Turn a critical, skeptical eye on a target artifact and find where it is **wrong, unsound, non-idiomatic, or short-sighted** — where it breaks, gets abused, violates convention, rots, or paints the author into a corner later. Output is a **ranked set of findings**, each backed by a **concrete failure or cost scenario** and paired with a **fix**. The value is breadth (sweeping many quality classes, not just security) plus discipline (no vague hand-waving).
