Code review stops after finding only a few issues, requiring repeated reviews

Resolved 💬 7 comments Opened Oct 4, 2025 by ghost Closed Jan 7, 2026

💡 Likely answer: A maintainer (etraut-openai, contributor) responded on this thread — see the highlighted reply below.

What version of Codex is running?

codex-cli 0.44.0

Which model were you using?

GPT-5-Codex

What platform is your computer?

Linux 6.6.87.2-microsoft-standard-WSL2 x86_64 unknown

What steps can reproduce the bug?

When running the review feature on Codex, the reviewer only reports 1 to 3 issues before stopping. It does not go through the whole codebase diffs to report all errors at once. As a result, a user have to repeatedly trigger the review, which wastes a lot of tokens and time. I would prefer if the reviewer could report all problems in one go, so I could fix them without multiple review cycles.

What is the expected behavior?

The reviewer should scan the entire codebase and report all errors/issues in a single review session.

What do you see instead?

The reviewer stops after finding just a few issues, requiring me to keep hitting review again and again. This leads to unnecessary token usage and delays.

Additional information

This behavior makes it inefficient to fix all problems quickly. The tool should surface all of them at once.

View original on GitHub ↗

7 Comments

Loongphy · 10 months ago

Same here, any workaround?

konwolters · 8 months ago

Same here. I can even see in the logs that it finds a lot more bugs but doesn't include them in the summary. I also always get only 1-3 issues in the summary.

Watcher3056 · 7 months ago

It found only one bug in a very huge pull-request, which is seems to behave like to stop on the first found issue.

Also "Fix All" button is disabled, so I'm not even able to proceed it to fix the issue.

etraut-openai contributor · 7 months ago

The "/review" command is designed to find and report only a few bugs in each pass, so this isn't a bug. You can run it multiple times if you'd like it to continue finding additional problems. There is an open feature request to support configurability of the number of code review issues. If you'd like to see that enhancement, please upvote.

clhereistian · 4 months ago

The "/review" command is designed to find and report only a few bugs in each pass, so this isn't a bug. You can run it multiple times if you'd like it to continue finding additional problems. There is an open feature request to support configurability of the number of code review issues. If you'd like to see that enhancement, please upvote.

Why? In what scenario is that desirable? That would be like a senior engineer only making one comment till the author fixed it and then making one more. That could string a PR along for weeks.

lostincomputer · 3 months ago

This violates the Principle of Least Astonishment. If you asked ten developers how they expect this to function, the vast majority, if not all, would assume the code review returns all findings by default.

Does anybody know if the Codex GitHub automatic pull request review has the same behavior?

mmcallister · 28 days ago

This violates the Principle of Least Astonishment. If you asked ten developers how they expect this to function, the vast majority, if not all, would assume the code review returns all findings by default. Does anybody know if the Codex GitHub automatic pull request review has the same behavior?

Yes, unfortunately it's the same behavior in GH auto PR review. This is an older comment on a closed issue but figured I'd post for anyone else who may be interested. Following in https://github.com/openai/codex/issues/5547