Ux Foundations

UX Foundations

This skill defines the structure and design language of the UI before any feature is built — the counterpart to a software-architecture document, but for the experience layer. It does not produce pixel-perfect screen designs; it produces the system those screens are later assembled from.

The governing idea, especially when an app has several interfaces: one shared core, defined once, plus a per-surface layer for each interface. A single design language (brand, tokens, accessibility, voice) spans every surface so the product feels like one thing; each surface (admin portal, website, mobile app) then gets its own profile because its users, navigation, components, and interaction context genuinely differ. Not one monolith that flattens the differences, and not independent design systems that drift apart.

Input

This skill consumes the architecture document produced upstream.

• If the user provides a path, use it.
• Otherwise default to docs/architecture.md.
• If no architecture document is found, tell the user, and offer to either point

to the right path or proceed in a reduced mode where you elicit the surfaces and users directly (the output will be weaker without the architecture's context, so prefer locating the document).

From the architecture, extract: the UI surfaces (they appear as building blocks / containers — e.g., "Admin Portal", "Web App", "Mobile App"), the user types / actors (from the context diagram), UX-relevant quality attributes (accessibility, latency/performance that shape perceived experience), constraints (existing brand, mandated frontend stack), and the domain and features that the UI must express.

Workflow

Phase 1 — Ingest and frame

1. Read the architecture document (per Input above) and pull out the surface

list, actors, constraints, and domain.

1. Play back the surfaces you found and confirm the set ("I see three UI

surfaces: an admin portal, a public website, and a mobile app — is that the full list?"). Surfaces are the spine of everything that follows.

Phase 2 — Elicit the UX specifics

Read references/elicitation-guide.md and run a short interview for what the architecture doesn't cover. Same philosophy as the architecture skill: ask in batches, infer aggressively from the architecture and state your inferences, offer defaults the user can accept in a word, and let them dump a brief if they have one. Cover the shared-core questions once (brand, voice, accessibility bar, visual direction, existing design system) and the per-surface questions for each surface (who uses it, their primary jobs, device/responsive targets).

Phase 3 — Establish the shared core (once)

Read references/design-system-guide.md and define the parts that span every surface: brand and voice, design tokens (color, typography scale, spacing, radius, iconography), the accessibility standard, and the cross-surface interaction principles. This is the single source of truth that makes the surfaces feel like one product.

Phase 4 — Profile each surface (loop)

Read references/surface-profile-guide.md. For each surface, produce a profile: its users and primary jobs, its information architecture and navigation (a sitemap), its key user flows, its screen/page inventory, and the surface-specific components and token overrides it needs on top of the shared core. Scale the depth to how much the surface actually diverges — two web surfaces may share a component library and differ mostly in IA, while a native mobile app warrants its own component layer and platform-convention notes.

Phase 5 — Reconcile across surfaces

Resolve the cross-cutting questions: which components are truly shared (live in the core) versus surface-specific, which flows span surfaces (e.g., act on the web, get notified on mobile), and the consistency rules that keep the surfaces coherent without forcing a desktop pattern onto mobile.

Phase 6 — Document and diagram

Read references/document-template.md and write the UX-foundations document. Read references/diagram-guide.md and embed the sitemaps and key user flows as Mermaid, so the document is one portable artifact. Right-size it: a single-surface tool gets a tight doc; a multi-surface product gets the shared core plus a section per surface.

Phase 7 — Deliver

Save to docs/ux-foundations.md by default (or where the user wants). Summarize the design direction and the surface set in a few sentences, and point out that the screen inventory is the handoff to implementation planning — it's what lets planning slice features vertically across backend and UI. Offer concrete next steps: emit a real design-tokens file (CSS variables / JSON), produce one or two anchor screen mockups to lock the visual direction, or proceed to implementation planning.

Output format

Default to a single Markdown file (docs/ux-foundations.md) with embedded Mermaid sitemaps and flow diagrams, and tokens/component inventories as structured tables. Optional, on request:

• A design-tokens file — tokens.css (CSS custom properties) or

tokens.json, as a concrete artifact the frontend build can consume.

• Anchor mockups — one or two key screens rendered to settle visual

direction, when the product is design-led enough to justify it early.

Scope boundaries

This skill defines the system and the structure. It deliberately does not:

• produce pixel-perfect, per-feature screen designs — those happen per slice in

the detailed-design step, assembled from this system;

• generate frontend component code — that's implementation.

Staying at the system/spec level is what lets the shared core span platforms: the same tokens compile to web and native even though components are built differently on each.

What good looks like

• One coherent shared core, with per-surface layers that diverge only where the

context truly demands it.

• Every design choice traces back to a user, a job, or a constraint — never

decoration for its own sake.

• A complete, per-surface screen inventory that implementation planning can

slice against.

• Diagrams render on the first try (see the diagram guide's reliability notes).
• Right-sized: structural and decision-dense, not a 60-page style bible.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: zeeshanhanif
• Source: zeeshanhanif/agent-skills
• License: MIT

Install and usage instructions live in the source repository linked above.