I’m considering DeepSeek as a first-pass reviewer for pull requests. Formatting and obvious mistakes are easy to spot, but I care more about whether a small change could alter edge-case behavior or break an assumption elsewhere.

How has DeepSeek performed on patch-level code review when the surrounding code is provided? Does it catch useful risks, or mostly return comments about naming and style? I’d like to hear how people check its findings before acting on them.