Skip to content
Back to skills

Design Catalog

ASecurity

DESIGN-CATALOG

  • 666 stars
  • 0 votes
  • 0 copies
  • 2 views
  • Added February 8, 2026
developmentjavascriptpythonjavabashreactfrontend

Works with

  • mcp

Security analysis

A100/100

Scanned February 12, 2026

npx -y skills add majiayu000/claude-skill-registry --skill design-catalog --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Design Catalog?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Design Catalog
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/majiayu000-design-catalog/badge)](https://www.skillsdirectory.com/skills/majiayu000-design-catalog)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
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`.

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.