I spent some time looking into the rsync regressions. I found the general shape of the cause of one of them, and a discord buddy was able to further nail the exact chain of failure. We reported our findings in the appropriate github issue. (It was a classic: security fix introducing a new bug. New tests for the fix didn't expose it.)

I took the opportunity to politely push back that there has been a sudden flurry of very large Claude-cosigned commits, and that seeing things like the entire test suite being rewritten from bash to python by an AI in a single commit sets off a lot of alarm bells in my head for ugly surprises. Too much, too fast.