What it does

book-to-skill ingests a technical book or document (PDF, EPUB, DOCX, and five other formats) and generates a structured Claude Code skill in ~/.claude/skills/. The output includes a master SKILL.md with core mental models, per-chapter summaries, a glossary, a patterns file, and a cheatsheet. You then type /your-book-slug <topic> and Claude loads only the relevant chapter on demand.

The interesting bit

The project treats “compile time” differently from RAG. Instead of chunking and embedding at query time, it runs one deep analysis pass that extracts the author’s actual frameworks, names them, and structures them for reasoning. The README is admirably direct about the trade-off: RAG wins for searching 50 books; this wins for having one author’s mental models sit next to you while you work.

Key highlights

• Supports 8 document formats with a cascading fallback chain (e.g., pdftotext → PyPDF2 → pdfminer for PDFs, ebooklib → stdlib zipfile for EPUBs)
• Chooses extraction strategy by book type: “technical” routes to Docling (~1.5s/page, preserves tables and code blocks), “text-heavy” routes to faster prose extractors
• Chapter files load on-demand, so a 400-page book doesn’t burn your token budget upfront
• Explicitly designed for books Claude doesn’t know: niche references, internal docs, recent publications, translated works
• Install is two curl commands into ~/.claude/skills/

Caveats

• Requires Claude Code; not a standalone tool or general LLM wrapper
• Docling extraction is slow: 164 seconds vs. 0.1 seconds for pdftotext on the benchmarked 103-page technical book
• The “effort: high” badge in the README is honest — this is not a one-click magic converter

Verdict

Worth it if you have one or two technical books you actually want to use while coding, not just say you read. Skip it if you’re looking to search across a whole library; the FAQ itself sends you to NotebookLM for that job.