// the find
virgiliojr94/book-to-skill
Turn any technical book PDF into a Claude Code skill — ready to study, reference, and use while you work.
Converts books and documents (PDF, EPUB, DOCX, etc.) into structured markdown 'skill' files that AI coding agents like Claude Code and GitHub Copilot CLI can load on demand. The core idea is to pay the structuring cost once at conversion time rather than burning tokens navigating raw PDFs repeatedly. Aimed at developers who want their reference books queryable inside their agent workflow without context-stuffing.
- The token efficiency claim (24x-51x fewer tokens vs context-dumping) has a documented measurement methodology in docs/performance.md, not just a marketing number — they explain the 'discovery loop tax' concept and how they measured it on real books.
- Format support is practical and tiered well: it probes for available tools and gracefully degrades rather than hard-failing, and `python3 scripts/extract.py --check` gives you an instant diagnostic of what's installed.
- Test coverage is notably wide for a utility tool — 20+ test files covering edge cases like CJK supplementary planes, BiDi control characters, unbalanced code fences, and scanned PDF detection, which suggests actual real-world breakage drove the test suite.
- The output format is an open standard (Agent Skills / SKILL.md) that works across multiple agent hosts, avoiding vendor lock-in to a single tool.
- The actual 'skill generation' step — turning extracted text into the structured SKILL.md + chapter files — is delegated to the agent itself following a spec, not done deterministically by the Python code. This means output quality is entirely dependent on your LLM and the quality of your prompts, which is a significant hidden dependency the README underplays.
- Quality Rule #7 ('never copy raw passages') is enforced only by prompt instruction, not by code. If the model drifts or you use a weaker model, you could inadvertently generate a near-verbatim reproduction of copyrighted content with no technical guardrails.
- MOBI/AZW support requires Calibre as an external dependency, which is a 200MB desktop application — a heavy hammer for what should be a CLI conversion step, and it won't work in headless/server environments.
- The 22k stars suggest a viral moment rather than sustained adoption — the project is at v1.4.0 but the forks-to-stars ratio (~1:10) is low for a tool people are actively building on, suggesting many starred-and-forgot rather than integrated it into workflows.