PDF Processing

PDF Processing Skill

When to use this skill

Use this skill when the user mentions:

• Reading or extracting content from PDFs
• Working with tables in PDF documents
• Merging, splitting, or reorganizing PDFs
• Creating new PDF documents
• Filling out PDF forms
• Processing scanned PDFs (OCR)

Progressive Disclosure Strategy

Level 1: Quick Start

When user asks about PDF:

1. Ask ONLY for the file path
2. Choose the appropriate script based on their need
3. Execute the script with minimal parameters
4. Show results

Level 2: Core Operations

Only reveal these options after initial execution or if user asks.

Level 3: Advanced Features

Only introduce when user explicitly needs them.

Available Scripts

All scripts are in the ./scripts/ directory. Execute them using bash or python.

1. Extract Text: extract_text.py

When to use: User wants to read PDF content, extract text.

Usage:

python ./scripts/extract_text.py <pdf_path> [--pages PAGES] [--output OUTPUT]

Parameters:

• pdf_path: (required) Path to PDF file
• --pages: (optional) Page range, e.g., "1-5" or "1,3,5". Default: all pages
• --output: (optional) Output file path. Default: stdout

Example:

# Extract all text
python ./scripts/extract_text.py report.pdf

# Extract specific pages
python ./scripts/extract_text.py report.pdf --pages "1-5"

# Save to file
python ./scripts/extract_text.py report.pdf --output extracted.txt

Progressive disclosure: Start with just the pdf_path, add options only if user needs them.

2. Extract Tables: extract_tables.py

When to use: User wants to extract tables, data, or spreadsheet content from PDF.

Usage:

python ./scripts/extract_tables.py <pdf_path> [--format FORMAT] [--output OUTPUT]

Parameters:

• pdf_path: (required) Path to PDF file
• --format: (optional) Output format: excel, csv, json. Default: excel
• --output: (optional) Output file path. Default: auto-generated

Example:

# Extract to Excel (default)
python ./scripts/extract_tables.py financial_report.pdf

# Extract to CSV
python ./scripts/extract_tables.py data.pdf --format csv

# Custom output path
python ./scripts/extract_tables.py data.pdf --output results/tables.xlsx

Progressive disclosure: Default to Excel format unless user specifies otherwise.

3. Merge PDFs: merge_pdfs.py

When to use: User wants to combine multiple PDFs into one.

Usage:

python ./scripts/merge_pdfs.py <output_path> <pdf1> <pdf2> [pdf3 ...]

Parameters:

• output_path: (required) Path for merged PDF
• pdf1, pdf2, ...: (required) PDFs to merge (in order)

Example:

# Merge PDFs
python ./scripts/merge_pdfs.py merged.pdf doc1.pdf doc2.pdf doc3.pdf

# Merge all PDFs in directory
python ./scripts/merge_pdfs.py combined.pdf *.pdf

4. Split PDF: split_pdf.py

When to use: User wants to split PDF into separate files.

Usage:

python ./scripts/split_pdf.py <pdf_path> [--mode MODE] [--output-dir OUTPUT_DIR]

Parameters:

• pdf_path: (required) Path to PDF file
• --mode: (optional) Split mode: pages (one per file), range. Default: pages
• --output-dir: (optional) Output directory. Default: ./split_output

Example:

# Split into individual pages
python ./scripts/split_pdf.py document.pdf

# Split into custom directory
python ./scripts/split_pdf.py document.pdf --output-dir pages/

5. Create PDF: create_pdf.py

When to use: User wants to create a PDF from text or content.

Usage:

python ./scripts/create_pdf.py <output_path> [--title TITLE] [--content CONTENT] [--input INPUT_FILE]

Parameters:

• output_path: (required) Path for new PDF
• --title: (optional) Document title
• --content: (optional) Text content (for short content)
• --input: (optional) Text file to read content from

Example:

# From text content
python ./scripts/create_pdf.py report.pdf --title "Monthly Report" --content "Report content here..."

# From text file
python ./scripts/create_pdf.py report.pdf --title "Report" --input content.txt

6. Fill PDF Form: fill_form.py

When to use: User needs to fill out a PDF form programmatically.

