# book-to-skill

A tool that converts a technical book PDF, document folder or other structured prose into a structured agent skill, so a coding agent loads only the relevant chapter on demand instead of the whole text.

Given a PDF, folder or glob, book-to-skill distills the source into a skill package: a core file with mental models and a chapter index, one file per chapter that loads only when needed, a glossary, a file of patterns and techniques, and a cheat sheet of decision tables. Asking about one chapter's topic pulls in only that chapter rather than the whole book, which the project reports uses 24 to 51 times fewer tokens than including the entire book in context, measured on real books.

It writes the generated skill to a shared, cross-agent skills directory so tools that support the open Agent Skills standard, including GitHub Copilot CLI, Amp, Codex and OpenClaw, can discover it directly; Claude Code additionally gets a verified symlink, and Hermes Agent instead uses its own partitioned per-category skills location. The same extraction approach works on other structured prose beyond books, such as a team's internal documentation.

It is written in Python and licensed under the MIT license.

Website: https://github.com/virgiliojr94/book-to-skill
