Design — Cross-Platform UI Skill Dispatcher

Overview

The cross-cutting design family is 16+ reference skills. Loading them all drowns the context; loading none means eyeballing it. Transcript evidence says they work in predictable clusters — this router loads the cluster for the task.

Resolve the target first: the screen/component/task in the argument, else the UI work active in the conversation. State it in one line, load the routed skills via the Skill tool, and apply them while working — router, not recital. Native iOS/SwiftUI → use /ios, not this.

Routing Table

| Task | Skills (in priority order) | |---|---| | New screen / component build (default row) | components-and-states, layout-and-composition, make-interfaces-feel-better | | Layout redesign, reflow, spacing pass | layout-and-composition, grid-and-spacing, color-and-elevation | | Table, chart, dashboard, metrics | data-visualization, components-and-states, accessibility-and-inclusive-design | | Dense single-viewport UI ("fit on one screen", admin tool, inspector) | dense-no-scroll-layout, layout-and-composition, grid-and-spacing | | Tufte-grade information display (data density, small multiples) | tufte, data-visualization | | Design audit / QA before ship | accessibility-and-inclusive-design, components-and-states, color-and-elevation (+ data-visualization if charts/tables present) | | Design case study / portfolio narrative | case-study dispatcher; it routes evidence, story, decisions, writing, and AI-agent supervision | | Polish / "make it feel better" | make-interfaces-feel-better, interaction-and-motion, app-motion-and-animation | | Motion / animation decisions | app-motion-and-animation, interaction-and-motion | | Design system / tokens | design-tokens, grid-and-spacing, color-and-elevation | | Color, contrast, dark mode | color-and-elevation, accessibility-and-inclusive-design | | Copy: forms, labels, errors, empty states | ux-writing-and-content, components-and-states | | Icons, avatars, imagery | iconography-and-imagery, design-tokens | | Persuasion surfaces: pricing, CTAs, consent, retention | ethical-design-and-dark-patterns, ux-writing-and-content | | "Why does this UX feel wrong" / behavioral check | laws-of-ux-and-psychology, usability-heuristics | | Greenfield page or full redesign with taste stakes | hallmark first (it owns anti-slop direction), then this table for the parts | | UX flows (auth, onboarding, checkout, ...) | userflow dispatcher (routes flow-), then this table for the screens |

Ambiguous target → the default row. Marketing/landing pages → hallmark owns the visual identity; pull structural rows from here only as needed.

Rules

• Load at most 4 per pass. Task matches several rows → primary row, plus the single most load-bearing skill from the secondary row.
• Case studies route through /case-study. For AI agents, that dispatcher must include ai-agent-case-study; do not reduce the work to a linear screen flow.
• Audits always include accessibility-and-inclusive-design, whatever the row says.
• Co-fire, don't sequence: evidence shows these skills are applied together (component anatomy while laying out, color while spacing) — load the cluster up front, not one at a time as gaps appear.
• The dec- canon is principles, this family is specs. For named-principle reasoning (Nielsen, Norman, cognitive load) route via /dec; come here for concrete values (tokens, ramps, grids, state matrices).
• Name which skills you loaded, per the always-on design-skill discipline in CLAUDE.md.