PDF Reader Skill

Extract and read PDF files. Two modes: chunked stdout (read-as-you-go, costs context per page) and full-dump to a markdown file (token-cheap, re-readable later via Read/Grep).

Setup (One-time)

pip install pymupdf

SKILL_DIR below = {home}/.claude/skills/tool-pdf-reader/ on the user's machine. Reference the script with that absolute path.

Usage

1. Get PDF Info

python "<SKILL_DIR>/pdf_tool.py" info "path/to/file.pdf"

Returns: total pages, title, author, file size.

2. Extract Text (Page Range, stdout)

python "<SKILL_DIR>/pdf_tool.py" extract "path/to/file.pdf" --start 1 --end 10

• --start: Starting page (1-indexed, default: 1)
• --end: Ending page (inclusive, default: 10)
• Recommended: Read 10-20 pages at a time
• Use when: you only need a few pages, or you're triaging which sections to dump.

3. Extract Table of Contents

python "<SKILL_DIR>/pdf_tool.py" toc "path/to/file.pdf"

4. Search Text

python "<SKILL_DIR>/pdf_tool.py" search "path/to/file.pdf" "keyword"

Returns pages containing the keyword.

5. Dump Full PDF to Markdown File (token-saver)

python "<SKILL_DIR>/pdf_tool.py" dump "path/to/file.pdf" "path/to/output.md"

Writes the entire PDF to a flat markdown file with ## Page N headers, faithful to the source. Blank pages are noted but not skipped (so page numbers stay aligned).

• Use when: you (or future-you in a new session) will want to reference the PDF more than once. Dump once, then Read / Grep the .md instead of re-parsing the PDF.
• Use when: the PDF is long (>30 pages) and you need most of it but not in stdout — saves context vs. iterated extract calls.
• Skip when: the PDF is mostly figures/scans (no extractable text), or you only need 1–2 pages.

For a vault PDF, dump beside it under a parallel directory:

{vault}/99_Attachments/cdegs/Foo.pdf
{vault}/30_Resources/Electrical-Engineering/cdegs-extracted/Foo.md       # dump output
{vault}/30_Resources/Electrical-Engineering/cdegs-extracted/Foo-digest.md  # your hand-written digest

---

Workflow for Reading a Book

1. Get info first to know total pages:

   python pdf_tool.py info "book.pdf"
   

2. Extract TOC to understand structure:

   python pdf_tool.py toc "book.pdf"
   

3. Read in chunks (10-20 pages per request):

   python pdf_tool.py extract "book.pdf" --start 1 --end 15
   python pdf_tool.py extract "book.pdf" --start 16 --end 30
   # ... continue as needed
   

4. Search for specific topics:

   python pdf_tool.py search "book.pdf" "emotion"
   

---

Common PDF Locations

Type	Path
Books	99_Attachments/books/
Papers	99_Attachments/papers/
Downloads	{home}/Downloads/

---

Trigger Phrases

User Says	Action
"读这个 PDF" / "read this PDF"	Get info → TOC → Extract chunks
"PDF 有多少页" / "how many pages"	Run info command
"提取第 X-Y 页" / "extract pages X-Y"	Run extract --start X --end Y
"搜索 PDF 中的 X" / "search PDF for X"	Run search command
"save the PDF as md" / "dump to markdown" / "extract whole PDF to file" / "save it programatically"	Run dump command, write .md next to source under a *-extracted/ parallel dir

---

Notes

• Text extraction quality depends on PDF type (scanned vs. native text)
• For scanned PDFs, consider OCR tools (not included in this skill)
• Large PDFs should always be read in chunks to preserve context window