Ui ux design
Skill muxammadmamajonov/dot-claude/.claude/skills/ui-ux-design
Use when designing user-facing surfaces — information architecture, user/data flows, screens or command surfaces, design-system baseline. Triggers — new web page, mobile screen, CLI command.From its SKILL.md
npx -y skills add muxammadmamajonov/dot-claude --skill ui-ux-designAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
One thing to look at
- 0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
SKILL.md
5.6 KB, ~1.2k tokens by cl100k_base, as published. Nobody here has run it
UI / UX & Interaction Design
When to use
- A spec or feature is approved and a human-facing or machine-facing surface must be designed before build.
- You are adding a new screen, view, command, endpoint shape, notification, or device interaction.
- Usability, accessibility, or consistency problems are reported and a redesign is needed.
- The project lacks a shared design baseline (tokens, components, interaction patterns) and drift is appearing.
"UI" here is broad: GUI screens, CLI/TUI command surfaces, conversational/agent turns, API request/response ergonomics, dashboards, and embedded/IoT displays all have an interaction layer worth designing.
Workflow
- Restate the job-to-be-done. From the spec, write one sentence per primary user (or consuming system) describing the outcome they need. Reference the founder interview and
.claude/templates/product-spec.md. If a primary user is unclear, ask the user — do not invent personas. - Map information architecture. List the top-level entities, their relationships, and how a user navigates between them. For non-GUI projects this is the command/resource hierarchy or the object model an API exposes. Produce a tree or sitemap.
- Design the critical flows. For each top-3 task, write the happy path as numbered steps, then enumerate the empty, loading, error, partial-permission, and offline states. A flow is not done until every state is named.
- Sketch screens / surfaces low-fidelity first. Lay out each surface as labeled regions (header, primary action, content, secondary). Decide the single primary action per surface. Keep it text/wireframe level — do not pick colors or pixel values yet.
- Establish the design-system baseline. Define tokens once: spacing scale, type scale, color roles (not raw hex scattered everywhere), radius, elevation, motion durations. Define core components (button, input, list, card, dialog, toast, empty-state) with their states. Record this in
.claude/templates/architecture.mdor a dedicated design-system doc. - Specify interaction & accessibility rules. Keyboard order, focus management, hit-target sizes, contrast ratios, error-recovery, copy tone. For CLI: flag conventions,
--helpoutput, exit codes, machine-readable output mode. For APIs: consistent naming, pagination, error envelope. - Validate cheaply. Walk each flow against the spec's acceptance criteria. Run a heuristic pass (see Standards). If feasible, do a 3–5 person hallway test or a quick task walkthrough before committing to build.
- Hand off. Produce the artifact described in Output format, link it from the spec, and flag the build phases that implement it.
Standards
- Do design the unhappy paths (empty, error, loading, no-permission, slow-network) for every flow — they are the bulk of real usage.
- Do define one primary action per surface; demote everything else visually or structurally.
- Do centralize design decisions as tokens/components so changes propagate; never hardcode the same spacing/color in twenty places.
- Do meet WCAG 2.2 AA as the default bar: 4.5:1 text contrast, 24px min target spacing, visible focus, no information by color alone. See
.claude/checklists/accessibility.md. - Do write real interface copy (labels, errors, empty states) — copy is design, not a fill-in-later task.
- Do design responsively/adaptively: small + large viewport for GUI, narrow + wide terminal for CLI, varying payloads for APIs.
- Do-not start visual polish (color, shadow, animation) before IA and flows are validated.
- Do-not introduce a new component when an existing one fits; extend the system instead.
- Do-not rely on hover-only or color-only affordances, or ship modals that trap focus or lose keyboard users.
- Do-not assume a web app — keep patterns translatable to whatever surface the project actually uses.
Common mistakes to avoid
- Designing only the success state and discovering error/empty/loading gaps during build.
- "Mystery meat" navigation: unlabeled icons, ambiguous primary actions, hidden destructive actions with no confirmation.
- Inventing personas or requirements not in the spec instead of asking the user one business-critical question.
- Token sprawl: a "spacing system" that has 14 ad-hoc values, or colors copied as hex everywhere.
- Inaccessible defaults: low contrast, tiny tap targets, focus that disappears, forms without labels.
- Over-designing low-traffic surfaces while under-specifying the daily-driver flow.
- Treating CLI/API/agent surfaces as "not UX" — they have the same need for clarity, consistency, and good error messages.
Output format
A design package (one document or folder) containing: (1) IA tree/sitemap, (2) the top flows with all states enumerated, (3) low-fidelity wireframes/region maps per surface, (4) the design-system baseline (tokens + core components with states), (5) interaction & accessibility rules, (6) real interface copy. Use .claude/templates/architecture.md for the system-level structure and link the package from the feature spec (.claude/templates/product-spec.md). Keep it diff-friendly (markdown + simple diagrams) so it lives in the repo and stays current.
Related checklists
.claude/checklists/accessibility.md.claude/checklists/qa.md.claude/checklists/security.md
Related agents
.claude/agents/core/orchestrator.md.claude/agents/design/ui-ux-designer.md.claude/agents/core/business-analyst.md
What ships with it
Read from the repository
Just SKILL.md. No reference files, no scripts.