A second opinion from OpenAI Codex, right inside Claude. Free & open source.
Claude writes the task, OpenAI Codex (default model gpt-5.6-sol) reviews your code as an independent agent — it reads the files itself in a read-only sandbox — and Claude presents structured findings with severity, line numbers and fix suggestions.
Requirement: local shell access. Codex runs on your computer, so Claude needs a tool that can execute terminal commands there:
- Claude Code (terminal, desktop app, IDE) — built in, nothing to add
- Claude Desktop app — add a local shell connector such as Desktop Commander (
npx @wonderwhy-er/desktop-commander@latest setup, then restart Claude Desktop)It does not work in claude.ai in the browser alone: code there runs in a cloud sandbox that cannot reach your machine, your Codex CLI or your ChatGPT login.
| ChatGPT account | OpenAI API key | |
|---|---|---|
| Who | ChatGPT Plus, Pro, Business … | anyone with an OpenAI API account |
| Cost | included in your plan's Codex quota | pay per token (typically cents per review) |
| Setup | codex login (browser) |
key stored locally in ~/.codex-codecheck/config.json |
The plugin always tries your ChatGPT account first. Your API key is used only when there is no ChatGPT login or the plan's limit is reached — and every result tells you which route was used. Configure both and you never get stuck.
| Command | Description |
|---|---|
/codex:setup |
Install check, link the runtime, choose ChatGPT account and/or API key |
/codex:review <file> |
General review (use "quick" for small files) |
/codex:multi <file1> <file2> … |
Cross-file review |
/codex:security <file> |
Security audit with CWE references |
/codex:optimize <file> |
Performance analysis with impact rating |
/codex:consult <question> [files] |
Let Codex cross-check a plan or decision |
- Install the plugin
- Run
/codex:setup - Run
/codex:review path/to/file.php
- Local shell access for Claude: Claude Code, or Claude Desktop with a local shell connector such as Desktop Commander
- Codex CLI (
npm install -g @openai/codex, tested with 0.155) - A ChatGPT account with Codex access or an OpenAI API key
- Python 3 and bash (macOS, Linux, WSL)
| Mode | Reasoning | Typical time |
|---|---|---|
| mini (quick review, < 300 lines) | low | ~15 s |
| review, consult | medium | 1–3 min |
| multi, security, optimize | high | 2–4 min |
Change the model with "model" in ~/.codex-codecheck/config.json or per run with
CODEX_CODECHECK_MODEL.
- Your code is processed by OpenAI, under your ChatGPT or API account's terms
- Codex runs in a read-only sandbox and never modifies files
- API keys stay in a local file (mode 600) and are passed only to the single Codex process
- The plugin itself stores nothing on any server — no telemetry, no tracking
MIT — see LICENSE