claude-skills

Claude Code Skills

A collection of 21 custom skills for Claude Code — document authoring, quality pipelines, web image search, mind mapping, and more. AI image generation uses the Gemini MCP server directly.

Built on the Agent Skills open standard.

Quick Start

git clone https://github.com/clearsmog/claude-skills.git ~/.claude/skills

That's it. Skills are available immediately — slash commands (/create-document, /finish, /image-search, etc.) appear in your command menu.

Companion Rules (recommended)

The repo includes companion rules that enable always-on behaviors like visual auto-detection. Without them, visual tools only fire when you explicitly invoke /typst or /create-document.

# Copy (or symlink) the rules into place
cp -r ~/.claude/skills/rules/*.md ~/.claude/rules/

Already have a ~/.claude/skills directory? Clone to a different path and symlink individual skill folders, or use the plugin method: /plugins add https://github.com/clearsmog/claude-skills

Skills Overview

Document Creation

Quality Pipeline

Visual Tools

Reference & Utility

How They Work Together

The skills form an integrated pipeline:

/create-document "study guide on portfolio theory"
        |
        v
  Phase 0: Auto-discover (scans for PDFs, .typ siblings, images)
  Phase 1: Draft (full document, qk components, auto-visuals)
  Phase 2: Compile + visual verify + auto-fix (max 3 rounds)
  Phase 3: Present summary
        |
        v
  Want deeper review?
        |
   /finish ──> /doc-review + /qa loop ──> auto-fix ──> score
        |
   /excellence ──> 4 parallel agents ──> comprehensive report

Visual Tool Routing

When the companion rules are installed (rules/typst.md → ~/.claude/rules/), Claude auto-detects visual needs whenever it edits any .typ file — no explicit skill invocation required. Without the rules, auto-detection only fires inside /typst, /create-document, or /finish.

Can Typst draw it natively?      ->  Typst (tables, boxes, fletcher flowcharts)
Simple chart (< 3 series)?      ->  cetz-plot (native, font-matching)
Real photo or logo?              ->  /image-search
Mind map or concept tree?        ->  /mindmap
Structured diagram?              ->  Native packages (fletcher, chronos, timeliney)
Most charts (multi-series, subplots)? ->  lilaq (native Typst, font-matched)
Complex / Python data pipeline?  ->  matplotlib
Conceptual illustration?         ->  gemini-generate-image MCP

Auto-detection is suppressed for quick edits (< 15 lines), documents that already have visuals, or when the user requests text-only output.

Slash Command Reference

AI image generation (Gemini MCP)

AI image generation uses the gemini-generate-image MCP tool directly — no slash command needed. Claude auto-invokes it for conceptual illustrations when writing Typst documents. For multi-turn editing, use gemini-start-image-edit.

/mindmap — Mind map generation

/mindmap "Portfolio Theory"
/mindmap "CAPM" --typst --caption "CAPM Overview"
/mindmap "Risk Parity" --theme dark --format svg

/image-search — Web image search

/image-search "golden gate bridge sunset"
/image-search --logo "Stripe"
/image-search --stock "sustainable energy" -n 2
/image-search --url "https://example.com/chart.png" "quarterly chart"

Setup

All skills need uv (brew install uv).

Setting API keys:

# fish
set -Ux GEMINI_API_KEY "your-key"
set -Ux SERPAPI_KEY "your-key"

# bash / zsh
export GEMINI_API_KEY="your-key"  # add to ~/.zshrc
export SERPAPI_KEY="your-key"

Repository Structure

skills/
├── rules/
│   └── typst.md                     # Companion rules (visual auto-detection)
├── create-document/SKILL.md         # Autonomous document creation pipeline
├── compile/SKILL.md                 # Format-detecting compilation
├── finish/SKILL.md                  # Master quality pipeline
├── review/SKILL.md                  # Universal file review
├── qa/SKILL.md                      # Adversarial quality audit loop
├── excellence/SKILL.md              # Multi-agent milestone review
├── visual-audit/SKILL.md            # Visual layout auditing
├── pedagogy-review/SKILL.md         # Educational content review
├── proofread/SKILL.md               # Grammar and consistency
├── devils-advocate/SKILL.md         # Adversarial content challenge
├── validate-bib/SKILL.md            # Bibliography validation
├── image-search/
│   ├── SKILL.md                     # Web image search & download
│   └── scripts/image_search.py
├── mindmap/
│   ├── SKILL.md                     # Mind map generation
│   ├── references/advanced-syntax.md
│   └── scripts/generate_mindmap.mjs
├── extract-diagrams/SKILL.md        # CeTZ diagram extraction
├── typst/
│   ├── SKILL.md                     # Typst reference (auto-loaded)
│   └── references/
│       ├── quick-ref.md             # Syntax, errors, special characters
│       ├── component-library.md     # @local/qk:2.2.0 API
│       ├── tool-routing.md          # Visual tool routing + examples
│       ├── templates.md             # 10 document preambles
│       ├── common-patterns.md       # Tables, show rules, large documents
│       ├── layout-patterns.md       # Spacing, figures, curve()
│       ├── math-pitfalls.md         # Math mode edge cases
│       ├── packages.md              # Popular packages with imports
│       ├── symbols.md               # sym.* arrows and symbols
│       ├── touying-guide.md         # Presentation framework
│       ├── quality-gates.md         # Rubrics and iteration limits
│       ├── visual-verification.md   # PNG rendering and spot-checks
│       └── data-driven.md           # JSON/CSV generation patterns
├── translate/SKILL.md               # Document format translation
├── learn/SKILL.md                   # Session knowledge extraction
├── deploy/SKILL.md                  # Deployment workflow
├── context-status/SKILL.md          # Context health monitoring
└── google-tasks/SKILL.md            # Google Tasks CLI reference

License

MIT