Turn Your Textbook Into a Claude Code Skill

An AI agent is only as good as what it can reach. Ask it about a specific textbook — your course’s actual textbook, with its exact definitions and notation — and it will paraphrase from memory, blur the details, and sometimes make things up. What if the agent could read the book itself, structured so it knows where every chapter and formula lives?

That is what book-to-skill does: it turns a PDF textbook into a Claude Code skill, so your agent reasons from the real material instead of a fuzzy recollection of it. It is free, open-source (about 24k stars on GitHub), and runs entirely on your machine.

What it produces

Point it at a PDF and it generates a proper Claude Code skill folder:

• a SKILL.md — the entry point Claude reads first, describing what the book covers and when to use it;
• per-chapter files — the content broken up so the agent can pull just the relevant chapter instead of the whole book;
• a cheatsheet — the key definitions, formulas, and takeaways condensed for quick reference.

Once installed, your agent treats the book like knowledge it owns: ask a question, it opens the right chapter, and answers from the source.

Install and run (one command each)

You need Node installed. Add the skill:

npx skills add virgiliojr94/book-to-skill

Then hand it your PDF:

/book-to-skill ./book.pdf

That is the whole flow. It processes the PDF locally — no upload, no API round-trip for the extraction — and writes the generated skill into your skills directory, ready to use in your next session.

Where this is genuinely useful

• Study. Convert your open-licence course textbook into a skill and ask the agent to quiz you, explain a proof, or trace a definition to its chapter — grounded in the actual book, not the internet’s average of it.
• Reference at work. A dense standard, a protocol spec, an open manual — anything you keep re-reading — becomes something your agent can cite chapter-by-chapter.
• Teaching. Build a skill per unit and let students ask questions that answer from the assigned reading.

The honest limits — read this part

Two things you must know before you run it.

1. It is slow on big books. Extraction runs at roughly 1.5 seconds per page. That is fine for a 40-page chapter; a 900-page textbook is a coffee-and-then-some job. Start with a single chapter or a slim book to see the output before you commit a doorstop to it.

2. Only feed it a book you are allowed to. This is the one that matters. Generating a skill from a copyrighted, paid textbook makes a derivative copy of it on your disk — do not do that. Use an open-licence book and name the licence when you share the result:

• OpenStax textbooks (CC-BY) — free, peer-reviewed, and huge across math, science, and CS.
• Anything explicitly CC-BY or CC-BY-SA.
• Public-domain works (older classics whose copyright has expired).

If you cannot state the licence, do not convert the book. The tool is free and legal; keeping it legal is on the input you give it.

Used right, this is one of the highest-leverage free tools out there: one command turns a book you are allowed to use into an agent that has actually read it.