OpenAI Just Shipped a Plugin So Codex Runs Inside Claude Code

The real story isn't the three commands. It's the admission hiding inside the architecture. I keep coming back to what OpenAI shipped on March 30 and 31. They put out codex-plugin-cc, open source under Apache 2.0, so Codex can run inside Claude Code. That caught me off guard a little. I could be wrong, but I can't remember another OpenAI move this direct into a rival dev surface. On paper it's a tiny surface area. Actually, wait, that's not quite right. It's narrow, not small. /codex:review does the read-only pass. /codex:adversarial-review is the skeptical one that goes after tradeoffs and failure modes. /codex:rescue hands the work to a Codex subagent, and the repo also ships /codex:status, /codex:result, plus /codex:cancel for background jobs. Small thing, big signal. What it isn't The part that makes this more interesting is what it isn't. At least from the repo, this doesn't read like a deep MCP-style bridge. It looks like Claude Code plugin plumbing: markdown command files, hooks