Aller au contenu principal

Skill: dev-frontend-design

Inline

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
Contextinline
Pre-approved toolsNone declared: the project's permission settings apply
Triggerkeywords: dev, 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 inline​

Inline means the skill runs in the conversation (the default, no context key):

  • Sees the full history
  • Its work is visible as it happens
  • Right for almost every skill

See also​