What problem does it solve? Flutter golden tests only prove a screen has not changed since its snapshot was taken; they cannot prove the screen ever matched the design. This Skill gates built Flutter screens against an independent design reference set (27 screens rendered from HTML/CSS in light/dark and LTR/RTL), catching design drift that goldens structurally cannot see. ## Core Features & Use Cases - Mechanical parity gates: Verifies the screen is in the requested theme, that every surface over 0.5% is within Δ24 of a Calm design token, and that at least 75% of the reference's horizontal band edges appear in the app within 4px. - Four-configuration coverage: Each referenced screen is checked in light/dark and LTR/RTL, with real Persian RTL references using Jalali dates and Eastern Arabic numerals. - CI gate script: scripts/check_parity.sh walks captured screenshots, runs the comparison tool per screen, and fails the build on wrong theme, untokenised colours, or broken vertical rhythm. - Use Case: Before opening a PR that changes any referenced UI screen, run the parity tests, execute the check script, and open the generated side-by-side sheet to confirm the app matches the design rather than merely matching last week's build. ## Quick Start Run the parity tests with flutter test test/parity/, then execute bash scripts/check_parity.sh and review the side-by-side sheets it writes to design/reference/_parity/.
