agentsclimarketplace

Ui design engineering

Skill ntaffzii/Skill-Agents/skills/engineering/ui-design-engineering

Design, implement, harden, and review production-grade user interfaces for web apps, dashboards, SaaS products, admin tools, forms, and responsive frontend components. Use when building or improving UI/UX, translating Figma or requirements into code, styling React/Vue/Svelte/HTML views, reviewing frontend quality, fixing accessibility or motion, refining interaction states, improving performance, or validating responsive layout and visual polish. Integrates with Skill-Agents frontend workflows and browser, Playwright, Figma, package, test, and validation MCP tools.From its SKILL.md

Install
npx -y skills add ntaffzii/Skill-Agents --skill ui-design-engineering

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

  • 4 stars4 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

9.7 KB, ~1.9k tokens by cl100k_base, as published. Nobody here has run it

UI Design Engineering

Build interfaces that emerge from the product, user, and workflow rather than from a generic template.

Choose The Mode

  • Build mode: inspect the product context, implement the interface, and validate it in a browser.
  • Review mode: report concrete violations first, explain impact briefly, and propose the smallest useful fix.

For a detailed review checklist, read references/ui-audit.md.

For attribution and the external ideas that informed this skill, read references/credits.md.

Skill-Agents Integration

When this skill is loaded through mcp-tools:

  1. Use route_request to select the workflow and related skills.
  2. Prefer the frontend-ui toolset for implementation and browser QA.
  3. Prefer the design-frontend toolset when Figma or design mapping is involved.
  4. Use build_agent_context to load only the selected workflow and skills.
  5. Use MCP tools in this order when available:
    • project, repo-index, package, and docs for discovery.
    • figma and browser-page-map for design and page context.
    • code-editing for implementation.
    • browser, playwright, and playwright-actions for visual and interaction QA.
    • test-inspection, ci, and validation for verification.
  6. Follow MCP safety policy and the project's allowed roots. Do not request broader tools than the task needs.

With Local LLM Orchestrator, use main-llm-tools while developing or auditing UI so the model receives this skill and the required MCP tools. A direct model such as main-llm or coding does not load skills automatically.

Workflow

  1. Inspect before designing

    • Read the existing frontend structure, design tokens, component library, routes, and nearby screens.
    • Reuse established primitives and patterns unless they are the source of the problem.
    • Identify the actual user, their immediate goal, the environment, and the frequency of the workflow.
  2. State the design intent

    • Name the product domain, desired density, visual tone, and interaction character.
    • Identify one signature element that belongs specifically to this product.
    • Name the obvious generic defaults to avoid.
    • For operational tools, prioritize scanning, comparison, repeated action, and predictable navigation.
  3. Design the information architecture

    • Lead with the user's primary task, not decorative content.
    • Establish navigation, hierarchy, grouping, and action priority before visual styling.
    • Keep page sections unframed; use cards only for repeated items, modals, or genuinely bounded tools.
    • Do not place cards inside cards.
  4. Implement complete behavior

    • Build real controls and wire their expected interactions.
    • Include loading, empty, error, disabled, selected, hover, focus, and success states when applicable.
    • Use icons for familiar tool actions and labels or tooltips for unfamiliar icons.
    • Preserve the project's framework, state-management, styling, and component conventions.
  5. Enforce accessibility

    • Prefer native semantic elements and established accessible primitives.
    • Give every interactive control an accessible name.
    • Make keyboard focus visible and preserve logical tab order.
    • Manage dialog focus, Escape behavior, and focus restoration.
    • Associate form labels, help text, errors, required state, and invalid state correctly.
    • Respect reduced motion and never rely on color, hover, or toast messages alone.
  6. Refine visual craft

    • Use typography, spacing, density, color, and elevation as one coherent system.
    • Match type scale to context; compact tools need compact headings.
    • Use tabular numerals for changing data and stable dimensions for controls and grids.
    • Avoid default-looking palettes, gratuitous gradients, glow effects, decorative blobs, and generic hero layouts.
    • Keep accent color deliberate and ensure the interface is not dominated by one hue family.
  7. Keep motion purposeful

    • Add animation only when it clarifies change, hierarchy, causality, or spatial continuity.
    • Prefer transform and opacity; avoid animating layout and large blur surfaces.
    • Keep interaction feedback fast and stop looping animation when it is not visible.
    • Do not let motion delay primary actions.
  8. Validate before handoff

    • Run the narrowest available lint, typecheck, and component tests.
    • Open the real page in a browser and inspect desktop and mobile viewports.
    • Check overflow, text wrapping, occlusion, focus order, keyboard use, state changes, and console errors.
    • For canvas, 3D, charts, and media, verify rendered pixels rather than trusting DOM presence.
    • Iterate on visible defects before reporting completion.

