Ideally, this tool should have an overall try/except so that if a failure is caused by the tool being broken then it won't block all the pending PRs. The tool failing is a different and less interesting case than when it finds the issue it was designed to detect.

Also, if a tool has a risk of false positives, it should be attached to one of the buildbots that isn't one of the few required to make a commit. It should simply report failures rather than bringing all development to a halt.