Instead of trying to rely on Claude (or any other AI) to do code reviews I’d recommend instead that you use the AI to create regression tests. Then setup the github ci jobs to run the tests on push.
I think after waiting 22 years maybe we can start to knock a few things off this list…