The Book-to-Skill Playbook — Turn Any Book Into an AI Skill That Never Forgets
The companion to the video. book-to-skill (github.com/virgiliojr94/book-to-skill, MIT, 16,528 GitHub stars as of Aug 5, 2026) converts any book or document you own — PDF, EPUB, DOCX, Markdown, HTML, RTF, MOBI — into an Agent Skills-standard skill that works in Claude Code, GitHub Copilot CLI, and Amp: a ~4K-token SKILL.md core that is the only part always loaded, one ~1K-token file per chapter pulled in on demand, plus glossary, patterns, and cheatsheet. INSIDE: (1) THE EXACT INSTALL — one git clone per agent (~/.claude/skills, ~/.copilot/skills + /skills reload, ~/.agents/skills) and the /book-to-skill command with folder, glob, and multi-source forms. (2) THE 4 SETUP QUESTIONS AND HOW TO ANSWER THEM — technical vs text-heavy (Docling at ~1.5s/page preserves code and tables vs instant pdftotext), proceed vs analyze-only, the purpose question ('All of the above' generates study-depth chapters), the slug — plus update mode to fold new material into an existing skill. (3) OUR MEASURED RECEIPTS — Think Python 2, 244 pages: 82,237 words / ~109K tokens / 19 chapters auto-detected in under a second, 4 parallel agents, ~15 minutes hands-off, ~33K-token skill; the repo's own tools/discovery_tax.py scored context-dumping at 109,649 tokens per turn vs ~5,000 through the skill = 21.9x fewer on our machine (the project's own benchmarks: 24x-51x). (4) 20 THINGS TO ABSORB FIRST — a concrete checklist from free CC-licensed books and arXiv paper clusters to ADRs, runbooks, API specs, and brand guidelines. (5) THE HONEST CATCHES — chapter detection needs explicit 'Chapter N' headings (Pro Git's 501 pages won't auto-segment), conversion costs about $1 of tokens per book once, and the copyright rule: a skill from someone else's book is private study notes, never something you publish or share. Every receipt from our own run Aug 5, 2026. Independent playbook from Hyperautomation Labs — not affiliated with the project.
Free. No spam. Unsubscribe anytime.