Yes, ChatGPT's Codex Connector Does Code Review on GitHub
Research note (via Felo AI) I’m keeping because the result surprised me: the connector works better than I expected for code review.
The question was simple: does the ChatGPT Codex Connector actually work for code review on GitHub? The answer is yes — and it’s specifically designed for that, according to OpenAI’s official documentation.
The three modes
On-demand review: you mention @codex review in a PR comment; Codex reacts (👀) and posts a review focused on high-priority issues (P0 and P1).
Automatic review: turning on “Auto reviews” in the settings reviews every new PR without needing a mention.
Customization: follows review guidelines defined in a repo’s AGENTS.md file — team rules and standards live versioned alongside the code.
That last point is what interests me most: the AGENTS.md pattern shows up again as the de facto standard for configuring agents (the same one I use with CLAUDE.md in Claude Code, and the one Denny Huang recommended at GDG to keep the problem from “drifting”).
Alternatives if you don’t want the connector
- GitHub Actions + OpenAI API: a custom workflow that reviews each PR’s code and posts comments — more control, more maintenance
- Deep Research Connector: ChatGPT’s GitHub connector for deep research, oriented toward asking questions about the codebase rather than reviewing PRs
My takeaway
For a team already living in the OpenAI ecosystem, the official connector is the lowest-friction path: manual via @codex review, automatic for every PR, and customizable via AGENTS.md. For teams on Claude, the equivalent is Claude Code’s code review — what matters is that AI-assisted review with rules versioned in the repo is already standard practice, not an experiment.