That tweet shown is most definitely not an innocent looking PR by any stretch. It would be impressive if it could be shown working on more subtle attempts, for example the xz one with the `.` that was left indented.
Agree, this is a rather trivial use case. Let me know if you have any specific PR that you'd like me to test this on, or feel free to test it yourself https://github.com/presubmit/ai-reviewer