Saturday, October 17, 2026Scoutari
Log inSign up
Sign up
ModelsAPI pricingSkill RadarSavedPersonalizeSubmit sourceLog in
Back to timeline

Sat, October 1700:00ToolsAPI & pricingMCP & SkillsAI codingAPI & pricing guide

book-to-skill: Turn technical book PDFs into Agent skills, cutting token usage by 24–51x

View original

Decision Brief

What changedThe open-source tool book-to-skill converts structured text like technical book PDFs and documentation tables of contents into cross-Agent skills that load chapters on demand to answer questions.
Why it mattersIt pays the structuring cost of a technical book once at conversion time, so later queries load only relevant chapters—using 24–51x fewer tokens than stuffing the whole book into context—making it ideal for developers who frequently consult technical books and internal docs.
Who should careAI coding tool users
Affected stackClaude CodeOpenAICopilot
Source confidenceHigh · Official release / blog / repo

book-to-skill is a converter built for the open Agent Skills standard. Given input in PDF, EPUB, DOCX, or HTML, it generates a SKILL.md, per-chapter files, a glossary, a pattern library, and a cheat sheet. SKILL.md is about 4,000 tokens, each chapter file about 1,000 tokens, loaded on demand. A deterministic Python extractor processes documents, then an Agent generates structured skills per the spec, output to cross-Agent directories like ~/.agents/skills/, supporting GitHub Copilot CLI, Amp, Claude Code, Codex, Hermes Agent, OpenCode, and OpenClaw. The tool targets developers who don't want to keep flipping through PDFs, teams turning internal docs into queryable skills, and engineers referencing specs or papers. It includes no book content, runs entirely locally, and produces structured derivative notes rather than verbatim copies—so skills generated from third-party books should be kept private.

Summary basis: full article readCompiled from the source scope noted above; the original remains authoritative.

Sources

Related intel

留言

登入后即可留言，和其他 builder 交换实测心得。

还没有留言，抢头香。