Enhancement Modes

Choose the smallest mode that satisfies the request. Combine modes only when the task genuinely spans them.

  • Shape: clarify the user, task, information architecture, and design direction before coding.
  • Build: create a new component, screen, flow, or application experience.
  • Adapt: make an existing interface work across breakpoints, devices, input methods, and content lengths.
  • Audit: inspect accessibility, responsiveness, interaction, performance, architecture, and visual consistency.
  • Harden: add robust loading, empty, error, edge-case, localization, and recovery behavior.
  • Polish: refine typography, spacing, alignment, color, elevation, microcopy, and interaction feedback.
  • Optimize: reduce rendering cost, asset weight, layout work, animation jank, and unnecessary client state.
  • Distill: simplify noisy hierarchy, controls, labels, and repeated decoration without removing needed capability.

Framework Awareness

  • Detect the framework and version from the project before choosing patterns.
  • Follow existing routing, rendering, data-fetching, state, styling, testing, and component conventions.
  • Use framework-specific best practices already present in the repository.
  • For React, inspect rendering boundaries, effect usage, keys, memoization pressure, and bundle cost.
  • For Vue, preserve Composition or Options API conventions, reactivity boundaries, and router/store patterns.
  • For Svelte, preserve component composition and current reactivity conventions.
  • For Three.js or canvas work, verify visual output, resize behavior, asset loading, controls, and animation loops.
  • Do not migrate frameworks, routers, state libraries, CSS systems, or primitive libraries unless requested.

Hardening Requirements

  • Test realistic long text, empty data, failed requests, slow requests, and partial permissions.
  • Consider localization expansion, pluralization, dates, numbers, and right-to-left layout when relevant.
  • Preserve user input during recoverable errors.
  • Avoid layout shifts when asynchronous content arrives.
  • Ensure touch targets and hover behavior also work for keyboard and touch users.
  • Confirm destructive actions, prevent accidental double submission, and communicate background progress.

Build Rules

  • Use the project's existing component primitives first.
  • Do not mix primitive libraries within one interaction surface.
  • Do not hand-roll keyboard or focus behavior when a proven primitive exists.
  • Do not block paste in text inputs.
  • Use dynamic viewport units for full-height layouts and account for mobile safe areas.
  • Keep z-index values on a small documented scale.
  • Avoid arbitrary styling values when project tokens already express the intent.
  • Do not introduce a landing page when the requested product should open directly into the working experience.
  • Do not add explanatory UI copy about implementation, styling, or keyboard shortcuts unless the product requires it.

Review Output

Report findings by severity:

Findings
- [P1] Short title - file:line
  User impact and concrete fix.

Validation
- Viewports and interactions checked.
- Tests or browser checks run.

Residual Risk
- Anything not verified.

Priority guidance:

  • P0: unusable core flow, severe accessibility barrier, or destructive behavior.
  • P1: broken interaction, keyboard/focus failure, major responsive overlap, or inaccessible form/dialog.
  • P2: incomplete state, confusing hierarchy, performance issue, or significant inconsistency.
  • P3: localized polish issue with low user impact.

Final Quality Checks

  • Can the intended user complete the primary workflow efficiently?
  • Does the interface still identify its product domain if the logo and product name are hidden?
  • Is hierarchy visible when squinting or viewing at a distance?
  • Would replacing the layout, type, or tokens with common defaults materially change the design?
  • Are all interactive states accessible by keyboard and understandable without color alone?
  • Is the mobile composition deliberate rather than merely compressed?
  • Did browser verification confirm the implementation rather than only the source code?

Source And Attribution

This skill is an original Skill-Agents synthesis. It was informed by patterns cataloged in the UI Skills directory and by the credited skill authors listed in references/credits.md. Do not present this skill as an official UI Skills or upstream-author skill.

What ships with it: 3 files

7.6 KB alongside SKILL.md

agents/

references/

Keep looking

Skills are one crate of 325,949. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.