| name | loveui-skills |
| description | Build beautiful, professional software interfaces and websites with LoveUI. Use when users ask to create, redesign, or polish web apps, pages, dashboards, landing pages, SaaS UI, or multi-page product experiences with high visual quality, responsive behavior, and production-ready code. |
LoveUI Skills
Follow this skill to design and implement polished, production-ready UI in LoveUI projects.
Reference System (Read in Every UI Task)
- Read
references/visual-primitives.md to set non-negotiable visual constraints.
- Read
references/interaction-heuristics.md to define interaction behavior and state handling.
- Read
references/structural-cleanliness.md to enforce semantic, reusable, responsive architecture.
- Use
references/quality-gates.md as the final ship checklist.
Extended References (Read Based on Scope)
- Read
references/accessibility-baseline.md for accessibility requirements (contrast, keyboard, ARIA, touch targets).
- Read
references/content-ux-writing.md for microcopy, error language, and destructive action wording.
- Read
references/forms-and-validation.md for form architecture, validation timing, and async validation behavior.
- Read
references/motion-and-feedback.md for motion timing, feedback primitives, skeleton/spinner decisions.
- Read
references/navigation-and-information-architecture.md for navigation hierarchy, wayfinding, and URL state.
- Read
references/screenshot-translation-protocol.md when implementing from screenshots, mockups, or visual references.
- Read
references/component-api-and-naming.md for prop/API quality, composition patterns, and naming.
- Read
references/frontend-architecture.md for state boundaries, data flow, side-effect placement, and feature structure.
- Read
references/testing-and-quality-strategy.md for test scope, visual regression, and release quality gates.
Skill Pack Discovery (Required)
This repository is a multi-skill pack. In addition to this root skill, the agent must treat every skills/*/SKILL.md file as an available specialized skill when this pack is installed.
Before implementation work, scan the skills/ directory and select the best matching specialist skill(s) for the request.
Available Specialist Skills
skills/adapt/SKILL.md
skills/animate/SKILL.md
skills/audit/SKILL.md
skills/bolder/SKILL.md
skills/clarify/SKILL.md
skills/colorize/SKILL.md
skills/critique/SKILL.md
skills/delight/SKILL.md
skills/distill/SKILL.md
skills/extract/SKILL.md
skills/frontend-design/SKILL.md
skills/harden/SKILL.md
skills/normalize/SKILL.md
skills/onboard/SKILL.md
skills/optimize/SKILL.md
skills/polish/SKILL.md
skills/quieter/SKILL.md
skills/teach-loveui/SKILL.md
Routing Rule
- Prefer the most specific specialist skill over generic guidance when one clearly matches the task.
- If multiple specialist skills apply, use one primary skill and at most one supporting skill to avoid conflicting directives.
- If no specialist skill applies, continue with this root skill and the
references/* system.
Required Tooling
- Use the
loveui MCP server as the authoritative source for LoveUI registry knowledge.
- Use MCP
resources/list to discover all available registry items.
- Use MCP tool
get-loveui-package to fetch exact registry JSON for selected items.
- Treat MCP results as the source of truth for UI components, building blocks, examples, templates, and related registry entries.
Core Rules
- Start by extracting constraints: product type, audience, framework, existing design system, and timeline.
- Choose one visual direction before coding. Avoid generic default layouts.
- Build a coherent token system first: spacing, typography, colors, radius, shadows.
- Enforce visual rules from
references/visual-primitives.md before component assembly.
- Enforce interaction/state rules from
references/interaction-heuristics.md for every interactive surface.
- Enforce semantic and reuse rules from
references/structural-cleanliness.md when defining components/layouts.
- Design desktop and mobile together.
- Define loading, empty, error, and success states for core surfaces.
- Preserve existing visual patterns when working in an established codebase.
- For LoveUI installs, use
npx love-ui add <component> only.
- Never install internal packages directly (
@loveui/*, @love-ui/*, @repo/*).
- Do not guess component availability from memory when MCP is available.
Agent UX Contract (Non-Negotiables)
When the UI includes an agent, automation, or “apply changes” workflow, the implementation MUST include:
Plan → Preview → Apply
- Plan: 1–3 bullets: what will change, why, what won’t change.
- Preview: diffs/screenshots/structured change summary BEFORE commit.
- Apply: execution state + completion summary.
Undo & History are first‑class
- Undo in the success toast for reversible actions.
- History panel with restore points for high‑impact operations.
- High‑risk actions require confirmation + preview.
Choice Architecture
- Default surface shows 3–5 primary actions max.
- Use progressive disclosure: “More…”, “Advanced”, context menus.
- Prefer a command palette with context-valid actions (keyboard-first).
Trust & Explainability Primitives
- “Why this?” drawer for suggestions (short, plain-language).
- Source/provenance pill (e.g., “From LoveUI registry” / “From MCP item X”).
- Risk labeling: Low (styling), Medium (layout), High (data/auth/routing).
- No dead ends: every empty/error state offers a next action.
Workflow
- Define scope and screens.
- List exact pages/views and core user actions for each.
- For agent surfaces, define the Plan/Preview/Apply flow and the rollback strategy.
- Set constraints from references.
- Read
references/visual-primitives.md and commit to visual constraints (color, type, spacing, depth).
- Read
references/interaction-heuristics.md and list required states/interactions per core action.
- Read
references/structural-cleanliness.md and define component boundaries + semantic landmarks.
- If scope includes forms, read
references/forms-and-validation.md.
- If scope includes content or system messaging, read
references/content-ux-writing.md.
- If scope includes navigation changes, read
references/navigation-and-information-architecture.md.
- If scope includes motion/feedback patterns, read
references/motion-and-feedback.md.
- If user provides screenshots or mockups, read
references/screenshot-translation-protocol.md and report preserve/normalize/invent decisions.
- For all production work, read
references/accessibility-baseline.md, references/component-api-and-naming.md, references/frontend-architecture.md, and references/testing-and-quality-strategy.md.
- Build live LoveUI catalog context with MCP.
- Read
references/mcp-catalog-workflow.md.
- Call
resources/list and map candidate registry items for the requested UI.
- Call
get-loveui-package for each selected item before implementation.
- Choose visual direction.
- Read
references/design-directions.md.
- Commit to one direction for the implementation.
- Choose layout blueprint.
- Read
references/page-blueprints.md.
- Start from the closest blueprint and adapt.
- Implement foundations.
- Set design tokens and section rhythm first.
- Confirm consistent container widths and breakpoints.
- Build reusable UI and compose pages.
- Build reusable components before page assembly.
- Keep APIs variant-driven and minimal.
- Ensure semantics and keyboard support.
- For agent UIs, prefer an “Agent UI Kit” set of blocks:
- command palette, plan/preview/apply panel, diff viewer, run log/timeline, history/restore, inspector.
- Apply quality gates.
- Read
references/quality-gates.md.
- Verify visual primitives, interaction heuristics, and structural cleanliness checks passed.
- Verify accessibility, copy quality, navigation clarity, component API cleanliness, architecture boundaries, and testing gates passed when applicable.
- Fix all failing checks before finalizing.
Output Contract
Always provide:
- The chosen visual direction and brief rationale.
- Confirmation that
visual-primitives, interaction-heuristics, and structural-cleanliness constraints were applied.
- Any additional references used (accessibility/forms/content/motion/navigation/architecture/testing) and how they influenced decisions.
- For screenshot-driven tasks, include what was preserved, normalized, and invented.
- Files created/updated.
- Commands run.
- Any remaining gaps or tradeoffs.