Codex plugin for Claude Code: review code and hand off tasks to Codex
OpenAI's plugin that adds Codex code review and task hand-off to Claude Code with 8 slash commands.
- Skills
- 3
- included
- Hooks
- 3
- included
- Agents
- 1
- included
- Commands
- 8
- included
- Stars
- 34.8K
- on GitHub
This plugin lets you run Codex code reviews and hand tasks to Codex without leaving Claude Code. It needs the Codex CLI and a ChatGPT sign-in or an OpenAI API key, because it drives your local Codex install. Use it to get a second model's review of Claude's changes, or to pass a stuck bug to Codex. Know before you install: the rescue command can edit your files, and every run counts against your Codex usage limits.
Good for
- Claude Code users who want a second model to review their changes
- Developers who want to hand a stuck bug or failing build to Codex
- Teams who pay for Codex already and want it inside Claude Code
- Anyone who wants a design challenged before it ships
Install codex
Claude Code
- 1
Copy this prompt.
Install the codex plugin in Claude Code. Run this command: /plugin marketplace add openai/codex-plugin-cc /plugin install codex@openai-codex /reload-plugins /codex:setup Then check that it loaded and tell me how to use it.
- 2
Paste it into Claude Code. It shows the change and asks before running it.
Requires Claude Code, Node.js 18.18 or later, Codex CLI (setup can install it with npm), ChatGPT account (Free or paid) or OpenAI API key.
Read-only mode: Ask for read-only or diagnosis only when you run /codex:rescue. Reviews are always read-only.
Run the first method in Claude Code. The plugin drives your local Codex CLI, so Codex must be installed and signed in. If /codex:setup finds Codex missing and npm available, it offers to install it. If Codex is installed but not signed in, run !codex login and pick a ChatGPT account or an API key; an existing Codex sign-in on this machine works as is. Usage counts against your Codex limits (pricing). Third-party marketplaces do not update by themselves, so update from the Installed tab in /plugin.
From the project README. Source.
Key facts
- Publisher
- OpenAI (official)
- License
- Apache-2.0
- Version
- 1.0.6 (released 2026-07-08)
- Last commit
- 2026-07-07
- First release
- v1.0.0, 2026-03-30
- GitHub stars
- 34,792 (2026-11-02)
What it includes
Codex reviews your uncommitted changes or branch, read-only
Runs the same review as /review in Codex on your working tree, or on your branch versus --base <ref>. Supports --wait and --background. Takes no custom focus text and never edits code.
Try: “/codex:review --base main”
Runs code ·
/codex:reviewA review that challenges your design, tradeoffs and assumptions
Same targets as /codex:review, but steerable: add focus text after the flags to pressure-test areas like auth, data loss, rollback or race conditions. Read-only.
Try: “/codex:adversarial-review --background look for race conditions and question the chosen approach”
Runs code ·
/codex:adversarial-reviewHand a bug, fix or follow-up task to Codex
Sends the task to the codex:codex-rescue subagent, which runs it in Codex. Supports --background, --wait, --resume, --fresh, --model and --effort. Codex can edit files in your workspace by default.
Try: “/codex:rescue fix the failing test with the smallest safe patch”
Runs code ·
/codex:rescueTurn this Claude Code session into a Codex thread you can resume
Imports the current Claude Code transcript into Codex and prints a codex resume <session-id> command. --source points at another transcript under ~/.claude/projects.
Try: “/codex:transfer”
Runs code ·
/codex:transferList running and recent Codex jobs for this repo
Shows job ID, kind, status, phase and elapsed time for current and recent jobs, plus review-gate status. Pass a job ID for full detail.
Try: “/codex:status task-abc123”
Runs code ·
/codex:statusShow the final output of a finished Codex job
Prints the stored result with findings, file paths and line numbers, and the Codex session ID so you can reopen it with codex resume.
Try: “/codex:result”
Runs code ·
/codex:resultStop a running background Codex job
Cancels the active background job, or the one whose ID you pass.
Try: “/codex:cancel task-abc123”
Runs code ·
/codex:cancelCheck Codex is installed and signed in; toggle the review gate
Reports whether Codex is ready. If Codex is missing and npm exists, it offers to run npm install -g @openai/codex. --enable-review-gate and --disable-review-gate turn the Stop hook review on or off.
