The Codex plugin for Claude Code, or Plyloft?

If you want Codex to look over what Claude Code just wrote, OpenAI has made that a one-line install: the Codex plugin for Claude Code. For an occasional second opinion on a diff it is the simplest option there is, and it costs nothing beyond your ChatGPT plan. Plyloft, the app I build, does something wider. It runs the whole loop around that review, from planning the card with both AIs to shipping the change, with both CLIs on your own subscriptions, on Mac and Windows.

Before I built Plyloft, I did this by hand: I copied plans and diffs between a Claude terminal and a Codex terminal. The plugin takes most of that copying out of the review step. I haven’t used the plugin myself, so everything about it on this page comes from its README, releases and issues, with links, so you can pick.

What the Codex plugin does

openai/codex-plugin-cc is OpenAI’s own plugin for Claude Code. It was created on March 30, 2026, is Apache-2.0, has about 34,500 GitHub stars, and its latest release is 1.0.6, from July 8, 2026 (repository).

From inside a Claude Code session it gives you:

• /codex:review, a read-only review of the work by Codex;
• /codex:adversarial-review, a review that pushes harder;
• /codex:rescue and /codex:transfer, to hand work over to Codex;
• /codex:status, /codex:result and /codex:cancel to follow a job, and /codex:setup;
• an optional review gate (/codex:setup --enable-review-gate): a Stop hook runs a Codex review when Claude wants to finish, and if the review finds issues, the stop is blocked so Claude deals with them first.

It signs in with a ChatGPT subscription (the Free plan included, according to its README) or an OpenAI API key, and its usage counts against your Codex limits. The README is honest about the cost of the gate: it “may drain usage limits quickly”.

When the plugin is the right choice

You want a second opinion now and then. One command in the session you already have open. Nothing new to learn, no new window.

You live in one Claude Code terminal. If your day is a single session, the plugin fits into it and Plyloft would be one more app.

It costs nothing extra. The plugin is free; the review uses your Codex limits. It even works on ChatGPT Free, which Plyloft doesn’t: Plyloft needs Claude Pro or higher, or ChatGPT Plus or higher.

If that describes you, use the plugin. Plyloft adds little for a review you ask for twice a week.

What the plugin leaves to you

The plugin is a set of commands, not a workflow. From its README and issues:

• there’s no board, no worktree per task, no record of review rounds, no findings placed on the lines of a diff, no shipping step and nothing for a team;
• it goes one way: Codex reviews Claude. There is no official plugin for Claude reviewing Codex, only community projects;
• reviewing the plan before code is written is an open request (issue #4).

Three issues are worth reading before you turn on the review gate. All three were open on October 22, 2026:

• brokers that never stop by themselves, 272 processes and about 2.2 GB of memory in one report (#543);
• ending one Claude session stops the shared broker, so jobs in other sessions die silently (#540);
• /codex:transfer always fails on Windows (#618).

What Plyloft runs around the review

Plyloft is a desktop app for Mac and for Windows 11, 64-bit (x64), with the same features on both. It runs the official claude and codex CLIs in real terminals, signed in to your own Claude and ChatGPT plans, and never asks for an API key. A card goes through these steps:

1. Plan with two AIs (optional, before any code): one AI writes the plan and the other critiques it, for two rounds, three at most, and Use this plan makes it the card’s plan. With only one AI signed in, two separate sessions of it take the two roles. More in Chat & Plan with two AIs.
2. Start build: the Builder works in the card’s own copy of your project, a git worktree, so your main branch is untouched.
3. Finish and check: when the Builder says it’s done, Plyloft saves its work, runs your project’s checks, and the Reviewer reviews the change read-only. Findings read like F1 · Serious — maxAge is in milliseconds in Express, sit under the lines they name, and go back with Send the findings to the Builder. Every round stays on the card. Either vendor can build, and either can review.
4. Try it: Plyloft starts your app from the card’s copy, so you can see the change running before you ship it (Try it).
5. Ship it: on a project on GitHub, opens a pull request and merges it once its checks pass, on every plan; Only on this Mac (Only on this PC on Windows) merges on your machine instead.

On Pro and Team, Build with both has Claude and Codex build the same card side by side, and you keep the better build. On every plan, Review a pull request points the Reviewer at any open pull request, yours or a colleague’s; its findings stay in Plyloft and nothing is posted to GitHub (Review a pull request).

The Board sorts every card by what it needs from you: Needs you, Not started, Working, On hold and Done. Team adds a shared board on GitHub Issues and shared skills.

The plugin and Plyloft, row by row

	Codex plugin for Claude Code	Plyloft
What it is	Commands inside Claude Code	Desktop app around both CLIs
Who reviews whom	Codex reviews Claude	Either way, you choose per card
Plan review	Open request (#4)	Plan with two AIs
Review rounds kept	No	Yes, on the card
Findings on the diff	No interface	Under the line they name
Worktree per task	No	One per card
Project checks before ship	No	At every Finish and check
Ship / pull request	No	Ship it opens a pull request and merges it when checks pass, on Free too
Review someone else’s pull request	No	Review a pull request
Team	No	Shared board on GitHub Issues
Windows	Runs in Claude Code; /codex:transfer fails (#618)	Windows 11, 64-bit (x64), same features as on Mac
Sign-in	ChatGPT plan (Free included) or API key	Claude Pro or higher, or ChatGPT Plus or higher; never asks for an API key
Price	Free	Free, one card building at a time; Pro €9 a month or €86.40 a year; Team €9 per seat a month

Plyloft’s Pro and Team prices are founding prices, kept for as long as a subscription runs (pricing).

Which one I’d pick

I haven’t run the plugin, so this goes by what its own pages say it does. For an occasional look at a diff, the plugin. It’s free, it’s where you already are, and it comes from the people who make Codex.

When I want the plan argued over before any code, rounds I can read again a week later, a worktree per card and a ship step with my checks in it, one command isn’t enough. That loop is why I built Plyloft, and it runs the same way on Windows.

For how the two agents compare on their own, see Codex vs Claude Code.

Plyloft is independent and not affiliated with Anthropic or OpenAI.

If you want the whole loop on your next card, download Plyloft and try it free, one card at a time.