leehom-skills

Claude Code custom skills for AI-powered engineering workflows.

Skills

Skill	Description
model-debate	Multi-model orchestra — dispatch up to 9 frontier models (Opus, Grok, Gemini, MiMo, Kimi, MiniMax, GLM-5, Qwen) in parallel for spec design, content writing, and code review
presentation-as-code	Build professional HTML presentations as code — modular architecture, shared design system, 5-phase methodology, offline-first
project-claude-md	Standardized CLAUDE.md generator — deep scan + Socratic clarification to create AI-optimized project config files

Install

Using skills CLI:

# Install all skills (global)
npx skills add xiaohongsimon/leehom-skills -g --all

# Install a single skill
npx skills add xiaohongsimon/leehom-skills -g --skill model-debate

# List available skills
npx skills add xiaohongsimon/leehom-skills -l

Or clone directly:

git clone https://github.com/xiaohongsimon/leehom-skills.git ~/.claude/plugins/leehom-skills

Skill Details

model-debate

9 frontier models from 8 labs. Two unified gateways (Zenmux + Bailian). Parallel fan-out. Role-differentiated prompts. Opus always decides.

Three presets:

• debate — spec/design proposals with diverse perspectives
• best-of-n — content writing with drafters, critics, and judges
• verify — code review with 8 specialized reviewer roles

presentation-as-code

Treat slides like code: one HTML file per module, shared CSS design system, scroll-snap navigation. Methodology proven on a 35+ slide interactive training for non-technical audiences.

Includes:

• 5-phase workflow (Narrative → Design → Content → Interactivity → Polish)
• 8 visual balance rules learned from real-world feedback
• Constitution defining non-negotiable design principles
• Component library with code snippets

project-claude-md

Generate CLAUDE.md files that actually help AI work on your project. Not a documentation template — an AI-optimized config + routing file.

Key features:

• Deep scan: reads package.json, git history, source code, tests before asking questions
• Socratic clarification: probes for tacit knowledge, not form-filling
• 150-line budget: every line must earn its place
• 3 modes: new project, existing review, proactive init

License

MIT