People who know the domain get far more out of LLMs than people who don't. It's not about prompt skill, it's about knowing what to ask

A secret doesn't look different from any other config line in a diff — why code review structurally misses this, and what closes the gap