What problem does it solve? Flutter widget tests silently pass while shipping broken layouts: the default 800x600 test surface is wider than any phone, overflow is reported only once per RenderObject, clipped text never reports an error, and the built-in contrast guideline false-passes white-on-near-white text. This Skill replaces those traps with a pinned-device harness, an overflow matrix, computed-geometry fit assertions, and pure-Dart WCAG/APCA contrast checks. ## Core Features & Use Cases - Pinned test harness: A single pumpApp harness that sets tester.view.physicalSize x devicePixelRatio and layers MediaQuery (textScaler, boldText, accessibleNavigation) above MaterialApp via copyWith. - Overflow and fit matrix: One testWidgets per (device, text scale, bold) tuple with getSize/getRect fit assertions that catch silently clipped text, plus a ban on takeException suppression, FittedBox, and text-scale clamping. - Honest accessibility gate: isSemantics node assertions, explicit getSize tap-target loops, traversal-order checks, and pure-Dart WCAG/APCA contrast tests on theme colour values instead of the known-broken textContrastGuideline. - Two golden lanes: Ahem geometry goldens plus one pinned-OS real-font lane with loadAppFonts, RTL goldens under Directionality, and CI blocking of accidental --update-goldens. - Use Case: A Flutter developer chasing an "overflowed by 47 pixels" failure at 200% text scale uses the matrix and fit assertion to find the silently clipped label, fixes the layout instead of clamping the text, and verifies the fix with scripts/check-test-hygiene.sh before the PR. ## Quick Start Ask the AI to write a Flutter widget test using the pumpApp harness that pins a 360dp device and asserts no overflow at 2.0x text scale with bold text enabled.