Usage:

python ./scripts/fill_form.py <form_path> <data_file> <output_path>

Parameters:

• form_path: (required) Path to PDF form
• data_file: (required) JSON file with field data
• output_path: (required) Path for filled PDF

Example:

# Fill form with data from JSON
python ./scripts/fill_form.py application.pdf data.json filled_application.pdf

See: ./references/form-filling.md for data format details.

7. OCR PDF: ocr_pdf.py

When to use: User has scanned/image-based PDF and needs searchable text.

Usage:

python ./scripts/ocr_pdf.py <pdf_path> [--lang LANG] [--output OUTPUT]

Parameters:

• pdf_path: (required) Path to scanned PDF
• --lang: (optional) OCR language code. Default: eng
• --output: (optional) Output PDF path. Default: _ocr.pdf

Example:

# OCR English document
python ./scripts/ocr_pdf.py scanned.pdf

# OCR with different language
python ./scripts/ocr_pdf.py document.pdf --lang chi_sim

Note: Requires tesseract-ocr installed on system.

Interaction Guidelines

✅ Good Interaction Pattern

User: "I need to read a PDF"
Agent: "I'll help you extract the text. What's the PDF file path?"
User: "report.pdf"
Agent: [Executes] python ./scripts/extract_text.py report.pdf
Agent: "✅ Extracted text from 5 pages. Here's a preview: [shows first 500 chars]
       Would you like to see the full text or extract tables?"

❌ Poor Interaction Pattern

User: "I need to read a PDF"
Agent: "I can extract text, tables, images, metadata. For text I can preserve layout or not, 
       use different encodings, extract specific pages using ranges or lists, 
       output to txt/docx/html. For tables I need to know format (Excel/CSV/JSON), 
       encoding, empty cell handling... Please specify all parameters."

Decision Tree for PDF Tasks

User mentions PDF
  ├─ "read", "extract text", "content" → execute extract_text.py
  ├─ "table", "data", "spreadsheet" → execute extract_tables.py
  ├─ "merge", "combine", "join" → execute merge_pdfs.py
  ├─ "split", "separate", "break" → execute split_pdf.py
  ├─ "create", "generate", "make" → execute create_pdf.py
  ├─ "form", "fill out", "application" → execute fill_form.py
  └─ "scanned", "image", "OCR" → execute ocr_pdf.py

Error Handling

If a script fails:

1. Show the error message clearly
2. Suggest common fixes (file not found, missing dependencies)
3. Don't repeat the same command - try alternatives or ask for clarification

Common errors:

• FileNotFoundError: Check if file path is correct, suggest listing directory
• ImportError (pdfplumber, pypdf, etc.): Install missing package
• No tables found: Suggest OCR if PDF might be scanned
• Permission denied: Check file permissions

Dependencies

Scripts require these Python packages:

pip install pdfplumber pypdf reportlab pandas openpyxl pytesseract pdf2image

System dependencies:

# Ubuntu/Debian
sudo apt-get install tesseract-ocr poppler-utils

# macOS
brew install tesseract poppler

Reference Documents

• ./references/form-filling.md: Detailed guide for PDF form data format
• ./references/ocr-guide.md: OCR best practices and language codes
• ./references/troubleshooting.md: Common issues and solutions

Examples

See ./examples/ directory for complete workflow examples:

• example_extract_report.sh: Extract text and tables from report
• example_merge_quarterly.sh: Merge quarterly PDFs
• example_batch_process.sh: Process multiple PDFs

Key Principles

1. Scripts are self-contained: Each script handles one task completely
2. Minimal parameters: Use smart defaults, accept overrides
3. Clear output: Scripts print success/error messages with emojis
4. Exit codes: 0 for success, non-zero for errors
5. Help text: All scripts support --help flag

Tips for Agent

• Start simple: Use minimal parameters first
• Show progress: For long operations, show that work is happening
• Be helpful: Suggest next logical steps after each operation
• Learn preferences: If user always wants CSV, remember that
• Handle errors gracefully: Don't just show stack traces, explain the problem

Skill Maintenance

This skill follows the Agent Skills standard:

• name and description in frontmatter enable discovery
• SKILL.md provides full instructions for activation
• scripts/ contain executable code for execution
• Progressive disclosure reduces cognitive load

Version: 1.0.0 Last updated: 2025-02-02