OpenAI Releases Codex Plugin for Claude Code to Delegate Tasks Locally
OpenAI launched an integration enabling Claude Code to delegate intensive tasks, trigger code reviews, and hand off background debugging sessions to Codex.
Impact: High
Why it matters
You can now execute targeted adversarial code reviews and trigger background task rescue operations without leaving your active Claude terminal.
TL;DR
- 01Enables bi-directional session handoff between Claude Code and local Codex environments.
- 02Allows asynchronous code reviews and debugging tasks to run without blocking your prompt line.
- 03Supports targeted adversarial reviews focusing specifically on design flaws and risk areas.
Key facts
- Supported Models
- gpt-5.4-mini, gpt-5.3-codex-spark (mapped via 'spark')
- Supported Flags
- --background, --wait, --resume, --fresh
Seamless Claude Code and Codex Integration
The new openai/codex-plugin-cc plugin embeds Codex's task delegation and code review workflows directly into your Claude Code workspace. By invoking /codex:setup, developers can verify local CLI authentication or prompt the plugin to auto-install Codex using npm if it is missing.
Adversarial Reviews and Task Rescue
Use /codex:adversarial-review with optional --base <ref> flags to pressure-test design tradeoffs, race conditions, or cache invalidation logic before pushing a branch. To offload heavy debugging or implementation tasks, run /codex:rescue (e.g., /codex:rescue investigate why the tests started failing). This executes the query through the codex:codex-rescue subagent in the background, minimizing prompt blocking.
If you need to switch tools mid-session, run /codex:transfer to export your active Claude Code transaction history into a JSONL session file under ~/.claude/projects. You can then pick up exactly where you left off by executing codex resume <session-id> inside the Codex TUI or desktop application.
Try it in 2 minutes
/plugin marketplace add openai/codex-plugin-cc
/codex:setupbash
✓ When to use
- To offload long-running debugging or test-repair tasks to background threads.
- When you want Claude Code to pause for a review gate before concluding a task.
✕ When NOT to use
- If you do not have an active OpenAI API key or ChatGPT subscription.
- In untrusted repositories where local .codex/config.toml files could automatically load malicious settings.
What to do today
- Install the plugin by running '/plugin marketplace add openai/codex-plugin-cc' inside Claude Code
- Run '/codex:setup' to configure local authentication and set up the review gate
- Create a project-level '.codex/config.toml' file to set your default reasoning model and effort level
Sources