Codex Plugin for Claude Code Review and Rescue Setup

If you implement in Claude Code but still want a second-line Codex review or a scoped handoff without leaving the session, OpenAI’s official codex-plugin-cc is the bridge.

📑Table of Contents

What codex-plugin-cc is
Install path: marketplace to setup
Command comparison table
Practical flow: Claude implements, Codex reviews
Limits, cost, and review gate
FAQ
Author Perspective
Summary checklist

It installs from the Claude Code marketplace and routes work to your local Codex CLI and app server, so existing login state, config, and MCP settings are reused.

What this article locks down

This article focuses on the following:

install steps
command boundaries (review / adversarial / rescue / transfer)
a practical Claude-implements / Codex-reviews loop
review-gate plus usage-limit risks

The goal is not star counts; it is knowing which command to run, and when to stop.

What codex-plugin-cc is

What it does

codex-plugin-cc is an official plugin that lets Claude Code call Codex for read-only reviews or task delegation.

It is not a separate remote runtime product. It delegates to local Codex CLI + app server and inherits:

authentication already used by Codex
~/.codex/config.toml and trusted project .codex/config.toml
repository context and MCP configuration

As of the July 2026 survey window:

repo: openai/codex-plugin-cc
license: Apache-2.0
example latest release: v1.0.6 (2026-07-08)
auth path: ChatGPT subscription (including Free) or an OpenAI API key

Core commands

Core commands:

/codex:review — standard read-only review
/codex:adversarial-review — steerable design/assumption challenge (still read-only)
/codex:rescue — investigate/fix handoff (writes possible)
/codex:transfer — move Claude session context into a Codex thread
/codex:status / /codex:result / /codex:cancel — background job control
/codex:setup — environment check and review-gate toggle

Best fit: developers already living in Claude Code who want Codex as a second opinion or a scoped worker.

If you need full standalone Codex workflows every day, pure CLI may still be better.

Install path: marketplace to setup

Official four steps

Official install flow:

/plugin marketplace add openai/codex-plugin-cc
/plugin install codex@openai-codex (choose user / project / local scope)
/reload-plugins if slash commands are missing
/codex:setup (may offer npm install when Codex is absent)

Optional manual path and prerequisites

Optional manual path:

npm install -g @openai/codex
!codex login

Prerequisites:

Node.js 18.18+
Claude Code
ChatGPT or API-key auth

Plugin usage counts against Codex usage limits, so “free install” is not the same as free reviews.

Done signals and config

Done signals:

slash commands such as /codex:review appear
/agents shows codex:codex-rescue
/codex:setup reports Codex ready

Tune defaults with .codex/config.toml (model, model_reasoning_effort).

Project config applies only when the repo is trusted
Rescue can override with --model / --effort

Command comparison table

Command matrix

Command	Mutates code?	Primary use	Key flags	Best for
`/codex:review`	Read-only	Standard review	`--base` `--background` `--wait`	Pre-ship second opinion
`/codex:adversarial-review`	Read-only	Challenge design/assumptions	`--base` + focus text	Auth, races, rollback risk
`/codex:rescue`	May write	Investigate / fix / continue	`--resume` `--fresh` `--model` `--effort` `--background`	Scoped bug hunts
`/codex:transfer`	Session handoff	Claude context → Codex	optional `--source`	Resume stuck Claude work in Codex
`/codex:status` / `result` / `cancel`	Control	Job progress / output / stop	task id	Background operations

Source

Source: openai/codex-plugin-cc README (as of July 2026)

Usage notes

review is the default second-pass review; custom focus text is not the main path.
adversarial-review accepts focus after flags for risk-shaped questions.
rescue maps spark to gpt-5.3-codex-spark; omitted model/effort uses Codex defaults.
multi-file reviews can be slow — prefer --background.

Official plugin vs DIY vs pure CLI

Independent write-ups often contrast three paths:

daily dual-check → official plugin
tightly controlled audit loops → DIY codex exec
full Codex power alone → pure CLI

Practical flow: Claude implements, Codex reviews

Default loop

Default loop:

Implement in Claude Code
Run /codex:review --background
Poll with /codex:status
Read /codex:result and fix in Claude

High-risk changes and rescue / transfer

For high-risk changes, use /codex:adversarial-review --base main plus focus text (auth, billing, migrations, concurrent writes, destructive schema).

