BugBot autofixed some issues it identified after being prompted to provide a review, despite the “Bugbot autofix” setting being disabled in the dashboard.
This interaction was via the Github UI, not the desktop app.
Additional Information
The “disable autofix” checkbox in the dashboard should act as a hard gate as it incurs a financial cost and could potentially violate SDLC requirements.
Bugbot is fast, but it is typically not “fixes issues in the same minute a review is asked for” fast. Is it possible that this came from another message you sent to a cloud agent asking for a review, or something else you asked @cursor in the GitHub thread?
When you commented @cursor bugbot review please , the system parsed everything after @cursor as instructions for a Cloud Agent (not Bugbot). The Cloud Agent interpreted “bugbot review please” as general instructions and proactively took action on the PR, which is its default behavior.
To trigger a Bugbot review specifically, use one of:
cursor review or @cursor review
bugbot run or @bugbot run
What do you think would make this more intuitive? Maybe @cursor bugbot should always route to Bugbot?
This behavior makes sense now that I understand the flow. But if the question is what’s intuitive to me, I think routing @cursor bugbot to the same handler as @bugbot feels right.
Maybe that change is even just a straightforward tweak to your system prompt?
Thanks for explaining! Please consider this issue resolved.
