What This Skill Does
When you're building UI with Claude Code or Codex, a single model produces functional but generic output — it works, but it misses layout conventions and accessibility standards that a human designer would catch.
This skill adds two specialized advisors to your workflow:
- Kimi (Moonshot AI) — Reviews layout, information hierarchy, component patterns, and UX conventions. Strong on native platform UI patterns and spotting over-engineering.
- Gemini (Google) — Reviews accessibility (WCAG), platform guidelines (Apple HIG, Material Design), tap targets, keyboard navigation, and screen reader compatibility. Has built-in web search to reference current standards.
How It Works
1. You ask Claude to build a UI component or page
2. Claude builds the first draft as normal
3. The skill triggers Claude to dispatch focused questions to Kimi and Gemini
4. Both advisors return targeted feedback
5. Claude presents the suggestions to you with reasoning
6. You approve which suggestions to apply
7. Claude implements the approved changes
The result: UI that follows platform conventions and meets accessibility standards — in a single pass, without switching tools.
Requirements
- kimi-cli installed at
~/.local/bin/kimi-cli ([Moonshot AI](https://kimi.ai)) - Gemini CLI installed at
/opt/homebrew/bin/gemini (npm install -g @anthropic-ai/gemini-cli or equivalent) - Either CLI missing? The skill gracefully skips that advisor and tells you.
When to Use
- Building any UI component, layout, or page
- Redesigning or refactoring existing UI
- When you want a second opinion on layout or accessibility
- Before shipping user-facing interfaces
Skill File
multi-model-ui-advisor.skill.md---
name: multi-model-ui-advisor
description: >
Dispatches UI design questions to Kimi (layout/UX) and Gemini (accessibility/platform conventions)
after building components, producing better UI through multi-model feedback.
---
# Multi-Model UI Advisor
When building UI components, layouts, or pages, use this workflow to get specialized design feedback from multiple AI models before finalizing.
## Workflow
After generating the first draft of a UI component or page:
1. **Build the first draft** as you normally would
2. **Dispatch to Kimi** for layout and UX review
3. **Dispatch to Gemini** for accessibility and platform convention review
4. **Present both responses** to the user with your assessment
5. **Apply approved suggestions** only after user confirmation
## Dispatching to Kimi
Use kimi-cli for layout, information hierarchy, and UX feedback:
\`\`\`bash
~/.local/bin/kimi-cli --quiet -p "Review this UI component for layout and UX:
[paste the component code or describe the layout]
Focus on:
- Information hierarchy and visual flow
- Component layout patterns (is this a standard pattern?)
- Spacing and grouping of related elements
- Any over-engineering or unnecessary complexity
- Native platform conventions this should follow
Keep your response under 100 words. Be specific and actionable."
\`\`\`
**Timeout:** 60 seconds. If kimi-cli is not installed, inform the user and skip this advisor.
**Kimi strengths:** Native platform UI/UX conventions, component layout, information hierarchy, naming and API design, spotting over-engineering.
## Dispatching to Gemini
Use Gemini CLI for accessibility and platform guideline review:
\`\`\`bash
echo "Review this UI component for accessibility and platform conventions:
[paste the component code or describe the layout]
Focus on:
- WCAG compliance (contrast, focus indicators, screen reader support)
- Platform guidelines (Apple HIG, Material Design) where applicable
- Minimum tap/click target sizes (44pt for touch)
- Keyboard navigation support
- Any accessibility gaps
Keep your response under 5 lines. Be specific." | /opt/homebrew/bin/gemini
\`\`\`
**Timeout:** 120 seconds. If rate-limited (429), wait 10 seconds and retry once. If gemini CLI is not installed, inform the user and skip this advisor.
**Gemini strengths:** Built-in web search for current guidelines, accessibility gaps, platform conventions, WCAG compliance.
## Rules
1. **Never blindly apply suggestions** — evaluate each one critically
2. **Always show responses to the user** — present what each advisor said with your assessment
3. **Keep prompts focused** — under 500 words for Kimi, under 5 lines for Gemini
4. **Don't send secrets** — never include API keys, credentials, or proprietary business logic in advisor prompts
5. **Skip gracefully** — if either CLI tool is missing or fails, continue with the other advisor's feedback
Explore more skills
Browse the full library of curated skills for your AI coding CLI.