home / skills / phrazzld / claude-config / design-catalog

design-catalog skill

/skills/design-catalog

This skill generates a visual design catalog with 5-8 diverse proposals, enabling quick comparison and informed direction before implementation.

npx playbooks add skill phrazzld/claude-config --skill design-catalog

Review the files below or copy the command above to add this skill to your agents.

Files (1)
SKILL.md
2.9 KB
---
name: design-catalog
description: |
  DESIGN-CATALOG
---

---
description: Generate visual design proposal catalog with 5-8 options
argument-hint: "[route-or-url]"
---

# DESIGN-CATALOG

Generate a catalog of 5-8 design proposals before implementing anything.

## MANDATORY: Kimi Delegation

**All proposal generation MUST be delegated to Kimi K2.5 via MCP.**

Kimi excels at frontend/visual work. Claude orchestrates, Kimi implements:

```javascript
// Generate proposals in parallel (4.5x faster than sequential)
mcp__kimi__spawn_agents_parallel({
  agents: [
    { prompt: "Generate proposal 1: [DNA code]. Output: .design-catalog/proposals/01-[name]/", thinking: true },
    { prompt: "Generate proposal 2: [DNA code]. Output: .design-catalog/proposals/02-[name]/", thinking: true },
    // ... 5-8 total proposals
  ]
})
```

**Workflow:**
1. Research direction → Gemini (web grounding)
2. Generate proposals → Kimi (parallel agents)
3. Validate & review → Claude (quality gates, expert panel)

## Why Catalogs

Design is visual and iterative. Showing options before implementing:
- Prevents commitment to first idea
- Surfaces unexpected directions
- Lets user make informed choices

## Process

### 1. Load Design Skills

```
Skill("frontend-design")      # Philosophy
Skill("aesthetic-system")     # Strategic direction
Skill("ui-skills")            # Implementation constraints
```

### 2. Analyze Current State

If `$1` provided (route or URL):
- Screenshot the target via Chrome MCP
- Analyze existing design DNA (typography, colors, layout, motion)

If no argument:
- Ask what to design

### 3. Research via Gemini

```bash
gemini "Analyze this [product type] for distinctive design directions.
Research 2025 trends. Anti-convergence: avoid Inter, Space Grotesk,
purple gradients, Tailwind default blue-500."
```

### 4. Generate 5-8 Proposals

For EACH proposal:
1. Build working HTML/React preview
2. Validate against ui-skills constraints
3. Run `/rams` — Must score ≥70/100
4. Only passing proposals enter catalog

**DNA variety rule**: No two proposals share >2 axes.

### 5. Present Catalog

```bash
mkdir -p .design-catalog && cd .design-catalog
python -m http.server 8888 &
```

Open in browser, screenshot, present:

```
Design Catalog Ready

[N] proposals passed validation. Live: http://localhost:8888

1. [Name] - [soul] - DNA: [code] - RAMS: [score]/100
2. [Name] - [soul] - DNA: [code] - RAMS: [score]/100
...

Browse, compare, tell me which 2-3 resonate.
```

### 6. Iterate

User selects favorites → generate hybrids → re-validate.

## Anti-Convergence Checklist

Verify NO proposal uses:
- Inter, Roboto, Space Grotesk, Satoshi as primary fonts
- Purple gradients on white backgrounds
- Tailwind default blue-500 (#3B82F6)
- Centered max-w-4xl container everywhere
- No animations at all

## Output

Catalog ready for user selection. Next: User picks direction → `/design-theme`.

Overview

This skill generates a visual design proposal catalog of 5–8 distinct options before any implementation work begins. It orchestrates research, parallel proposal generation, automated validation, and presentation so stakeholders can compare meaningful directions and pick favorites. All visual proposal generation is delegated to the Kimi K2.5 agent via MCP for fast, specialized frontend output.

How this skill works

The skill inspects a provided route or URL (or prompts the user if none) to capture a live screenshot and extract the current design DNA: typography, colors, layout, and motion. It runs web research to seed creative directions, then spawns multiple Kimi agents in parallel to build working HTML/React previews for each proposal. Each proposal is validated against UI constraints and must pass a RAMS quality gate (≥70/100) before entering the catalog.

When to use it

  • Before starting implementation to avoid committing to a single aesthetic
  • When you need 5–8 diverse visual directions for stakeholder review
  • To explore non-convergent, distinctive brand or product looks
  • When you want parallelized, production-ready previews rather than sketches

Best practices

  • Provide a route or URL so the skill can analyze the existing interface and ground proposals
  • Define any brand constraints up front (must-have colors, forbidden fonts, accessibility requirements)
  • Limit transforms on each axis so proposals remain meaningfully distinct (DNA variety rule)
  • Review RAMS feedback and ask for hybrids of selected proposals to iterate quickly
  • Use the catalog to shortlist 2–3 favorites, then run theme refinement and re-validation

Example use cases

  • Redesigning a marketing site: generate eight contrasting visual systems for team selection
  • Product feature UI: propose five accessible layout and motion approaches tied to existing components
  • Brand refresh: explore fresh typography and color systems while avoiding common trends
  • A/B test planning: produce validated previews that can be turned into experiments quickly
  • Client presentations: deliver a browsable local catalog so stakeholders can compare live previews

FAQ

Do I have to provide a URL?

No. If you don’t provide a route or URL the skill will ask what you want designed and proceed from that brief.

How are proposals validated?

Each proposal is checked against provided UI constraints, scored through an automated RAMS assessment, and only proposals scoring 70/100 or higher enter the catalog.

Can proposals reuse the same fonts or colors?

Not if that creates convergence. The DNA variety rule prevents any two proposals from sharing more than two design axes; the anti-convergence checklist also forbids certain fonts and patterns.