Invoice & Receipt Extractor

Typing invoices and receipts into a spreadsheet is slow, and small mistakes are easy to miss: a misread digit, a dropped discount, a date read the wrong way round.

This skill gives your AI a fixed procedure for it. It reads text PDFs, photos, scans or pasted text, and writes one JSON file per document in a fixed schema: vendor, customer, invoice number, issue and due dates, line items, subtotal, tax lines, tip, total, amount paid and due, currency, payment terms and method. It also writes one CSV row per document.

It handles the cases that usually go wrong: VAT-inclusive receipts, European number formats like 1.234,56, ambiguous dates like 03/04/2026, a dollar sign that isn't USD, discounts, shipping lines, tips and credit notes. Missing values are left empty and flagged, never guessed. Full card numbers and bank account numbers are left out.

Inside: SKILL.md, a field-by-field schema reference, extraction rules for hard cases, a blank template, a JSON Schema file, and a Node validator (no dependencies) that checks required fields, date formats and arithmetic, flags duplicates across folders and writes the CSV. Examples include three sample documents with expected output, a deliberately broken file and two duplicates.

For freelancers, small businesses and bookkeepers. It is a data-entry aid, not tax or accounting advice, and it cannot read what the AI cannot see clearly. Always check results against the originals.