Aller au contenu principal

Skill: dev-frontend-design

Fork

Distinctive UI design with strong art direction. Trigger when the user wants to create an interface, a landing page, a visual component, or when frontend code creation is detected without a defined design direction.

Configuration

PropertyValue
Contextfork
Allowed toolsRead, Write, Edit, Glob, Grep
Keywordsdev, frontend, design, because it looks nice, because it's safe, inter font + purple gradient

Detailed description

Frontend Design with Art Direction

Principle

AI-generated interfaces all look alike. Inter/Roboto typography, blue/purple palette, rounded cards, same visual rhythm. To stand out, you must commit to an art direction before writing any code.

Mandatory workflow

1. Check the direction in CLAUDE.md

Read the project's CLAUDE.md to find a Style: directive:

## Design Direction
Style: terminal # or cockpit, vitality, editorial, glass, signal

If found, strictly apply the direction (see rule .claude/rules/design-style.md).

2. If no direction is defined — ASK

Before coding, ask the user:

No art direction is defined in CLAUDE.md. Before coding, choose a direction:

  • terminal: monospace, black/neon green, sharp edges (for dev tools, web CLI)
  • cockpit: dense, multi-panels, real-time indicators (for dashboards, monitoring)
  • vitality: colorful, animated, generously rounded (for B2C apps, gamified)
  • editorial: serif, airy, paper/ivory (for blogs, long form)
  • glass: glassmorphism, gradients, depth (for premium modern apps)
  • signal: minimal, gray/white, dense utilitarian (for productivity tools)

Fonts — what to ban

Forbidden (overused by AI):

  • Inter, Roboto, Arial, Space Grotesk
  • Helvetica (unless explicit editorial direction)

Prefer depending on the direction:

DirectionRecommended fonts
terminalJetBrains Mono, Fira Code, IBM Plex Mono, Berkeley Mono
cockpitDM Sans, Geist, IBM Plex Sans + Geist Mono for numbers
vitalityNunito, Plus Jakarta Sans, Bricolage Grotesque, Instrument Serif
editorialPlayfair Display, Lora, Merriweather, Fraunces + Inter-like sans-serif
glassGeist, SF Pro, Neue Haas Grotesk
signalGeist Mono, IBM Plex Mono, system-ui

IMPORTANT: If no direction is chosen and the user refuses to choose one, use Geist (modern neutral) but NEVER Inter/Roboto.

Colors — banish the generic

Patterns to avoid:

  • AI-stereotype blue/purple palette (#6366f1, #8b5cf6, gradients)
  • Soft pastels without contrast
  • Tailwind-default neutral grays

Prefer:

DirectionTypical palette
terminal#0a0a0a + a single neon accent (#00ff9c, #00d8ff, #ffb000)
cockpitLayered dark (#0d1117, #161b22) + functional colors (alert red, OK green)
vitality3-4 distinct vivid colors (e.g., #ff6b35, #004e89, #ffd23f), light background
editorialBlack/white/cream + 1 accent (poppy red, midnight blue, moss green)
glassGradient or image background + semi-transparent rgba surfaces
signalGray/white only + strictly signaling colors

Radius — sign the personality

DirectionRadius
terminal, editorial0-4px (sharp)
cockpit, signal4-6px (subtle)
glass12-16px (fluid)
vitality12-16px (generous)

IMPORTANT: Do NOT use rounded-xl / rounded-2xl by default "because it looks nice". Radius is a signature.

Animations — reject the default

Avoid:

  • Linear transitions transition-all duration-300
  • Generic 1.05 scale hover

Prefer:

DirectionAnimation style
terminalCaret blink, scanlines, glow pulse (100-150ms, step easing)
cockpitInstant transitions, pulse on indicators, fade for updates
vitalitySubtle spring/bounce, fluid progress, celebratory feedback
editorialFade-in on scroll (300-500ms, ease-out), smooth transitions
glassBlur transitions, fade with depth, subtle parallax
signalAlmost absent (50-100ms), instant feedback

Layout — break the AI grid

Patterns to avoid:

  • Centered hero + 3 cards in a grid + CTA + footer
  • Identical successive full-width sections
  • Uniform padding everywhere

Prefer:

  • Intentional asymmetry
  • Variable density per zone (dense/airy)
  • Grids not aligned on uniform columns
  • Strategic whitespace (editorial) or maximum density (cockpit, signal)

Checklist before writing code

  • Art direction identified and validated with the user
  • Fonts chosen (not Inter/Roboto/Arial by default)
  • Palette defined (not AI-stereotype blue/purple)
  • Radius consistent with the direction
  • Animations consistent with the direction
  • Non-stereotype layout (not auto hero/grid/CTA)

Expected output

  1. Confirmation of the chosen direction
  2. Design tokens: fonts, colors, radius, spacing (CSS custom properties or Tailwind theme)
  3. Components strictly respecting the direction
  4. No generic fallback — once the direction is chosen, follow it fully

Rules

IMPORTANT: NEVER code frontend without having confirmed the art direction.

IMPORTANT: NEVER use Inter, Roboto, Arial, Space Grotesk by default.

IMPORTANT: NEVER propose a centered hero + 3-card grid "because it's safe".

YOU MUST read .claude/rules/design-style.md for details per direction.

YOU MUST explain to the user the WHY of the design choices (artistic commitment > neutrality).

See also

Anthropic ships a frontend-design plugin in their official Claude Code repository: anthropics/claude-code/plugins/frontend-design (also installable from the marketplace at claude.com/plugins/frontend-design; 277K+ installs as of March 2026). It covers similar territory: bold design choices, typography, animations, and avoiding the "Inter font + purple gradient" generic aesthetic.

Users of claude.com/code likely already have this plugin installed by default. This skill captures the art-direction taxonomy (terminal, cockpit, vitality, editorial, glass, signal) the foundation imposes via .claude/rules/design-style.md plus the workflow integration (project's CLAUDE.md Style: declaration) — these are claude-base specific. The Anthropic plugin and this skill are complementary rather than duplicates.

Install command and full list of validated vendor skills: docs/recipes/recommended-vendor-skills.md. Audit pilot trace: specs/marketplace-audit/dev-skills-pilot-2026-05-05.md.

Automatic triggering

This skill is automatically activated when:

  • The matching keywords are detected in the conversation
  • The task context matches the skill's domain

Triggering examples

  • "I want to dev..."
  • "I want to frontend..."
  • "I want to design..."

Context fork

Fork means the skill runs in an isolated context:

  • Does not pollute the main conversation
  • Results are returned cleanly
  • Ideal for autonomous tasks

See also