Try: “/codex:setup --enable-review-gate”
Runs code ·
/codex:setupSubagent that forwards a task to Codex in one call
Runs on the sonnet model with only Bash. Forwards your request to the Codex companion script once and returns Codex's output unchanged. Its description tells Claude to use it proactively when stuck, and it adds --write unless you ask for read-only.
Try: “Ask Codex to redesign the database connection to be more resilient.”
Runs code ·
codex:codex-rescue (subagent)Records session ID and transcript path for later commands
Runs a local Node script (5 second timeout) that appends three env vars to Claude Code's env file. No network.
Runs code ·
SessionStart hookShuts down the Codex helper and this session's running jobs
Runs a local Node script (5 second timeout) that tells the shared Codex app-server broker to shut down, kills this session's queued or running jobs and removes them from state.
Runs code ·
SessionEnd hookOptional: Codex reviews Claude's last edits and can block stopping
Runs on every turn end (900 second timeout). With the gate off it only notes running jobs. With it on, it runs a Codex task reviewing the previous turn's edits and blocks the stop if Codex answers BLOCK, fails or times out.
Try: “/codex:setup --enable-review-gate”
Runs code ·
Stop hook (review gate)Used by the assistant behind the scenes
Internal rules for how the rescue subagent calls Codex
Not user-invocable. Tells the subagent to make exactly one task call, map flags like --resume and spark, and default to write-capable runs.
codex-cli-runtime (skill)Internal guide for writing tight prompts for Codex
Not user-invocable. Lets the subagent rewrite your request into a structured, XML-tagged Codex prompt, with reference recipes and anti-patterns.
gpt-5-4-prompting (skill)Internal rules for showing Codex output back to you
Not user-invocable. Keeps findings ordered by severity with exact file paths, and tells Claude not to apply fixes from a review until you say which ones.
codex-result-handling (skill)
From the official docs
Permissions
Medium risk
Your code goes to OpenAI for every review, and the rescue command can change files in your project. Ask for read-only in rescue prompts, keep the review gate off, and keep secrets out of the repo you review.
What it can do
- Read your data
- Create and change things
- Delete things
- Read-only modeHow to turn it on
Watch out for
Rescue can edit your files by default
The codex-rescue helper adds write access unless you ask for read-only work, so Claude can start an edit-capable Codex run you did not request. Say "read-only" or "diagnosis only" in the prompt. github.com
Claude Code deny rules do not apply to Codex
Codex reads and writes through its own sandbox, so Read and Edit deny rules in .claude/settings.json do not stop it. This is reported and still open. Do not rely on those rules to protect secrets. github.com
Reviews send your code to OpenAI
A review sends your diff and untracked text files to Codex, which calls OpenAI. An open report says adversarial review follows symlinks and can include files outside the repo. Keep secrets and links to other folders out of the repo you review. github.com
Runs spend your Codex usage or API credit
Each review and task counts against your Codex limits, or is billed at API rates if you sign in with an API key. Watch /codex:status for long jobs. learn.chatgpt.com
Stop hook: the optional review gate can block Claude
With the gate on, the Stop hook has Codex review each turn that edited code, for up to 15 minutes, and stops Claude from finishing if it finds issues. It can loop and use up limits. It is off by default. github.com
SessionStart hook writes to Claude's env file
It runs a local Node script (5 s timeout) that adds three variables to Claude Code's env file. No network calls. Open reports say it repeats on resume. github.com
SessionEnd hook stops the helper and kills jobs
It runs a local Node script (5 s timeout) that shuts down the Codex helper and kills this session's running Codex jobs. Finish or collect results before you close a session. github.com
Setup can install a global npm package
If Codex is missing and npm is present, /codex:setup asks once, then runs the Codex install. Say no if you install tools yourself. github.com
Security advisories
No published security advisories as of Nov 2, 2026.
Technical details
| Turning on read-only | /codex:review and /codex:adversarial-review always run Codex in a read-only sandbox. For /codex:rescue, ask for read-only or diagnosis only, because the helper adds write access by default. Leave the review gate off (default). |
|---|---|
| Talks to | OpenAI |
| Sign-in and keys | Uses the sign-in of your local Codex CLI (ChatGPT account or API key). |
Works in
Built for Claude Code. It drives your local Codex CLI, and jobs it starts can be reopened in Codex with codex resume.
