pdf-processing（PDF processing 文档）

PDF Processing Skill

Overview

This skill provides comprehensive PDF processing capabilities leveraging Claude's native PDF support, which can:

• Extract and parse text content
• Identify and extract tables with structure preservation
• Analyze charts, graphs, and visual elements
• Understand document layout and formatting
• Generate summaries and insights
• Convert PDFs to various formats (Markdown, JSON, CSV, plain text)

When to Use

Claude should automatically activate this skill when:

• User provides a PDF file path or wants to process a PDF
• User asks to extract text, tables, or data from PDFs
• User requests PDF analysis, summarization, or conversion
• User needs to understand charts, diagrams, or visual content in PDFs
• User wants to transform PDF content to another format

Key Capabilities

1. Text Extraction

• Full document text extraction with formatting preservation
• Page-by-page text extraction
• Section and paragraph identification
• Header and footer detection

2. Table Extraction

• Automatic table detection and extraction
• Structure-preserving conversion to CSV, JSON, or Markdown
• Multi-page table handling
• Cell merging and complex table support

3. Visual Content Analysis

• Chart and graph interpretation
• Diagram and flowchart understanding
• Image and figure description
• Infographic analysis

4. Document Understanding

• Layout analysis and structure detection
• Multi-column text handling
• Form field identification
• Metadata extraction (title, author, creation date, page count)

5. Format Conversion

• PDF to Markdown (preserving headings, lists, tables)
• PDF to JSON (structured data extraction)
• PDF to CSV (table extraction)
• PDF to plain text

Instructions

Step 1: Validate PDF Input

First, determine how the PDF is provided:

Option A: File Path

# Verify the file exists and is a PDF
ls -lh /path/to/document.pdf
file /path/to/document.pdf

Option B: Base64-Encoded PDF If the user provides base64-encoded content, save it first:

# Decode and save base64 PDF
python .claude/skills/pdf-processing/scripts/decode_pdf.py --input base64_string.txt --output document.pdf

Option C: URL If the user provides a URL, download it:

# Download PDF from URL
python .claude/skills/pdf-processing/scripts/download_pdf.py --url "https://example.com/doc.pdf" --output document.pdf

Step 2: Read and Analyze PDF

Use the Read tool to access PDF files. Claude's native PDF support will:

• Display the PDF content visually
• Extract text and structure automatically
• Identify tables, charts, and images

# Example: Read PDF using the Read tool
# The Read tool handles PDFs natively and extracts text + visual content
Read(file_path="/absolute/path/to/document.pdf")

Step 3: Process Based on User Request

A. Text Extraction

For simple text extraction:

python .claude/skills/pdf-processing/scripts/extract_text.py \
  --input document.pdf \
  --output document.txt \
  --preserve-formatting true

For page-specific extraction:

python .claude/skills/pdf-processing/scripts/extract_text.py \
  --input document.pdf \
  --pages 1,3,5-10 \
  --output selected_pages.txt

B. Table Extraction

Extract all tables to CSV:

python .claude/skills/pdf-processing/scripts/extract_tables.py \
  --input document.pdf \
  --format csv \
  --output-dir ./extracted_tables/

Extract tables to JSON with structure:

python .claude/skills/pdf-processing/scripts/extract_tables.py \
  --input document.pdf \
  --format json \
  --output tables.json

C. Document Summarization

Generate a summary of the PDF:

python .claude/skills/pdf-processing/scripts/summarize_pdf.py \
  --input document.pdf \
  --output summary.md \
  --style concise  # Options: concise, detailed, executive

D. Format Conversion

Convert PDF to Markdown:

python .claude/skills/pdf-processing/scripts/convert_pdf.py \
  --input document.pdf \
  --output document.md \
  --format markdown \
  --preserve-images true

Convert PDF to structured JSON:

python .claude/skills/pdf-processing/scripts/convert_pdf.py \
  --input document.pdf \
  --output document.json \
  --format json \
  --extract-metadata true

E. Visual Content Analysis

Analyze charts and graphs:

python .claude/skills/pdf-processing/scripts/analyze_visuals.py \
  --input document.pdf \
  --output analysis.json \
  --elements charts,graphs,diagrams

Step 4: Post-Processing and Output

After extraction/conversion:

1. Validate Output: Check that the output file was created successfully

ls -lh output_file.{txt,md,json,csv}

1. Preview Results: Show the user a preview of the extracted content

head -n 20 output_file.txt  # For text files
cat output_file.json | python -m json.tool | head -n 50  # For JSON

1. Provide Summary: Summarize what was extracted and offer next steps

Step 5: Handle Edge Cases

Password-Protected PDFs

python .claude/skills/pdf-processing/scripts/extract_text.py \
  --input document.pdf \
  --password "user_provided_password" \
  --output document.txt

Scanned PDFs (OCR Required)

# Use OCR for scanned PDFs
python .claude/skills/pdf-processing/scripts/ocr_pdf.py \
  --input scanned_document.pdf \
  --output document.txt \
  --language eng  # Language code: eng, fra, deu, etc.

