I think that's perfectly fine to upstream, provided there's understanding and you're communicating as yourself. Having Claude generate all the communication for you without your input makes it really, really hard to tell who you are. For some smaller packages, the review may be small and "LGTM". For others, like Django, there's going to be a few rounds of back and forth, and for that a reviewer shouldn't have to communicate with a bot without their consent. Does that make things a bit clearer?