How to enable Codex automatic code review for PRs

TL;DR

Enable Codex code review in Codex web settings, then any PR submitted to the repo is automatically reviewed by codeex. The system can access the whole repository, run tests, and check logs, not just the diff, and you can tailor reviews with custom instructions via AGENTS.md or PR comments. Local CLI reviews are also supported to catch issues before submission.

Transcript

Code review of two comment mark. Hey everyone, I'm Roma and I'm Maya. Codex needs to do two things really well to be an effective coding teammate. First, it needs to work with all your tools and second, it also needs to plug into all of your team workflows. Code review is one of the most important workflows for any engineering team and we want to h... Read More

Key Insights

• Codex code review reacts to PRs automatically once enabled in the Codex web settings. The system is designed to work with openai tools and integrate into existing team workflows.
• Codeex can review not only diffs but the entire repository, including dependencies and broader codebase contexts, enabling more accurate bug detection.
• Users can instruct the model with specific focus areas by commenting on the PR or via AGENTS.md to tailor the review behavior.
• The training emphasizes high precision and real-world usefulness, prioritizing bug detection that matters to real projects without excessive noise.
• Codeex supports logging of its review process so teams can see the reasoning and data behind findings, improving transparency and trust.
• OpenAI has deployed codex code review internally and externally to reduce critical issues during model releases and complex configurations.
• A CLI integration allows developers to run reviews locally before pushing changes, ensuring quality gates before PRs are created.
• The system can form hypotheses, generate test scripts in Python, and validate its conclusions against the codebase and tests.

Questions & Answers

Q: How does Codex code review get enabled in practice?

Codex code review is enabled by going into the Codex web settings and turning on the feature. Once enabled, any pull request submitted to the repository will be automatically reviewed by codeex. This setup makes it possible to start receiving automated feedback on code changes immediately, aligning with OpenAI's iterative deployment philosophy.

Q: What does Codex code review look at besides the diff?

Beyond the diff, Codex code review has access to the entire repository, can track down dependencies, and understands the broader codebase. This means it can reason about context, verify whether a change affects related modules, and assess how a fix might ripple through the system, leading to more robust conclusions.

Q: Can you instruct Codex to focus on specific areas in a PR?

Yes, you can add instructions for Codex to focus on particular areas by commenting on the PR with guidance for the reviewer or by using custom guidelines stored in AGENTS.md. This allows teams to steer the review toward critical hotspots, such as security, performance, or API stability, without overwhelming developers with irrelevant notes.

Q: How are reviews presented when Codex completes a pass on a PR?

When Codex completes a review, its findings are surfaced as comments and suggested changes within the PR thread. The system can also be prompted to re-run reviews after fixes, enabling an iterative cycle where human reviewers and the model collaborate to validate fixes and ensure the changes meet project standards.

Q: What role do tests and logs play in Codex reviews?

Tests and logs are integral to Codex reviews. The model can write and execute Python code to test hypotheses about potential bugs, examine test results, and show evidence from logs. This approach helps verify the validity of findings beyond textual comments and static analysis alone.

Q: How does Codex help in internal development workflows at OpenAI?

Codex code review has been used internally to catch issues before they become critical, such as preventing training run bugs or misconfigurations from delaying releases. By reviewing changes across the codebase, it enables contributors to work confidently on unfamiliar parts of the project.

Q: What is the benefit of reviewing code locally with the CLI?

The CLI review feature allows developers to examine changes locally before pushing to GitHub, serving as a local gatekeeper. This reduces the risk of introducing bugs into a PR, complements editor-based reviews, and broadens the set of scenarios the model can test by running commands directly in the terminal.

Q: How does AGENTS.md influence Codex code review behavior?

AGENTS.md provides a standard format for coding agents and lets teams embed custom instructions for Codex code review. By including agent-specific guidance, developers ensure the model adheres to project-specific rules, style preferences, and particular review targets, improving alignment with team expectations and reducing unnecessary comments.

Summary & Key Takeaways

• Codex code review can be enabled in the web settings to automate PR reviews, integrating with GitHub workflows and broader repository context.

• The review process leverages the model's access to dependencies, logs, and test results, providing high-signal feedback beyond static analysis.

• Custom guidance and local CLI support allow teams to steer reviews, set focus areas, and perform pre-submit verifications for safer, faster ship cycles.