE, 2026-03-27. https://siliconangle.com/2026/03/27/openai-introduces-plugins-codex-programming-assistant/ ↩

8. Introducing GPT 5.4, OpenAI, 2026-03-05. https://openai.com/index/introducing-gpt-5-4/ ↩