To have Claude Code and Codex review each other's code, run them as two members of one Crewly team, one on the Claude CLI runtime and one on the Codex CLI runtime. The author commits, then messages the other member to review the commit. We ran Claude Code writing and Codex reviewing. The reverse uses the same setup, but we did not run it.
Set it up
- In the Crewly dashboard, create a team with two members. In the New Team form, each member has a Runtime Type dropdown: set one to Claude CLI and the other to Codex CLI. The full walkthrough, with screenshots, is in Claude Code and Codex Working Together.
- Start the team and note both session names in the terminal panel's Session dropdown.
- Give the author a task that includes the review. This is the wording we used:
Fix the bug in slugify.js and add a boundary test, run node --test, commit. Then send a message to the agent session <reviewer-session> asking it to re-review the new commit (git show HEAD) and reply LGTM or concrete problems. Keep the request to 3 lines.
To reverse the direction, give the task to the Codex member and put the Claude Code member's session name in the message.
What to ask the reviewer
- Ask for concrete problems, not a verdict. "LGTM or concrete problems" gives you something to check.
- Ask it to read the commit, not the author's summary. It did not write the code, so it should judge the code.
- Keep the request short. Three lines was enough in our run.
What we saw
In our run, Claude Code committed a function with a planted bug. The Codex reviewer read the commit, ran the tests and named a specific failing case: slugify("hello world", 6) returned hello-. Claude Code fixed it, and the reviewer answered LGTM on the second round.
That is one planted bug on one small task. It shows the channel works and the reviewer did its job. It does not show that either CLI reviews better than the other.
Limits
- macOS on Apple Silicon, one two-member team, one task.
- Which direction works better is untested; do not pick a reviewer because of our single run.
- The Codex member runs without approval prompts or a sandbox, so use a scratch repo first.
FAQ
Can Claude Code review code that Codex wrote?
Yes, with the same setup in the other direction: give the task to the Codex member and name the Claude Code member as the reviewer. We tested only Claude Code writing and Codex reviewing, so treat the reverse as untested.
Do I have to copy the diff between the two tools myself?
Not in a Crewly team. The author sends the reviewer a message through Crewly's send-message skill and the reviewer reads the commit itself with git show HEAD.
Is it safe to let the reviewer run?
Crewly launches the Codex member with -a never -s danger-full-access: no approval prompts and no sandbox. Use a scratch or disposable repo until you are comfortable.