What it does This plugin embeds OpenAI’s Codex agent inside Anthropic’s Claude Code. You get slash commands—/codex:review, /codex:rescue, /codex:status, etc.—that hand off tasks to Codex while staying in Claude’s chat interface. It wraps your local codex binary and app server, so authentication, config, and runtime all piggyback on whatever Codex setup you already have.

The interesting bit The “adversarial review” mode is the standout: you can steer Codex to challenge your design choices, tradeoffs, and failure modes rather than just linting your syntax. There’s also a Stop hook review gate that can block Claude from proceeding if Codex finds issues—though the README warns this can spawn “long-running Claude/Codex loops” that chew through usage limits.

Key highlights

• /codex:review for read-only code review; /codex:adversarial-review for steerable design critique
• /codex:rescue delegates bug investigation or fixes to Codex as a background subagent
• Background jobs with /codex:status, /codex:result, /codex:cancel—fire and forget, check back later
• Resumes seamlessly in native Codex via session IDs if you outgrow the chat interface
• Picks up existing ~/.codex/config.toml or project-level .codex/config.toml automatically

Caveats

• Requires separate Codex authentication (ChatGPT sub or OpenAI API key) even if you’re already a Claude Code user
• The review gate is explicitly flagged as potentially expensive and loop-prone
• Multi-file reviews and rescue tasks “might take a while”; background mode is strongly recommended

Verdict Worth a look if you’re already living in Claude Code but want Codex’s agentic execution for reviews or background tasks. Skip it if you only use one tool or dislike the idea of two AI agents billing against your limits simultaneously.