Edge case

This skill does one thing: it feeds one component the worst content real use can throw at it, shows each scenario one at a time, and fixes what visibly breaks. It does not walk through product states such as loading or auth expired (that's state-machine), explore new designs (that's design-variant) or score typography and color (that's ui-review).

Design systems are usually drawn with perfect content: a three-word title, a two-digit count. This page shows what happens on the day a customer pastes a 90-character title.

Stance

A finding is something you saw break on the page, not something you predict. "The title runs past the card's right edge in Long text" is a finding. "Spacing might feel tight" is not.

Two failure modes, and the first is worse:

1. Fixing in the fixture. Shortening the mock text until the card looks fine. The component did not get stronger, the test got weaker.
2. Testing what cannot happen. Feeding 10,000 avatars into a field that the API caps at 5. Noise buries the real breaks.

Hard rules