Claude Code calls in Codex for backup

What it does The plugin wires OpenAI Codex into Claude Code as a set of slash commands. You can ask for a standard read-only code review, run an adversarial review that challenges your design choices and trade-offs, or hand off a task to a Codex subagent that investigates or patches while you keep working. It simply wraps your local codex binary, so it piggybacks on whatever auth and config you already have.

The interesting bit The adversarial review mode is unusual: instead of checking syntax, it pressure-tests your assumptions, failure modes, and whether a simpler approach existed. There is also an optional review gate that hooks Claude’s Stop event, forcing Codex to review Claude’s response before the turn ends—effectively making one AI the QA checker for the other.

Key highlights

• /codex:review diffs uncommitted changes or a branch against a base ref, entirely read-only.
• /codex:adversarial-review accepts custom focus text to challenge architecture, hidden assumptions, and risk areas like auth or race conditions.
• /codex:rescue delegates bugs or fixes to a background Codex subagent, with --resume to continue prior threads.
• Job control via /codex:status, /codex:result, and /codex:cancel for long-running background work.
• Reuses your existing Codex CLI install, session auth, and .codex/config.toml settings.

Caveats

• The optional review gate can spin up long-running Claude/Codex loops and drain usage limits rapidly.
• Reviews of multi-file changes and rescue tasks are slow enough that the README explicitly recommends background mode.
• You still need a separate Codex login (ChatGPT subscription or OpenAI API key); Claude Code credentials do not cover Codex usage.

Verdict Try it if you want Codex’s review or task-delegation muscle inside Claude Code without switching terminals. Pass if you are happy with a single model or do not want to manage two separate AI usage meters.