OpenAI just made a genuinely smart distribution move: they shipped an official Codex plugin that runs inside Claude Code. In this video I break down what it is, how to install it, and how the interaction works under the hood—Claude Code collects your diffs, hands them to the locally running Codex CLI/app server, and then Codex sends a review report back into your workflow. The point is you don’t have to leave Claude Code to get a second model’s opinion, which is exactly how OpenAI gets Codex in front of more developers without asking them to switch tools. I walk through the three patterns you’ll actually use: basic `codex review` (one-way review, no file edits), gated/adversarial review (a stop hook triggers a review loop that challenges design decisions, not just syntax), and `codex rescue` (Codex as a sub-agent that can implement changes while Claude Code orchestrates). My core takeaway is simple: don’t use the same model to write and review code—biases compound and you get stuck in loops. Use two providers on purpose, but be aware of token burn when you set up iterative gated review, because you’ll pay Anthropic for regeneration and OpenAI for review in the same loop.