book-to-skill turns a book or document into an agent skill. It extracts the author’s named frameworks, principles, techniques and anti-patterns rather than writing a summary, and saves them as a skill your agent can load while you work.
It reads local PDF, EPUB, DOCX, HTML, Markdown, TXT, RTF and MOBI/AZW files, or searches for a book through a connected book MCP server. It works with GitHub Copilot CLI, Amp, Claude Code, Hermes Agent, OpenCode and OpenClaw.
What You Get
- A generated skill folder with a master SKILL.md, a chapters folder, glossary.md, patterns.md and cheatsheet.md
- Chapter summaries that list the frameworks, key concepts, mental models, anti-patterns and takeaways of each chapter
- A chapter index and a topic index in the master SKILL.md, so the agent reads only the chapter a question needs
- An analyze-only report of frameworks, principles and techniques, for review before any skill files are generated
Common Use Cases
- Turn a technical book PDF into a skill you can query while coding
- Study a book by asking the agent questions answered from its chapters
- Apply an author’s frameworks to a work task using the generated skill
- Update an existing book skill with new source files or a revised edition
What Makes It Unique
The skill is built around extracting structure instead of producing a summary: named frameworks keep the author’s exact formulation, and large books get reference files and on-demand chapters. It supports five modes (full conversion, analyze only, generate from prior analysis, update of an existing skill, and search through an MCP book provider) and installs into the skill folders of several agent hosts from one converter. It also offers to publish the generated skill to GitHub, private by default.
Under The Hood
The master SKILL.md is a long step-by-step procedure the agent follows. It runs scripts/extract.py through python3, which imports the book_to_skill package and writes full_text.txt and metadata.json into a per-run work directory. Extractors depend on the format: pdftotext, pypdf, pdfminer.six or docling for PDFs, ebooklib or the standard zip library for EPUB, and optional packages for DOCX, HTML and RTF. The extractor can offer to install missing Python packages, which reaches the network, and it asks first. The agent then writes the skill files into your skills folder, creates a symlink under the Claude Code skills folder, scans the output with tools/scan_generated_skill.py and deletes its work directory. The search mode talks to a book MCP server such as libgen-mcp, reads book text through it, and only helps install that server after you confirm. The optional publish step uses the gh CLI and git to create a GitHub repository. The scan of this repository flagged that shell, file-write, git and network use is not declared in the SKILL.md frontmatter, and flagged characters that render as nothing in book_to_skill/utils.py and CHANGELOG.md. It also flagged credential handling in a test mock, text in a test for the scan tool that tells a model to ignore its instructions, and the file docs/CNAME, which was not read. The last three sit in test and docs files, not in the steps the skill runs.