wtf is codex-review?

Codex Review is a Claude Code skill that gives you a second opinion on a code diff from a different AI model family. It wraps the OpenAI Codex CLI in a thin, read-only script, so instead of only having Claude review your own code, you also get a review from Codex, which can catch mistakes or blind spots that a same-model reviewer might share with the author. To use it, you copy the skill into your Claude Code skills directory, or install it through SkillsMP. It requires the Codex CLI to already be installed and logged in, if it is not, the skill fails with a clear message rather than trying to work around the missing tool. The tool supports several ways to scope a review: the current branch against an automatically detected base branch like main or master, just the uncommitted changes in your working directory, an explicit base branch, or a single commit. You can also pick which Codex model to use, and instead of a scoped review you can give it a plain prompt, such as asking it to focus on the authentication changes, though a prompt and a scope flag cannot be combined. The script is built to be read-only: it always calls codex review rather than any command that could write to your repository, and it rejects any flag that is not on its allowed list. The author frames Codex output as untrusted input to a review decision that the calling AI agent still owns, meaning findings should be checked and confirmed rather than accepted automatically. The skill is meant to be used for non-trivial features, fixes, or important writing changes such as README updates, and skipped for trivial edits or throwaway scripts. It fails gracefully if the Codex CLI is missing, if you are not inside a git repository, or if Codex is not authenticated. The project is released under the MIT license.