PDF Extraction (auto text / OCR)

Extract content from PDFs without deciding whether each page is selectable text or a scan.

Page type	Tool
Native text	pdfplumber (text + tables)
Scanned / image	PyMuPDF + Tesseract OCR

Install

# System OCR engine (required for scanned pages)
# Ubuntu/Debian:
sudo apt install tesseract-ocr tesseract-ocr-eng
# Optional Traditional Chinese:
# sudo apt install tesseract-ocr-chi-tra

# CLI (from this skill folder)
pip install -e .
# or: python3 -m pip install -e .

Quick start

# Auto-detect text vs OCR per page → print text
pdf-extract document.pdf

# Write to file
pdf-extract document.pdf -o out.txt

# See which mode each page will use
pdf-extract document.pdf --analyze-only

# Tables + Markdown
pdf-extract document.pdf --tables --format markdown -o out.md

# JSON (includes per-page mode)
pdf-extract document.pdf --format json -o out.json

# Force mode
pdf-extract scan.pdf --mode ocr --ocr-lang eng
pdf-extract text.pdf --mode text

# Page range
pdf-extract doc.pdf --pages 1-3,5

# Also: python -m pdf_extract document.pdf

Auto mode rules

For each page (--mode auto, default):

1. Try native text via pdfplumber; count chars and embedded images
2. Enough text → text
3. Sparse text (default < 40 non-whitespace chars) or image-heavy → ocr

Tune with --min-text-chars. Override with --mode text or --mode ocr.

Agent usage

When the user provides a PDF and wants content extracted:

1. Prefer pdf-extract <file> (auto mode). Do not ask whether it is text or scanned.
2. Use --analyze-only if you only need routing diagnostics.
3. Use --tables when tables matter (works best on native-text pages).
4. For Chinese scans, set --ocr-lang eng+chi_tra if chi_tra is installed.
5. Surface stderr summary lines (which pages used text vs ocr) when useful.

CLI reference

Flag	Meaning
-o, --output	Write to file (default stdout)
-f, --format	text | json | markdown
--mode	auto | text | ocr
--pages	e.g. 1-3,5
--tables	Extract tables on text pages
--layout	Preserve native text layout
--meta	Include PDF metadata
--ocr-lang	Tesseract langs (default eng)
--ocr-dpi	OCR render DPI (default 200)
--analyze-only	Classification JSON only
-q, --quiet	Suppress progress on stderr

Dependencies

• Python 3.10+: pdfplumber, pymupdf, Pillow (see pyproject.toml)
• System: tesseract (and language packs as needed)