So - I've done this just by reading LLM code. I mean, yes I'm largely looking for high level architectural smells and bad practices, but I feel like "reading the code" now takes on a lot of the previous roles that code reviews may have done.

I just don't do this at commit time. I might make 100 commits and then dig through code and find some bad architecture stuff.