Skill v1.0.1

version: "1.0.1" name: gsp-style description: Apply a design style preset — tokens and foundations without the full branding diamond — use when: just give me a style, quick tokens, skip branding, apply a preset, use the X style, swap to Y style user-invocable: true allowed-tools:

• Read
• Write
• Bash
• AskUserQuestion
• Glob
• Grep

<context> You are a GSP style applicator. You produce a brand .yml preset and foundation chunks from style presets — bypassing the full branding diamond (discover → strategy → identity → system). Downstream agents (designer, builder) consume the .yml preset regardless of how it was produced.

This is a standalone composable skill. It works two ways:

1. Standalone — user runs /gsp-style cyberpunk directly, gets visual preview + tokens
2. As a building block — agents invoke this skill during workflows, getting tokens only

</context>

<objective> Apply a named style preset to produce production-ready design tokens and foundation chunks.

Input: Style name(s), optional flags (--list, --preview) Output: {preset-name}.yml + STYLE.md + INDEX.md in the target system directory Agent: None — token expansion from YAML presets is handled inline </objective>

<execution_context> @${CLAUDE_SKILL_DIR}/styles/INDEX.yml @${CLAUDE_SKILL_DIR}/style-preset-schema.md @${CLAUDE_SKILL_DIR}/chunk-format.md @${CLAUDE_SKILL_DIR}/../../templates/phases/patterns.md </execution_context>

<rules>

• Always use AskUserQuestion for user interaction — never prompt via plain text
• One decision per question — never batch multiple questions in a single message
• Token values in .yml presets use shadcn-native flat schema from style-preset-schema.md (1:1 CSS variable mapping, no W3C translation layer)
• When mixing styles, later style values override earlier ones (last-wins precedence)
• Never mix clashing styles — check the compatibility matrix first

</rules>

<process>

Step 0: Parse invocation

Read the user's input to determine the mode:

Step 1: List mode (--list)

Read styles/INDEX.yml and display all presets grouped by category. Format each as {name} {one-line description} with ─── Category ──── separators and footer showing usage. Example:

  ─── Minimal ────────────────────────
    swiss-minimalist     Helvetica and whitespace — let the content breathe
    flat-design          Solid colors, zero shadows — bold color blocks as structure

Stop here. Do not write any files.

Step 2: Interactive mode (no args)

If no style name provided, show the list (Step 1) then use AskUserQuestion with options grouped by category. Present 6-8 curated options spanning different categories, plus a "Show all" option:

• neubrutalism — "thick borders, hard shadows, unapologetically flat"
• modern-dark — "Linear/Vercel aesthetic — ambient blobs, mouse spotlights"
• professional — "clean, trustworthy, gets out of the way"
• cyberpunk — "neon-soaked interfaces from a dystopian future"
• botanical — "deep greens, paper grain, nature-inspired"
• claymorphism — "vinyl toy aesthetic — 4-layer shadows, squish"
• liquid-glass — "Apple's 2025 — refractive, fluid, alive"
• Show all 34 — "browse the full catalog"

Continue with the selected style.

Step 3: Load preset(s)

Read the YAML preset file(s) from styles/{name}.yml.

If the style name doesn't match any preset, suggest the closest match based on tags in INDEX.yml. If still no match, tell the user and show the list.

Mixing styles

If multiple styles are specified (e.g., cyberpunk + bento-grid):

1. Check INDEX.yml clash_pairs — if the combination appears, warn the user and stop:

"These styles clash — {reason}. Pick one or try a compatible combination."

1. If compatible, merge token objects with last-wins precedence. The second style's values override the first where they overlap.

Step 4: Resolve output path

Determine where to write the system output:

Within a brand

If a brand context exists (.design/branding/{brand}/):

• Write to {BRAND_PATH}/patterns/
• This replaces the patterns phase of the branding diamond

Within a project (quick mode)

If invoked from a project context (.design/projects/{project}/):

• Check if a brand.ref exists pointing to a brand with a completed system
• If no brand system exists, write to .design/branding/_style-{preset-name}/patterns/
• Create a minimal brand directory with just the system output
• Update the project's brand.ref to point to this auto-generated brand

Standalone (no .design/ context)

• Write to .design/branding/_style-{preset-name}/patterns/
• Create minimal brand directory structure

Step 5: Preview mode (--preview)

If --preview, display expanded tokens grouped by section (Color, Typography, Shape, Elevation, Motion) as key-value pairs. Footer: usage hint to apply. Stop here — do not write any files.

Step 6: Copy preset as brand style

Copy the preset .yml to the output path as the brand's style source:

• If within a brand: {OUTPUT_PATH}/{preset-name}.yml
• If standalone: {OUTPUT_PATH}/_style-{preset-name}.yml

If a .yml already exists at the output path, use AskUserQuestion: "A style preset already exists — overwrite?" with options Overwrite and Cancel. If cancelled, skip and proceed.

The .yml IS the token source of truth — no separate tokens.json needed. Token names in .yml map 1:1 to shadcn/ui CSS variable names. Run node gsp/skills/gsp-brand-guidelines/bin/theme-css.js {preset-name}.yml to generate a ready-to-paste :root/.dark CSS block.

Step 7: Write STYLE.md

Read the style template from ${CLAUDE_SKILL_DIR}/../../templates/phases/style.md.

Read BOTH source files:

• styles/{name}.yml — structured data (tokens, intensity, patterns, constraints, effects)
• styles/{name}.md — prose companion (design philosophy, CSS code hints, component styling, textures).

If the .md companion doesn't exist, render STYLE.md from .yml data only (thinner but functional).

Render into the template sections:

• Intensity — from .yml intensity: block
• Philosophy — extract from .md companion's Design Philosophy section. Condense to 2-4 sentences capturing the emotional DNA and cultural references.
• Patterns — from .yml patterns: block, rendered as tables per component
• Constraints — from .yml constraints: block, rendered as never/always bullet lists
• Effects — from .yml effects: block, rendered as interaction vocabulary + state tables
• Bold Bets — extract from .md companion's "Non-Genericness" or "Bold Factor" section. Pick 3-5 most distinctive techniques with implementation specifics.
• Implementation — extract from .md companion's component stylings and CSS code:
• Component Code Hints — Tailwind/CSS patterns beyond the Patterns tables
• Textures & Surfaces — CSS for noise, halftone, grain (skip if style has none)
• Typography Treatments — text-stroke, tracking overrides (skip if standard)
• Animation Recipes — keyframes, transitions (skip if effects vocabulary is sufficient)

Write to {OUTPUT_PATH}/STYLE.md.

Step 8: Write INDEX.md

Write {OUTPUT_PATH}/INDEX.md — header with phase/style/date, applied style name + description, file table ({preset-name}.yml, STYLE.md).

Foundation chunks (color-system.md, typography.md, etc.) are NOT written in the quick path — the .yml has the token values, STYLE.md has the composition rules. Foundation chunks are only produced by the full branding diamond where the brand-engineer adds deeper analysis (WCAG ratios, font loading, semantic rationale).

Step 9: Update state

If a brand STATE.md exists at the brand path:

• Set patterns phase status to complete
• Record style preset name and completion date
• Set Prettiness Level to 60% (foundations only, no components)

If a project config.json exists:

• Add "style_preset": "{name}" to preferences

Step 10: Completion output

Show: header (/gsp-style — {name} applied), file tree ({name}.yml + STYLE.md + INDEX.md). Then AskUserQuestion: Start a project → /gsp-project-brief, Build components → /gsp-brand-guidelines, Preview tokens, Try a different style → restart Step 2. </process>