Bugbot appears to resolve prior inline findings after an unrelated commit

Support Bug Reports
,
1

Where does the bug appear (feature/product)?

BugBot

Describe the Bug

We recently started using Cursor’s Bugbot for PR reviews, and we also enabled automatic review on every new commit.

We observed behaviour that seems wrong, or at least surprising.

I understand that AI reviews can be somewhat non-deterministic, so I would not expect every rerun to produce identical findings. That part is understandable.

What seems problematic is the state behaviour, if Bugbot previously raised an inline issue, and a later commit does not fix that issue or touch the relevant code, I would expect the original finding to remain open. A fresh review pass might choose not to raise the same issue again, but it should not automatically resolve a prior open finding unless the issue was actually addressed or the comment became outdated due to changes in that exact area.

So the concern is less “the model changed its mind” and more “previous review state does not seem to persist correctly across commits.”

Is this expected behaviour, or could this be a bug in how Bugbot tracks findings between review runs?

Steps to Reproduce

  • A PR was opened and Bugbot reviewed it.
  • Bugbot left an inline comment on a code block.
  • Another commit was pushed to the same PR, but it only changed other parts of the codebase and did not address the issue Bugbot had commented on.
  • Bugbot ran again on the new commit.
  • After that rerun, the earlier Bugbot comment was marked resolved / cleared, even though the underlying issue had not been addressed.

Operating System

MacOS

Version Information

it’s bugbot

Does this stop you from using Cursor

No - Cursor works, but with this issue

5

Hi @KemalY

This is, to some extent, a known limitation of Bugbot’s handling of previously raised findings.

When a new commit is pushed, Bugbot computes a diff between the code at the time a finding was raised and the current HEAD, then passes that to a model to determine whether the issue has been addressed.

The problem is that the model can produce false positives and conclude a finding was resolved even when the relevant code wasn’t actually changed between commits. There’s no easy fix today, but we’re aware this is a gap and want to improve the accuracy of the resolution check going forward.

Does this happen to every finding Bugbot raises (all get auto-resolved on the next push), or just some of them?

7

This topic was automatically closed 22 days after the last reply. New replies are no longer allowed.