I have a `codex-review` skill with a shell script that uses the Codex CLI with a prompt. It tells Claude to use Codex as a review partner and to push back if it disagrees. They will go through 3 or 4 back-and-forth iterations some times before they find consensus. It's not perfect, but it does help because Claude will point out the things Codex found and give it credit.