book-distill

Distill any book into an executable Claude Code skill pack. Evaluates skillability, generates structured skills with modules/templates/examples, and supports multi-book synthesis.

What This Is

A meta-skill — it doesn't teach you a book's content, it converts a book's methods into Claude Code skills you can actually use. Feed it a title, it scores the book's "skillability," then generates a complete skill pack: SKILL.md files, modules, templates, decision frameworks, and guardrails.

Installation

In Claude Code:

/install gh:michaelpersonal/book-distill

Usage

/book-distill "Never Split the Difference"
/book-distill "Radical Candor" "Crucial Conversations"
/book-distill "The Manager's Path" --notes ~/highlights/managers-path.md
/book-distill "人性的弱点" --lang zh
/book-distill "嫌われる勇気" --lang ja
/book-distill "影响力" "Never Split the Difference" --lang en

Arguments:

• One or more book titles (required)
• --lang — output language: en (English), zh (Chinese), ja (Japanese). Auto-detected from title if omitted
• --notes — path to highlights/excerpts/notes file (optional)
• --output — output directory (optional, defaults to cwd)

Language Support

Supports English, Chinese, and Japanese — not as translation, but as native generation:

• Auto-detects language from book title, or set explicitly with --lang
• Localizes all content: methods, examples, templates, guardrails, README
• Culturally adapts: workplace examples reflect local norms (e.g., 面子 dynamics in Chinese, 根回し/報連相 in Japanese)
• Preserves original terms: key concepts show the source language term alongside the output language
• File names stay English: SKILL.md, prepare.md etc. for CLI compatibility
• Cross-language synthesis: can merge books from different languages into one skill pack

How It Works

flowchart LR
    A(["📖 Book Title(s)"])
    B["Evaluate\nskillability"]
    C["Synthesize\noverlaps"]
    D["Choose\nstructure"]
    E["Generate\nskill pack"]
    F(["✅ Done"])

    A --> B --> C --> D --> E --> F

Loading

Phase	What happens	You do
Evaluate	AI scores the book on 5 dimensions (max 25)	Validate, correct, go/no-go
Synthesize	Overlap analysis across books (if 2+)	Confirm which modules to merge
Structure	Router+modules or standalone skills?	Override if you prefer
Generate	Skill files built section by section	Review each section
Finalize	Files written, README generated, git commit	Ship it

Skillability Score

Books are scored on 5 dimensions (1-5 each, max 25):

Dimension	What it measures
Actionability	Methods vs. narrative
Modularity	Separable techniques vs. one big idea
Situational Triggers	Clear "use when X" vs. general wisdom
Concreteness	Scripts/templates vs. abstract principles
Scope Clarity	Tight domain vs. covers everything

Score	Verdict
20-25	Excellent — full skill pack
14-19	Good — solid with some thin areas
8-13	Marginal — extract strongest parts only
5-7	Poor fit — better as reference doc

Multi-Book Synthesis

When given multiple books, the skill analyzes overlap:

• Unique concepts stay in each book's pack
• Complementary concepts get synthesized into unified modules (stronger than either source)
• Redundant concepts use the stronger treatment

The output includes a shared router that maps situations to the right skill regardless of source book.

Skill Structure

book-distill/
├── skills/
│   └── book-distill/
│       ├── SKILL.md              # The meta-skill
│       └── modules/
│           ├── evaluate.md       # Skillability scoring
│           ├── generate.md       # Skill pack generation
│           ├── synthesize.md     # Multi-book overlap & merging
│           └── templates.md      # Output format templates
└── README.md

Examples Built With This Pattern

• negotiate-skills — Never Split the Difference (router+modules)
• radical-candor-skills — Radical Candor (standalone skills)