Large PDFs (Memory Optimization)

# Process large PDFs in chunks
python .claude/skills/pdf-processing/scripts/extract_text.py \
  --input large_document.pdf \
  --output document.txt \
  --chunk-size 10  # Process 10 pages at a time

Error Handling

Common Issues

1. File Not Found

- Verify the path with ls or Glob - Check for typos in the filename - Ensure absolute paths are used

1. Corrupted PDF

- Try reading with the Read tool first - Use repair mode: python scripts/repair_pdf.py --input corrupted.pdf --output repaired.pdf

1. Unsupported PDF Features

- Some PDFs with complex DRM or encryption may fail - Inform the user and suggest alternatives

1. OCR Failures

- Check if tesseract is installed: which tesseract - Verify image quality is sufficient - Try different language settings

Best Practices

1. Always use the Read tool first - This leverages Claude's native PDF support for best results
2. Preserve structure - When extracting tables or converting formats, maintain the original structure
3. Validate outputs - Always check that output files were created successfully
4. Provide context - Tell the user what was extracted and what they can do next
5. Handle errors gracefully - If processing fails, explain why and suggest alternatives
6. Respect privacy - Remind users not to upload sensitive documents without proper authorization

Output Formats

Text Output

Plain text with optional formatting preservation
Line breaks and paragraphs maintained
Special characters preserved

Markdown Output

# Document Title

## Section Heading

Paragraph text with **bold** and *italic* formatting.

| Column 1 | Column 2 | Column 3 |
|----------|----------|----------|
| Data 1   | Data 2   | Data 3   |

![Chart Description](chart_extracted.png)

JSON Output

{
  "metadata": {
    "title": "Document Title",
    "author": "Author Name",
    "pages": 42,
    "creation_date": "2024-01-15"
  },
  "content": [
    {
      "page": 1,
      "type": "text",
      "content": "Page 1 text content..."
    },
    {
      "page": 2,
      "type": "table",
      "headers": ["Col1", "Col2", "Col3"],
      "rows": [["A", "B", "C"], ["D", "E", "F"]]
    }
  ]
}

CSV Output (for tables)

Column1,Column2,Column3
Value1,Value2,Value3
Value4,Value5,Value6

Advanced Features

Batch Processing

# Process multiple PDFs
python .claude/skills/pdf-processing/scripts/batch_process.py \
  --input-dir ./pdfs/ \
  --output-dir ./extracted/ \
  --format markdown

Custom Templates

# Use custom conversion templates
python .claude/skills/pdf-processing/scripts/convert_pdf.py \
  --input document.pdf \
  --output document.md \
  --template .claude/skills/pdf-processing/assets/custom_template.md

Selective Extraction

# Extract only specific sections
python .claude/skills/pdf-processing/scripts/extract_sections.py \
  --input document.pdf \
  --sections "Introduction,Methods,Results" \
  --output extracted_sections.md

Integration with Other Tools

This skill works well with:

• Data analysis tools - Extract tables and feed to pandas/numpy
• Documentation generators - Convert PDFs to Markdown for wikis
• Search systems - Extract text for indexing
• Automation workflows - Batch process invoices, reports, forms

Examples

Example 1: Extract and Summarize

# User: "Please extract the key points from this research paper"
1. Read(file_path="/path/to/paper.pdf")
2. python scripts/summarize_pdf.py --input paper.pdf --output summary.md --style executive
3. Show the user the summary with key findings highlighted

Example 2: Extract Tables to CSV

# User: "Get all tables from this financial report"
1. Read(file_path="/path/to/report.pdf")
2. python scripts/extract_tables.py --input report.pdf --format csv --output-dir ./tables/
3. List the extracted CSV files and preview the first table

Example 3: Convert to Markdown

# User: "Convert this PDF to markdown"
1. Read(file_path="/path/to/document.pdf")
2. python scripts/convert_pdf.py --input document.pdf --output document.md --format markdown
3. Show preview of the markdown and confirm successful conversion

Dependencies

The scripts in this skill require:

• Python 3.8+
• PyPDF2 or pypdf (PDF parsing)
• pdfplumber (table extraction)
• pdf2image (image extraction)
• pytesseract (OCR for scanned PDFs)
• Pillow (image processing)
• requests (URL downloads)

These are installed via the requirements file in assets/requirements.txt.

References

See the references/ directory for:

• pdf_capabilities.md - Detailed breakdown of Claude's PDF support
• api_reference.md - Complete API documentation for all scripts
• examples.md - More usage examples and use cases
• troubleshooting.md - Common issues and solutions

Notes for Claude

• Always read PDFs with the Read tool first - This is the most reliable method
• After reading, analyze what the user needs - Text, tables, summary, conversion?
• Use the appropriate script - Don't try to do everything manually
• Validate outputs - Always check that files were created successfully
• Provide helpful context - Explain what was extracted and suggest next steps
• Handle errors gracefully - If something fails, explain why and offer alternatives
• Be efficient - Use batch processing for multiple PDFs
• Preserve structure - Maintain document formatting when converting

Version History

• 1.0.0 (2025-11-18) - Initial release with core PDF processing capabilities