Claude codes, Codex reviews

Claude Code implements a change on your machine, Codex reviews the diff, and nothing is pushed before you validate it. Two CLIs, your subscriptions.

Two models are better than one at catching what a single agent misses. This workflow has Claude Code write a change in your repository, then hands the diff to Codex for a review, the way a colleague would read your pull request. Both run on your machine, with the subscriptions you already pay for: SideHub asks for no API key and resells no model access.

How it works

1. Claude Code implements the change you describe, runs the project's tests, and commits on the current branch. It never pushes.
2. Codex reads that diff and the code around it, without changing anything, and lists its findings by severity: blocking, important, minor.
3. The work waits in To review: you read the diff and Codex's review side by side, from your computer or your phone. Ask for a fix with the remarks you keep (Claude picks its session back up), or validate.
4. Once validated, the commits show up in To push: you confirm, and the agent pushes exactly the head you read.

The cost of each step is measured, per CLI.

What you need

• An agent connected to the repository's machine, with Claude Code and Codex both logged in. Nothing else: no secret, no GitHub token, no extra tool.

Steps

1. 1. Claude Code: implement the change

   Implement this change in the git repository of the current directory:
   
   {{inputs.change}}
   
   1. Record the current commit before touching anything: `git rev-parse HEAD` (write `none` if the repository has
      no commit yet). This is the base of your work.
   2. Read the project's conventions first: CLAUDE.md, AGENTS.md, CONTRIBUTING.md, README, and the linter and
      formatter configuration.
   3. Implement the change, keeping it focused: no unrelated refactoring. Add or update the tests it needs.
   4. Build the project and run its tests. Fix what you broke.
   5. Commit your work on the current branch with a message that describes the change. Never push, never switch
      branch, never rewrite existing commits.
   Write a summary as your output: what you changed and why, the files touched, the tests run and their result,
   and anything left open. Complete the step with `--value base=<the commit recorded in step 1>`.

2. 2. Codex: review the change

   You are reviewing a change another coding agent (Claude Code) just made in the git repository of the current
   directory. Its summary is in your input. Review it read-only: do not modify, create or delete any file in the
   repository, do not commit, do not push. Use a temporary directory outside the repository if you need one.
   
   1. Read the project's conventions: CLAUDE.md, AGENTS.md, CONTRIBUTING.md and the linter configuration.
   2. Read the change: `git log --stat {{steps.1.outputs.base}}..HEAD` and `git diff {{steps.1.outputs.base}}`
      (if the base is `none`, review every commit: `git log --stat` and `git show`). If there is no change, write
      one line saying so and complete the step.
   3. Read the code around each change, not only the diff, and check it against the request:
   
      {{inputs.change}}
   
   4. Look for bugs, missing edge cases, security issues, missing or weak tests, and departures from the project's
      conventions. Ignore style preferences the conventions do not mention. Run the tests if that helps you
      confirm a finding.
   Write your review as your output, in Markdown: a one-line verdict (ready to ship, or needs a fix), then the
   findings grouped by severity (blocking, important, minor), each with file:line, the problem and the fix you
   suggest. Write "No finding" under a severity that has none. Complete the step with
   `--value blocking=<number of blocking findings>`.

FAQ

Does anything get pushed without me?

No. Claude commits locally, Codex only reads. Pushing waits for you, in To push, after you validated the work in To review.

Why a second CLI for the review?

A different model reads the change with fresh eyes and catches what the author misses. You keep the remarks that matter and drop the others.

Can I swap the roles?

Yes. Each step has its own CLI: open the workflow and pick Codex for the code and Claude Code for the review, or any CLI your agent has.

Can I ask for another review after a fix?

Yes. On the review page, "Have Codex review it" runs a new peer review of the fix, with one checkbox and one comment per recommendation.

Use this template (free) · All templates