@simpsoka I think Codex should have a native skill like the Cursor Babysit skill where you feed it a PR URL and it will monitor it and address comments, etc until it’s ready to merge

@simpsoka I don't see a fix all button after code review is completed lol

@simpsoka @jxnlco It already helps me and my little 16 person eng team do both! Very happy with Codex + GPT 5.5 as a code reviewer.

It catches soooo many little nuances that we would definitely miss!

@simpsoka 1. Run a dedicated security review on every PR

2. for big PRs it should summarise discussion and list unresolved concerns

3. Have it assign risk probability

@simpsoka If the drop down model selector had 5.6 available I think it would greatly help with code review and managing PRs

@simpsoka Hire @mattpocockuk as a consultant... Maybe counteract the nonsense coming from Peter 😂

The current limitation with code review is that it often reports only one issue per review, at least I haven’t seen it report more than one. I often need to start several separate review sessions to catch everything, even after it previously says “no actionable items found.”

Would be great if it could surface multiple actionable issues in one pass.

@simpsoka everything is loop

I created a skill that has codex security review on most pushes - would be cool if the codex team was able to iterate on that type of workflow vs manually having to call the skill every time something needs to be reviewed - automations work as well but if feels like there is a more robust solution we can put in place

@simpsoka 0% anybody does that?!

@simpsoka Best review use case is making Codex argue with the diff before a human does. The win is not fewer comments. It is finding the weird one earlier.

@simpsoka auto-triage stacked PRs by risk + blast radius so I know which one to actually open first. the queue is the bottleneck, not the review itself

@simpsoka Maybe by increasing rate limits.

@simpsoka The biggest pain point for our team is openai-GitHub code reviewer and codex going into code review loops with 10-30 turns.

@rgagnier06 @simpsoka Just add a sentence to AGENTS and it will do that

@simpsoka Walking through the changes step by step

@simpsoka for me it’s most useful when it turns review comments into concrete patches, not just summaries. the boring win is fewer stale PR threads.

@rgagnier06 @simpsoka I was planning to write a skill to do that.

@simpsoka Would be great to have dictation added to review comments

@simpsoka I have this skill that I actually made to review editors notes for a paper, but I think it's a nice basis for serially processing code reviews too. But it needs more integration to really fly. https://gist.github.com/robbiemu/74581d01f145de1d5198dc9e389c7fb1