Extracts structured data from Japanese receipts, invoices, and hometown tax donation certificates. The implementation is thoughtful: it tries text extraction first for PDFs before falling back to OCR, and the recommended approach runs two parallel reads of the same image to cross-validate amounts and dates, prompting you to choose when they disagree. Handles common Japanese date formats including converting from imperial calendar years. Outputs clean structured data with vendor names, itemized amounts, and dates in YYYY-MM-DD format. It's designed to be called by other skills but works standalone too. If you're building expense tracking or tax prep tools for Japanese users, this handles the tedious part of data entry with appropriate paranoia about accuracy.
npx skills add https://github.com/kazukinagata/shinkoku --skill reading-receipt