Convert technical books and documents into agent-ready skills
If you have ever struggled to get accurate answers from an agent about a technical book you read long ago, this tool addresses that issue. By pointing it at a book or document folder, it converts the content into structured skill files that the agent loads only when needed to answer with actual content. The resulting output is ready to use in various agent hosts, including GitHub Copilot CLI and Claude Code.
How book content is structured
The tool does not simply summarize books; it creates a structure composed of frameworks, decision rules, anti-patterns, and chapter-specific files. The conversion process begins by specifying a file or folder, after which the tool analyzes it to generate a skill directory containing SKILL.md, chapter markdown files, a glossary, a pattern list, and a cheat sheet. These generated skills are designed so that the agent loads only the necessary chapter files when asked about a specific topic. This eliminates the need to put the entire book into the context, reducing token usage by 24 to 51 times when measured against the actual book.
Support for non-book documents
Although named 'book-to-skill', the input can be any structured text. You can bundle internal company materials, such as internal documents, architecture decision records, runbooks, and onboarding guides, into a single skill to query them during coding. It also converts frequently referenced but hard-to-memorize materials like brand guidelines, design system documentation, research paper collections, RFCs, and API contracts. Any document you frequently open and want to keep in mind is a valid target for this tool's conversion.
Installation and requirements
The tool consists of a Python-based extractor and a spec-based generator that the agent follows. You can install it by running the npx skills add virgiliojr94/book-to-skill command or by cloning the repository and placing it in the agent skill folder. For PDF processing, it uses different extraction tools depending on whether the book is text-centric or technical. Text-centric books use pdftotext or pypdf, while technical books with lots of code or tables recommend docling; the tool automatically selects the appropriate tool based on user choice. Scanned PDFs lack a text layer and require prior OCR processing, meaning you must run ocrmypdf first. Other formats such as EPUB, DOCX, HTML, and RTF are also supported, and the tool provides installation commands if required dependencies are missing.
Copyright and usage scope
This tool does not contain book content; it is a tool for converting files that the user already owns. The extraction and analysis process occurs locally, and files are not uploaded externally. The generated skills are structured synthesis notes rather than copies of the original text, making them suitable for personal use. However, redistributing or sharing generated skills for copyrighted books may infringe on the rights holder's interests, so skills for others' books should be kept private. Internal documents, your own works, and publicly licensed materials can be shared within the scope of their respective licenses. The project itself is distributed under the MIT license, which applies only to the converter code and skill definitions.
