Book2Skills: turn books into agent skills that actually work

Book2Skills is an open-source project that distills classic book methodologies into structured AI agent skills. Each skill is a folder with a SKILL.md core file and reference documents — the agent reads the framework, applies the actual decision rules from the book, and produces reasoning traceable to specific chapters. The project is on GitHub under MIT license, currently released with Philip Fisher’s Common Stocks and Uncommon Profits, with more on the roadmap.

Telling an AI to “act like a value investor” produces a generic blend of everything it has ever read about value investing. The result is smooth, agreeable, and useless when you need a specific decision: at what P/E ratio do you walk away? What counts as a sufficient margin of safety? The AI cannot answer because “act like” is not a knowledge injection. It is a probabilistic approximation based on training data.

How book2skills works

The project converts books into skills following a standard format:

skills/your-skill-name/
├── SKILL.md              # Core file with YAML frontmatter + main framework
└── references/           # Detailed reference files, loaded on demand
    ├── framework.md
    ├── evaluation-criteria.md
    ├── common-mistakes.md
    └── ...

SKILL.md contains the core methodology: decision processes, evaluation criteria, what the author advises against. References hold deeper content loaded only when the agent needs it. Each claim is traceable to a specific chapter — no invented frameworks, no paraphrased approximations.

The quality standards are specific: faithful to source (annotated by chapter), actionable (produces specific judgments, not generalities), clear structure (headings, tables, lists), SKILL.md under 500 lines. Contributors must read the full original book — summaries and secondhand materials are rejected.

Skill vs prompt

Approach	What the AI receives	Result
Prompt engineering	”Act like Philip Fisher”	Generic impression from training data
book2skills skill	15-point scuttlebutt framework + evaluation criteria + chapter-annotated rules	Specific buy/sell judgments with traceable reasoning

Prompt engineering is cheaper and works for broad tasks. Skill injection is better when the decision framework has nuance that training data cannot capture. Fisher’s “scuttlebutt” method — talk to competitors, suppliers, and former employees before buying a stock — is a specific investigative process. An AI prompted to “do scuttlebutt research” will produce a plausible-sounding but generic investigation plan. An AI that reads the skill applies the actual 15-point checklist from the book.

Where the project is now

Currently released: Common Stocks and Uncommon Profits by Philip Fisher. Roadmap includes The Intelligent Investor (Graham), One Up On Wall Street (Lynch), Poor Charlie’s Almanack (Munger), Good to Great (Collins), and more in investing, business thinking, and mental models.

The library is small because each skill requires reading the full book and distilling it into an agent-readable format. Quality over quantity is explicit in the contribution standards.

The project accepts contributions. Anyone can submit a book request via GitHub Issues, or contribute a new skill by following the CONTRIBUTING guide: read the full book, create the skill folder, test it in Claude.ai or Claude Code, and submit a PR.

Book2Skills does not make your AI smarter. It replaces approximation with citation. Whether that matters depends on how often your AI’s “good enough” answer turns out not to be.

Installation

The skills follow the Agent Skills open standard and work with Claude.ai, Claude Code, Cursor, VS Code, and compatible tools.

# Clone the repo
git clone https://github.com/simbajigege/book2skills.git

# Copy a skill into your project
cp -r book2skills/skills/common-stocks-uncommon-profits .claude/skills/

# Or download the .skill file from Releases

For Claude.ai: Settings → Features → Custom Skills → Upload the .skill file. For any platform that supports Agent Skills: copy the skill folder to the platform’s skills directory.

The hosted library at book2skills.com/zh/ provides a browsable web version of the same skills with additional collection features.

GitHub: https://github.com/simbajigege/book2skills Website: https://book2skills.com/zh/ License: MIT (code), skill content for personal learning and research