docs: improve codex review for github integration
Resolved 💬 4 comments Opened Feb 5, 2026 by Daanvdplas Closed Feb 5, 2026
💡 Likely answer: A maintainer (etraut-openai, contributor)
responded on this thread — see the highlighted reply below.
What issue are you seeing?
I have codex integrated with my github and I ask it for reviews. I notice that I always only get one comment.
I found two methods to integrate it:
- via ci: https://github.com/openai/openai-cookbook/blob/main/examples/codex/build_code_review_with_codex_sdk.md
- via comment directly in github mentioning (
@codex): https://developers.openai.com/codex/integrations/github/
Now the latter document doesn't specify at all what P0 or P1 is and therefore how to improve the reviews via the AGENTS.md.
I am also not sure whether the first link is still the recommended way of using codex in github for PR reviews.
Please tell me how openai uses codex as reviewer and how to give it direction / context.
What steps can reproduce the bug?
Search for how to integrate codex and how to improve it, no clear one way of doing it.
What is the expected behavior?
A way to improve / change the way codex does reviews.
Additional information
_No response_
4 Comments
When you use the
@codexinvocation, you can provide additional instructions like@codex review for security vulnerabilities. Is that what you're looking for?Summary
Testing
View task →
No, I am building e.g. a cli tool and I want it to do high quality in depth reviews. Those are not always security related. Now I have to ask
@codex10 times a review in my PR because it only comes with one comment per time. This is not efficient at all.We're tracking that feature request in #5547. Please upvote that issue if you'd like to make the number of code review comments each time.