safe-agentic-workflow: Command for Gemini CLI

description = "Extract structured data from PDF documents"

prompt = """

Extract PDF Command

Extract structured data from PDF documents using Gemini's multimodal document understanding.

Target: {{args}}

Capabilities

• Text extraction with formatting preservation
• Table detection and CSV/JSON export
• Form field extraction
• Invoice/receipt data parsing
• Multi-page document processing

Workflow

1. Identify PDF File

!{ls -la {{args}} 2>/dev/null || echo "File not found"}

2. Read and Analyze PDF

Read the PDF file and analyze its structure:

• Identify document type (invoice, form, report, contract, etc.)
• Detect tables, forms, and structured content
• Extract text with section awareness

3. Extract Structured Data

Based on document type:

For Invoices/Receipts:

{
  "vendor": "Company Name",
  "invoice_number": "INV-12345",
  "date": "2025-01-14",
  "items": [
    {"description": "Item 1", "quantity": 2, "price": 99.99}
  ],
  "subtotal": 199.98,
  "tax": 16.00,
  "total": 215.98
}

For Tables:

Column1,Column2,Column3
Value1,Value2,Value3

For Forms:

{
  "field_name": "field_value",
  "checkbox_field": true
}

4. Output Options

Specify output format:

• --json - Structured JSON output
• --csv - Table data as CSV
• --markdown - Formatted markdown
• --text - Plain text extraction

Usage Examples

/media:extract-pdf ./invoice.pdf
/media:extract-pdf ~/Documents/contract.pdf --json
/media:extract-pdf ./forms/application.pdf --csv

Batch Processing

For multiple PDFs:

/media:extract-pdf ./invoices/

Will process all PDFs in directory and generate combined output.

Notes

• Gemini supports PDFs up to 100MB (Gemini 3)
• Complex layouts may require multiple passes
• Handwritten text: OCR quality varies
• Scanned documents: Ensure good image quality """