Invoice Extractor 📄

Turn invoices and receipts into structured expense data. Extract from PDFs and images, auto-categorize spending, and maintain a running CSV ledger.

Hybrid approach: A Python script handles PDF text extraction and ledger management, while you (the agent) parse the invoice content — LLMs understand varied formats far better than regex.

When to Use

• "Extract data from this invoice"
• "Track my expenses" / "Add to my expense ledger"
• "Categorize this receipt"
• "Process these invoices" / "Batch process receipts"
• "Show me my spending summary"
• "Prepare tax documents" / "Get my expenses for April"

Setup

pip install pdfplumber
# Fallback: PyPDF2 (auto-used if pdfplumber unavailable)

Script: scripts/extract.py (relative to this skill directory) Config: expense-config.json (same directory)

⚡ Single Invoice Workflow

PDF Invoices

python3 scripts/extract.py pdf <file-path>

Read the output text, parse it into structured JSON (see schema below), then confirm with the user before adding to ledger.

Image Invoices (jpg, png, webp, gif)

Use the image tool with a prompt like: "Extract all invoice/receipt data from this image. Return vendor, invoice number, date, line items, subtotal, tax, total, and currency."

Parse the result into structured JSON, then confirm with the user before adding to ledger.

🔒 Confirm Then Add

Always present extracted data for user review before writing to the ledger:

📋 Invoice Extracted
Vendor: Amazon
Date: 2026-04-01
Invoice #: INV-2026-001
Description: Office supplies — keyboard and monitor
Total: €539.96 (incl. €100.97 tax)
Category: office (auto)

Add to ledger? (yes/edit/skip)

Format output for the current channel — adapt formatting to match what the platform supports. See references/formatting.md for platform-specific examples.

On confirmation, write the JSON to a temp file and run:

python3 scripts/extract.py ledger add /tmp/invoice-entry.json

Or pipe via stdin:

echo '<json>' | python3 scripts/extract.py ledger add -

If the user says "edit", modify the requested fields and re-confirm. If "skip", discard.

📦 Batch Processing

python3 scripts/extract.py batch <folder-path>

1. Run the batch command to get a JSON list of all PDFs and images
2. Process each file one at a time (PDFs via pdf command, images via image tool)
3. Collect all results — do NOT confirm each one individually
4. Present a summary of ALL extracted data at the end
5. Ask the user to confirm once: add all, edit specific entries, or skip

Show this summary after processing all files:

📦 Batch Results — 8 files processed