ShopBooks

Turn a month of scanned receipts and supplier invoices on your NAS into a checked expense ledger that flags double payments, without your records leaving home.

Comment

Inspiration

Small shops in Nigeria keep their books in a shoebox: phone photos of receipts, faded POS slips and PDF supplier invoices. Totalling a month is slow, the same invoice sometimes gets paid twice, and cloud bookkeeping apps mean uploading private financial records. I wanted a shop owner to say one sentence to their NAS and get a checked monthly ledger back, with the receipts never leaving home.

What it does

ShopBooks is a Skill for UGREEN AI NAS. The owner says something like "Make my ShopBooks ledger for 2026-11 from /Shop/Receipts and save it in /Shop/Ledgers." The Skill:

  1. Finds the month's receipts, POS slips and invoices in that folder, and spots identical duplicate files.
  2. Reads each one: supplier, date, invoice or receipt number, total, VAT and payment method.
  3. Compares with last month's ledger, if the owner points to one.
  4. Builds the ledger: totals by supplier, possible double payments (same number, same amount from the same supplier within 7 days, flagged copies, identical files), items that need a human check, and suppliers whose spending changed by 25% or more.
  5. Saves the ledger as Markdown only after approval. Tagging the processed files is optional and also needs approval.

It never moves, edits or deletes a receipt, never adds amounts in different currencies, and never guesses a missing value.

How I built it

With the official UGOS Skill Creation Tool (uliya-skill-kit 0.7.7). Every step maps to a capability in the public catalog and is delegated to the registered expert:

  • ug-filesystem find / cat and ug-filemgr scan-dedupe → file-agent
  • ug-document extract (receipt and invoice schemas, duplicate detection) → document-agent
  • ug-knowledge save-text → knowledge-agent
  • ug-filemgr tag-add / tag-add-file (optional) → file-agent

Each instruction is self-contained, each step uses only the real result of the step before it, and every public status (completed, failed, cancelled, waiting_user, waiting_approval) plus empty and partial results is handled. The package passes the kit's static checker with 0 errors and 0 warnings.

Challenges

Deciding what a Skill may and may not do on someone's financial records. The rule I settled on: read freely, write only with platform approval, never touch the source receipts, and report "possible" double payments rather than accuse.

What's next

If selected as a finalist: validate on a real NAS with a month of real Nigerian receipts, including two planted double payments, and tune for faded thermal POS slips.

Built With

Share this project:

Updates

Submission history