OpenAI Codex engineer Jason Liu solicits feedback on using Codex for pull request reviews
Engineer Vaibhav Srivastav proposed letting Codex automatically approve PRs
Positive users praise Codex for PR code reviews as better than Cursor BugBot after heavy use, while negative users call it weak for lacking codebase context and repeatedly flagging the wrong issues.
No Digg Deeper questions have been answered for this story yet.
Most Activity
@jxnlco Codex should approve PRs once it’s happy!
Do you use Codex for PR Review?
What could we improve on the code review experience in the application?
@jxnlco Yes, I use it extensively - around 600 times in past few months. I find Codex review better than Cursor BugBot. How does "exhaustive code review" works exactly?
@jxnlco yes but it has no idea what 'good' looks like for our codebase. flags the wrong stuff every time
@jxnlco well the hosted version codex in github catches all kinds of stuff while the local version is happy as a clam. maybe put the same prompt in local codex?
@jxnlco We used lots of .feature files for writing our tests as we use Cucumber. For Java it gives lots of feedback but for feature files there are very few comments.
@jxnlco Often I open pr review in a session and the branch displayed is for a different codex session. sometimes unrelated to a session where I am interacting with multiple prs. I'd love to see only branches/prs that a codex session has interacted with.
@jxnlco ho... good to remember, never used. and I run 3 to 4 codex session all time pushing things to gitlab, idk if it work with it, need to test.
@jxnlco /autoreview runs the full autoreview skill by @steipete
@jxnlco a more 'baked in' way to monitor for thumbs up from codex-review on the PR or comments. right now to do it i just have to set an automation that runs repeatedly and checks PRs often for it.
@jxnlco codex is pretty weak at reviews, might right up a proper skill for this actually.
@jxnlco An inbuilt code review loop, perhaps adversarial.
PR comment based workflows feel too limited and short sighted perhaps, trying to abstract agents over a human workflow.
@jxnlco https://github.com/openai/codex/issues/26867#issuecomment-4866415707
please
@jxnlco more focus on code quality. currently it feels like that mostly correctness is being reviewed.
@jxnlco I just say to codex, spin up sub agents to review PR, merge when clear.
I basically want fully automated PR reviews nearly as standard and I don’t want have to look at github.
@jxnlco For some reason tagging codex in GitHub for a review always finds issues the review in the app.didnt catch. Ex in the app codex can say it didn't find any issues to report but when using the pr review it can find like 3 P2s, and even do so multiple times.
@jxnlco threads nesting and some help making them work together? I have tried different ways to orchestrate but sometimes weird things happen like orchestrator stops reviewing or things like that.
@jxnlco Mostly i dont to pr, just push to main. But i like the github app way to handle prs
@jxnlco Works well tbh but the isolation kills me, it has no idea what the last 3 PRs did and that's exactly where things break in our codebase.
@jxnlco Please first fix this: https://github.com/openai/codex/issues/29353
I have verified it on my end
@jxnlco More native implementation. Integrate Codex Review feedback in the desktop app.
Go from thumbs up to in-app sidebar feedback with one-click actions (merge, solve-or-comments, etc).
Conductor does a decent job, but it’s not native/integrated. Codex should be so much better.