/review-bugbot on a large local branch reports “Bugbot found no bugs” even when the review is incomplete and the subagent actually found a bug.
The diff embedded in the bugbot subagent prompt is silently truncated. The prompt was about 244k characters, contained only 34 of 71 changed files, and still closed with with no truncation notice. The diff looks complete.
The subagent’s final message contained a valid …. The Task tool result returned to the parent was exactly “Bugbot found no bugs.” The parent then told the user there were no findings.
Expected:
Truncated diffs are marked as truncated, or the full diff is provided.
If the subagent emits entries, those findings are returned to the parent. An empty is the only case that should become “no bugs.”
Both of the things you noticed are happening. On very large branches, only part of the diff fits into a single Bugbot review, and right now that isn’t called out. Bugbot also double-checks each candidate finding before reporting it, and when every candidate gets set aside, the summary reads the same as a clean review. We’ve let the team know about both.
In the meantime, splitting the review gives much better coverage on a branch this size:
Ask the agent to run /review-bugbot against an intermediate commit on your branch. That covers the newer work.
Then check out that intermediate commit and run /review-bugbot against your base branch. That covers the older work.
If a lot of the change is still uncommitted, you can review just the uncommitted changes first, then stash them before reviewing the committed work.
When a branch is too large for a single review, Bugbot now says the review is partial, lists the files it left out, and suggests splitting a follow-up review across them. Its double-check step is also told which files were left out, so it no longer sets findings aside just because their file wasn’t in the reviewed part.
When Bugbot sets aside every candidate finding during that double-check, the result no longer looks like a clean review. It now says how many findings were reported and set aside.
Both changes were made on our side, so no Cursor update is needed. Could you give /review-bugbot another try on a large branch and let me know if anything still looks off?