Tap an app to see how to set it up and any known problems.
Claude Code
Has issuesDocumented by OpenAI. Also installs in the desktop Code tab and VS Code extension, not cloud sessions.
Known problems
- Open
Session start hook keeps adding the same lines to the env file, so long sessions can break Bash commands
Since Jul 2026 · GitHub
- Open
Background Codex helper processes are never shut down and pile up, using gigabytes of memory
Since Jul 2026 · GitHub
- Open
Ending one Claude session kills Codex tasks running in another session in the same folder
Workaround: Run /codex:cancel on the stuck job before starting a new one.
Since Jul 2026 · GitHub
- Open
On Windows, reviews and rescue tasks come back empty because Codex's sandbox fails to start
Since May 2026 · GitHub
- Open
Several long-lived Codex helpers can reuse a spent sign-in token and log every Codex client out
Workaround: Sign in again with codex login.
Since Sep 2026 · GitHub
Examples
Review your uncommitted changes in the background
/codex:review --background
Starts a read-only Codex review. Check it with /codex:status and read it with /codex:result.
SourceReview a branch against main
/codex:review --base main
SourceQuick, cheaper fix with a smaller model
/codex:rescue --model gpt-5.4-mini --effort medium investigate the flaky integration test
SourceDelegate in plain words without a slash command
Ask Codex to redesign the database connection to be more resilient.
SourceHand a failing build to Codex
/codex:rescue investigate why the build is failing in CI
Source
What you can do with it
Pressure-test a design before it ships
Challenge the approach on a branch, not just the syntax.
/codex:adversarial-review --base main challenge whether this was the right caching and retry design
Run long work in the background
Start a Codex task and check in later.
/codex:rescue --background investigate the flaky test
Continue a previous Codex fix
Pick up the last rescue run where it stopped.
/codex:rescue --resume apply the top fix from the last run
Get a fast pass from a smaller model
Use the Spark model for a quick fix.
/codex:rescue --model spark fix the issue quickly
Move a Claude Code chat into Codex
Turn the current session into a Codex thread you can reopen with codex resume.
/codex:transfer
Get a second opinion before you commit
Have Codex review your uncommitted changes.
/codex:review
What people say
People like the second-opinion reviews and dislike the speed, the open bugs and the way Codex ignores Claude Code's deny rules.
What people like
- A second model reviewing Claude's work turns up real findings that Claude missed.
- Pairing Claude's plan mode with /codex:rescue gives a plan that has been reviewed by two models.
- Claude and Codex can review each other's work on their own once permissions are set.
Common complaints
- Reviews can be slow, slower than /review in the Codex CLI itself.
- Codex runs ignore Claude Code deny rules in settings.json.
- The repo has many open bugs, and some users ask whether an active fork exists.
- One user found delegating to a GPT model through the plugin produced more over-engineering.
In their words
“I use codex plugin to often validate claude results, surprisingly many findings, so for things like code-review becomes very helpful.”ilia-aHacker NewsApr 2026
“there is an official codex plugin for claude. I just have them do adversarial reviews/implementations. etc with each other. adds a bit of time to the workflow but once you have the permissions sorted it'll just engage codex when necessary”ipkstefGitHub issueSep 2026
“In about two days of use in one workspace: 226 stops were reviewed, with 202 ALLOW, 23 BLOCK and 1 failure. Most blocks were correct and useful.”wardgerard01-hashHacker NewsOct 2026
“Then using the Codex plugin in Claude Code was slow, and I wanted it to feel like a team, not just a supervisor.”chasemonkey
2 more quotesFewer quotes
“I used /codex:review in Claude Code, but it took a long time to return the results. However, when I used codex cli to execute /review (for same diff), it returned quickly.”yinzhou-jcGitHub issueMar 2026
“The Codex plugin bypasses the deny rules and reads files that are explicitly blocked.”deokju
Included in this plugin
3
- codex-cli-runtime
Internal helper contract for calling the codex-companion runtime from Claude Code
800 installs - codex-result-handling
Internal guidance for presenting Codex helper output back to the user
680 installs - gpt-5-4-prompting
Internal guidance for composing Codex and GPT-5.4 prompts for coding, review, diagnosis, and research tasks inside the Codex Claude Code plugin
732 installs
Questions
How do I install the Codex plugin in Claude Code?
