What problem does it solve?
UI polish work often relies on subjective, ungrounded feedback. This Skill replaces guesswork with reproducible screenshots reviewed by a panel of design-expert agents and user-persona agents that return calibrated numeric scores, so redesign iterations converge on a measurable quality bar.
Core Features & Use Cases
- Multi-agent expert panel: Runs one agent per craft dimension (visual hierarchy, design-system consistency, color/contrast, typography, spacing, interaction flow) in parallel, each returning a 1-10 score with severity-tagged, evidence-backed issues.
- Optional persona panel: Simulates users with different cognitive styles (ADHD, power user, low-vision, minimalist, novice) returning verdicts, blockers, frictions, and delights.
- Iterate-to-target loop: Re-screenshots and re-rates after each fix round until both panel averages clear a configurable target (default 8/10), then hardens to PR-ready with tests, l10n, and changelog fragments.
- Use Case: When redesigning a modal or screen in the Lotti Flutter app, capture baseline screenshots, summon the panel, and iterate on the ranked must-fix list until the workflow returns cleared=true.
Quick Start
Ask the assistant to run the design review panel on a specific UI surface, for example: summon the design panel to rate the AI popup model picker and iterate until both panels average at least 8 out of 10.
