.ai/code-review.md
Work through these layers in order when reviewing any change:
Prefer the smallest change that correctly solves the problem. When reviewing, flag unnecessary scope creep — changes to files or layers not required for the fix add risk without benefit. If a change touches more than what the task requires, question whether each additional change is genuinely load-bearing.
When a bug resists 1–2 fix attempts, stop speculating through code alone. Ask the author or user to verify runtime state directly: console output, DevTools element inspection, computed styles, network requests. Formulate precise questions ("what classes are on this element?", "what does this variable log as?") — one targeted observation resolves what many iterations of code guessing cannot.