A standalone Agent Skill for source-grounded reading of nonfiction ebooks. It works across nonfiction, with focused checks for AI, science, technology, and philosophy.
The skill offers a preliminary positioning card when only a title or table of contents is available. With readable book text, it builds a map of the central question, the role and relationship of major parts, evidence or derivations, conclusions, and limits. Selected notes retain source locations and distinguish the author's words from the reader's own thoughts. It can add short closed-book recall questions.
Its local extractor accepts PDF, EPUB, TXT, Markdown, and HTML, and writes source.md plus manifest.json. PDF annotations can be recovered; separate platform exports of highlights or notes can be supplied as companion material. See SKILL.md for agent behavior and extraction details for format caveats.
python3 -m pip install -r requirements.txt
python3 scripts/extract_book.py /path/to/book.epub --output /path/outside/repo/book-packageThe output directory must not exist and must be outside this skill repository. The package may contain the book's full text, so keep it in a user-chosen location. TXT and Markdown extraction use the Python standard library; other formats need only their relevant listed packages. Image-only PDFs require a separate OCR source, and figures or complex layouts require review.
The analysis can be used directly in conversation or saved as analysis.md. The extraction package keeps a source hash, locations, and coverage information for later review. The user supplies the book and any separate highlight exports; notes with uncertain ownership are labeled accordingly.