For investigation or minimal fixes, use /codex:rescue --background ....

Continue context with --resume
Reset viewpoint with --fresh
Because rescue can write, keep the scope narrow and review results before merge

Use /codex:transfer when you want Claude history seeded into a Codex thread rather than a pure review pass.

Decision shortcuts

Decision shortcuts:

second opinion only → review (and adversarial-review when risk is high)
scoped investigation/fix → rescue
full context move → transfer
always-on stop gate → generally no (see next section)

Limits, cost, and review gate

Three boundaries matter before heavy use.

Usage limits and pricing

Usage contributes to Codex limits.

The official Codex pricing page documents Free and paid ChatGPT paths plus API-key billing, and plan windows where message ranges vary widely by model.

Key points:

Plan windows share Codex usage limits
Message ranges vary widely by model
Prompt length alone is a weak estimator

Multi-file review latency

Multi-file review is slow.

Background jobs plus status/result are more operable than blocking waits.

Review gate caution

Review gate (/codex:setup --enable-review-gate) injects a Codex review on stop and can block Claude from finishing when issues remain.

The official README warns that long Claude/Codex loops can burn usage quickly. Enable only when you can supervise a high-risk session.

Troubleshooting cues

Troubleshooting cues:

Node below 18.18 → upgrade
auth failures → codex login + /codex:setup
huge diffs / buffer errors → shrink the base/diff, use background
runaway review gate → stop the loop under supervision and --disable-review-gate

FAQ

Q1. Do I need a separate Codex account?

If Codex is already logged in on the machine, reuse that auth. Otherwise sign in with ChatGPT or an API key and verify with /codex:setup.

Q2. Is the plugin a separate runtime?

No. It delegates to local Codex CLI and app server and shares config/MCP.

Q3. review vs adversarial-review?

Both are read-only. review is the standard pass; adversarial-review is steerable toward design and assumption challenges via focus text.

Q4. Should review gate stay on always?

No. Official guidance warns about rapid usage burn. Keep it off unless the session is high risk and supervised.

Q5. How do I change model or effort?

Via .codex/config.toml or rescue flags --model / --effort. Project config applies only when trusted.

Author Perspective

In my experience, reviewing code with a different model can reveal issues the original model missed. That makes /codex:review a practical way to use a Codex model while continuing to work in Claude Code.

Even then, the center of value is bounded dual-check—not always-on review:

Keep implementation in Claude when that is the primary workflow
Use Codex as a second opinion when needed, not as a permanent gate

Run one background review first; leave review gate off unless the session is supervised.

For standalone Codex CLI command depth, see the Codex CLI command guide.

Summary checklist

marketplace add → install → reload → /codex:setup until Codex is ready
first run: /codex:review --background → status → result
use adversarial-review only for high-risk changes with focus
keep rescue scoped and review outputs before merge
leave review gate off unless supervised
size budget with Codex pricing and model choice

codex-plugin-cc is useful when you want orchestration, not lock-in: Claude for implementation tempo, Codex for a second opinion or a scoped handoff, with cost and loop risk designed in from day one.

Related articles:

Related new article:

Keystroke n8n Alternative for Claude Code and Cursor: Setup and Fit Check – This published update adds current operational context for Codex Plugin for Claude Code Review and Rescue Setup.
1Password for Claude zero-exposure: login and OTP without sharing secrets – This published update adds current operational context for Codex Plugin for Claude Code Review and Rescue Setup.
Docker Sandboxes (sbx) Guide: Run Claude Code Safely in a microVM – This published update adds current operational context for Codex Plugin for Claude Code Review and Rescue Setup.
Adversarial Verification for Claude Agents: Practical Checklist – This published update adds current operational context for Codex Plugin for Claude Code Review and Rescue Setup.
Role-Based Model Routing for Claude Agent Teams (2026) – This published update adds current operational context for Codex Plugin for Claude Code Review and Rescue Setup.

Author

krona23

Over 20 years in the IT industry, serving as Division Head and CTO at multiple companies running large-scale web services in Japan. Experienced across Windows, iOS, Android, and web development. Currently focused on AI-native transformation. At DevGENT, sharing practical guides on AI code editors, automation tools, and LLMs in three languages.

DevGENT about →

📚 Read Next