Receipt to Excel

What you're buying: a download package (skill instructions + Python scripts) that runs on your own computer inside your own AI agent — it is not a cloud service.
You need: an AI agent that supports skills and can read images (e.g. Claude Code or Codex) — its subscription or API costs are yours; Python 3.9+ with pillow, pypdfium2 and openpyxl.
Third-party services: none. The scripts work offline. Receipt images are read by your own AI agent, so its provider's data policy applies.

A folder of receipts in, a tidy Excel workbook out. No typing, no subscription, and your data stays on your computer.

What you get

• 📊 Receipts sheet: one row per receipt — date, vendor, seller and buyer tax IDs, invoice number, currency, net, tax, tip, total, category, payment method, items, confidence and source file. Filters and frozen panes; rows needing review are yellow, duplicates grey.
• 📈 Summary sheet: month × category SUMIFS formulas, per currency. Edit a row and the totals update.
• 🔍 Review sheet: every receipt that needs a human check and why, plus duplicate files and unread pages.
• ℹ️ About sheet: when it was generated, plus a disclaimer.

Example

A folder of 6 files — a Taiwan e-invoice (plus an identical copy), a triplicate invoice dated in the ROC calendar, a US coffee-shop receipt with tip, an English PDF invoice and a blurry, skewed taxi receipt:

• 5 pages read; the duplicate file was skipped and listed on the Review sheet
• ROC date 115/04/02 correctly converted to 2026-04-02; invoice numbers normalised
• Summary: TWD 1,463 (3 receipts) and USD 45.50 (2 receipts) — live SUMIFS formulas that recalculate correctly in Excel
• The blurry taxi receipt got 70% confidence, was marked yellow and listed for review
• A planted bad record (unconverted ROC year, totals that don't add up, wrong tax-ID checksum, unknown category, duplicate invoice number, broken JSON) — every error was caught

The Summary sheet it built — month × category totals per currency, as live formulas:

Try it — copy a prompt

• "Put every receipt in receipts/2026-10 into an Excel file, categories from my accountant's list, and tell me what needs review."
• "Here are 40 photos from my business trip — build the expense sheet in USD and EUR, tips included."

Why buy it

One payment, every month's receipts — no per-scan or subscription fees for an expense app.

Who it's for

Freelancers, small-business owners, bookkeepers, assistants filing expense reports and anyone doing their own books.

Supported receipts

• 🇹🇼 Taiwan e-invoices, uniform invoices (duplicate, triplicate, cash register), taxi and parking receipts — ROC years converted automatically
• 🌏 English and other international receipts and invoices (USD, JPY, EUR…), including tips
• JPG, PNG, WEBP and TIFF photos, and multi-page PDFs (one receipt per page)

Sample input — a Taiwan e-invoice photo:

Automatic checks

• Date format, including unconverted ROC years
• Net + tax + tip ≈ total
• Taiwan invoice-number format (AB-12345678)
• Taiwan tax-ID checksum (official Ministry of Finance rule)
• Duplicate invoice numbers, and likely duplicates (same day, vendor and amount)
• Identical image files are read only once
• Valid currency codes, categories and payment methods

When the AI misreads something, your agent re-reads that image and fixes it; receipts that are genuinely unreadable keep their review flag. The AI never changes numbers just to make a check pass.

How it works

1. Tell your agent where the receipt folder is (optionally attach your accountant's category list).
2. The script converts PDFs, resizes images and skips duplicates.
3. The agent reads each receipt against a strict field spec — anything it can't see is left blank, not guessed.
4. The script validates and builds the Excel file; the agent spot-checks 3 receipts against the originals.
5. You get the totals, plus how many items need review and why.

Requirements

• Python 3.9+
• pip install pillow pypdfium2 openpyxl
• An AI agent that can read images (Claude, Codex, etc.)

Limitations (please read)

• This is AI reading and can be wrong — the original receipts always prevail. Clear the Review sheet before using the numbers.
• A bookkeeping aid only — not tax or accounting advice. Ask a qualified accountant or tax professional whether an expense is deductible.
• Handwritten, faded thermal-paper or badly cropped receipts are less accurate and will be flagged.
• No connection to banks, government e-invoice platforms or accounting software.
• Categories are suggested bookkeeping labels, not tax classifications.

Files

• SKILL.md — agent workflow
• scripts/prepare.py — folder scan, PDF to image, de-duplication
• scripts/build_xlsx.py — validation and Excel output
• references/schema.md — field spec
• references/categories.md — default categories (English and Chinese)