You read the book. Then you forgot chapter 7 existed.
If you're a student or a business owner who reads a lot, you know this pattern: you finish a great technical book, and three months later you can't tell someone what was in the back half of it.
The usual fixes don't work. Searching the PDF gives you a list of pages, not an answer. Asking your AI about the book gets you a hallucinated summary, or it just tells you it doesn't have the content. Taking notes while you read gets you a 200-line doc you never open again.
This free plugin fixes it by saving the book as a Claude skill. Instead of your AI either forgetting the book or rereading the whole thing every time you ask something (and blowing through your context in the process), it just flips straight to the right chapter and answers from the real content. Over 12,000 people are already using it.
book-to-skill, in one sentence
It's a free, open-source agent skill (MIT licensed) that reads a book, PDF, or folder of documents, and distills it into a structured skill your AI loads on demand: core mental models, per-chapter files, a glossary, a pattern list, and a cheatsheet. It works in Claude Code, GitHub Copilot CLI, and Amp, since they all read the same open Agent Skills format.
1. One command, two minutes
If you're on Claude Code, run this in your terminal. It clones the skill straight into your skills folder, and Claude Code picks it up next session, no restart or config needed.
git clone https://github.com/virgiliojr94/book-to-skill ~/.claude/skills/book-to-skill On GitHub Copilot CLI, clone it to ~/.copilot/skills/ instead, then run /skills reload. On Amp or cross-agent, use ~/.agents/skills/. Claude Code and Amp pick the skill up automatically on your next session.
2. Run the command
In a fresh chat, run the skill against a file, a folder, or a glob pattern, and optionally give it a short slug name. That slug becomes the command you'll use to talk to the book later.
/book-to-skill ~/books/your-book.pdf your-book-slug Claude will ask you one thing first: is this book technical (code, tables, formulas) or text-heavy (mostly prose)? That answer decides which extractor it uses. Text-heavy books extract instantly. Technical books take longer (roughly 1.5 seconds a page) because it's preserving tables and code blocks instead of flattening them into plain text.
What you need installed, by format
Plain text and Markdown need nothing. Everything else has a best-quality tool and a built-in fallback, so you're never fully blocked. Run python3 scripts/extract.py --check inside the cloned folder any time to see exactly what's installed and what command fixes what's missing.
| PDF (prose-heavy) | pdftotext or pypdf | Instant. No setup needed on most machines. |
| PDF (technical, tables/code) | docling | `pip3 install docling` — slower, but keeps tables and code readable. |
| EPUB | ebooklib + beautifulsoup4 | Falls back to a built-in reader with zero installs if you skip this. |
| DOCX / HTML / RTF | python-docx / beautifulsoup4 / striprtf | Each has a built-in fallback if you don’t want to install anything. |
| TXT / Markdown / AsciiDoc | nothing to install | Works out of the box. |
| MOBI / AZW / AZW3 | Calibre | The only format that needs a separate app, not a pip install. |
Five files, one skill
When it finishes, you've got a full skill folder sitting where your agent already looks for skills. The chapter files don't count against anything until you actually ask about that chapter, so the skill stays cheap to keep around.
- SKILL.md The core mental models plus a chapter index. This is what loads first, every time.
- chapters/ch01-*.md ... One file per chapter. Only loads when you actually ask about that chapter.
- glossary.md Every key term from the book, alphabetized, with a pointer to the chapter it came from.
- patterns.md Every technique, framework, or design pattern the book teaches, pulled out on its own.
- cheatsheet.md Decision tables and quick-reference rules you can scan in seconds.
How you'll actually use this
Once the skill exists, you talk to it like any other skill in your agent. No re-running the install, no reprocessing the book.
/your-book-slugloads the core mental models, for a quick refresh./your-book-slug replicationfinds and explains a specific topic, wherever it lives in the book./your-book-slug ch05dives straight into one chapter.- Use it while you study, while you reference a specific part of a text, or while you work and need to check something without stopping to dig through the PDF.
Beyond books
The name says "book," but the input is any structured prose. If you re-open a document often enough to wish you'd memorized it, it's a candidate. That includes internal documentation, brand or voice guidelines, a stack of research papers plus your own notes, or specs and contracts you reference but never memorize.
Feed it multiple sources at once. You're not limited to one file. Point it at several related files, or a whole folder, and it merges them into one unified skill.
Update it instead of rebuilding it. When new material comes in on the same topic, point the command at the existing skill folder and it folds the new content in rather than starting over.
Pick the right tool for the job. If you genuinely need to search across dozens of unrelated books ("find me the part that mentions X"), that's a job for a RAG tool. This is built for going deep on one book, or a tight cluster of related sources you keep coming back to.
Troubleshooting
- "It says a tool is missing."
Run
python3 scripts/extract.py --checkfrom inside the cloned folder. It tells you exactly which format is missing an extractor and the exact install command to fix it. - "The chapters didn't split up right."
Auto-detection needs explicit headings like "Chapter 5." Books that only use section titles or roman numerals won't auto-segment. Extraction and conversion still work fine, you just point it at sections manually instead.
- "Should I pick technical or text-heavy?"
If the book has code blocks, tables, or formulas you'd actually want preserved, say technical. If it's mostly straight prose, say text-heavy. When unsure, text-heavy is faster and you can always redo it.
- "This feels like a lot for one book."
It's a two-minute install and one command per book. Full extraction and skill generation typically costs under two dollars in API usage per book, and you only pay it once.