Use the install box at the top of this page. It adds OpenAI's openai-codex marketplace, installs the plugin and reloads plugins. Then /codex:setup checks that Codex is installed and signed in, and can install it with npm.
Do I need a ChatGPT subscription for the Codex plugin?
No. A ChatGPT account on any plan works, including Free, and so does an OpenAI API key. OpenAI lists Codex in Free, Go, Plus, Pro, Business, Edu and Enterprise.
Is the Codex plugin free?
The plugin is free and Apache-2.0 licensed. The Codex runs it starts count against your ChatGPT plan's Codex limits. With an API key they are billed at API rates.
Does the Codex plugin use my Codex usage limits?
Yes. The README says plugin usage counts against your Codex limits. Long reviews and the optional review gate can use them up fast, so run /codex:status to see what is running.
Can Codex edit my files through the plugin?
Reviews cannot. They run in a read-only sandbox. /codex:rescue can, and it adds write access by default unless you ask for read-only work. Say "read-only" or "diagnosis only" in the prompt if you only want an explanation.
What is the Codex review gate?
It is an optional Stop hook. After you run /codex:setup --enable-review-gate, Codex reviews Claude's last edits at the end of each turn and can stop Claude from finishing if it finds issues. It can loop and drain usage limits, so OpenAI says to enable it only while you watch the session. Turn it off with /codex:setup --disable-review-gate.
Does the Codex plugin work on Windows?
It runs, but there are open Windows reports. Reviews and rescue tasks can come back empty when Codex's sandbox fails to start, and /codex:transfer has failed. Several earlier Windows bugs were fixed.
What is the difference between /codex:review and /codex:adversarial-review?
/codex:review is the same review as /review in Codex, on your uncommitted changes or a branch. It takes no focus text. /codex:adversarial-review challenges the design and lets you add focus text, such as auth or rollback. Both are read-only.
Codex plugin vs using the Codex CLI directly?
The plugin runs the same local Codex CLI and keeps you inside Claude Code. Using the CLI directly skips the plugin layer. One user reported /review returning quickly in the CLI while /codex:review took a long time on the same diff.
Is the Codex plugin safe?
No security advisories are published. It sends code to OpenAI through Codex, the rescue command can edit files, and Claude Code deny rules do not apply to Codex. Reviews are read-only and the review gate is off by default.
Information
updated Nov 2, 2026
- Runs in
- Claude Code
- Repository
- openai/codex-plugin-cc
Sources
Docs, issue trackers and discussions this page is built from.
Show all 50Hide
- README · github.com
- README: Requirements · github.com
- Codex pricing · learn.chatgpt.com
- Keep plugins updated · code.claude.com
- Codex plugin for Claude Code (README) · github.com
- Publisher · github.com
- License · github.com
- Version · github.com
- Last commit · github.com
- First release · github.com
- Contents · github.com
- Discover plugins (Claude Code docs) · code.claude.com
- Marketplace · github.com
- Announced · community.openai.com
- /codex:review · github.com
- /codex:adversarial-review · github.com
- /codex:rescue · github.com
- /codex:transfer · github.com
- /codex:status · github.com
- /codex:result · github.com
- /codex:cancel · github.com
- setup command · github.com
- codex-rescue agent · github.com
- codex-cli-runtime (skill) · github.com
- gpt-5-4-prompting (skill) · github.com
- codex-result-handling (skill) · github.com
- session-lifecycle-hook.mjs · github.com
- stop-review-gate-hook.mjs · github.com
- hooks.json · github.com
- Community report · github.com
- Rescue edits files by default · github.com
- Adversarial review follows symlinks · github.com
- README: Enabling the review gate · github.com
- Windows: empty review and rescue results · github.com
- README: /codex:adversarial-review · github.com
- Community report · github.com
- README: Moving the work over to Codex · github.com
- README: /codex:rescue · github.com
- README · github.com
- README · github.com
- README: Review before shipping · github.com
- README: Long-running work · github.com
- README: /codex:transfer · github.com
- Community report · news.ycombinator.com
- Community report · news.ycombinator.com
- Community report · github.com
- Community report · news.ycombinator.com
- Hacker News discussion · news.ycombinator.com
- Hacker News discussion · news.ycombinator.com
- Is there an active fork? · github.com