pdf-reader

# PDF Reader Extract structured markdown from PDF files using pymupdf4llm. Two approaches depending on content type. ## Primary: pymupdf4llm (structured markdown) Best for text-heavy documents, articles, and slide decks where you need parseable content. ```bash # Full document ~/.claude/skills/pdf-env/bin/python3 -c "import pymupdf4llm, sys; print(pymupdf4llm.to_markdown(sys.argv[1]))" "/path/to/file.pdf" # Specific pages (0-based) ~/.claude/skills/pdf-env/bin/python3 -c "import pymupdf4llm, sys; print(pymupdf4llm.to_markdown(sys.argv[1], pages=[int(p) for p in sys.argv[2].split(',')]))" "/path/to/file.pdf" "0,1,2" ``` Output is clean markdown with headers detected by font size, tables preserved where possible, and text flow maintained. ## Fallback: Read tool multimodal (visual layout) Best for PDFs where visual layout matters (diagrams, charts, infographics, heavily visual slide decks). ``` Read(file_path="/path/to/file.pdf", pages="1-5") ``` - Renders pages as images (multimodal reading) — preserves visual layout - Requires poppler (`brew install poppler`) - Max 20 pages per request - Uses more tokens but captures what text extraction misses ## When to use which | Content type | Best approach | |-------------|---------------| | Text-heavy documents | pymupdf4llm | | Slide decks with text | pymupdf4llm first, Read fallback if diagrams lost | | Visual/diagram-heavy | Read tool (multimodal) | | Large PDFs (100+ pages) | pymupdf4llm with page ranges | ## Known limitations of pymupdf4llm - Nested lists lose hierarchy - Tables without borders may parse as plain text - No slide-awareness (treats each page as a text page) - Links can produce unexpected results (entire line becomes hyperlink) ## Wrapper note processing When processing PDF wrapper notes from `Inbox/`: 1. Extract content using pymupdf4llm 2. Write a concise summary into the wrapper note's `## Summary` section 3. Update frontmatter (categories, areas, tags) 4. Move wrapper note to its permanent location (`Notes/` or `References/`)