Year-End Tax Pack

A UGOS Skill that turns a year of scattered receipts, invoices and statements on UGREEN AI NAS into a checked, accountant-ready folder.

Comment

Inspiration

Freelancers and small-business owners who keep their records on a NAS end up with a year of expenses scattered across emailed PDFs, scans, phone photos of paper receipts and bank statements. At year end they spend hours collecting and retyping them, and missing months surface only when the accountant asks.

What it does

Year-End Tax Pack is a UGOS Skill. One request ("prepare my 2025 tax pack") runs a guided workflow over the folders the user names:

  1. Finds receipts, invoices, receipt photos and statements for the tax year, plus exact duplicates.
  2. Reads date, vendor, total and tax from documents (ug-document extract) and photos (ug-photo vision-ocr / vision-vlm).
  3. Shows a preview: category totals, likely duplicates (for example a photo and a PDF of the same receipt), missing statement months, statement charges without a receipt, unreadable files.
  4. The user edits categories, excludes items, chooses summary-only, or cancels.
  5. Only after approval: copies (never the originals) go into a new year folder grouped by category and named date_vendor_amount, checksums are verified, and a summary plus a CSV ledger are saved.

Originals are never moved, renamed or deleted. The Skill organizes bookkeeping records; it gives no tax advice and makes no network calls of its own.

How we built it

An orchestration-only SKILL.md with three linked references (ledger and folder format, user guide, disclosures), built with the official UGOS Skill Creation Tool (uliya-skill-kit 0.7.3). Nine steps delegate to four registered experts: file-agent, document-agent, photo-agent and knowledge-agent, using only capabilities listed in the public capability catalog. The kit checker reports 0 errors and 0 warnings. Claude Code was used to read the kit documentation, draft the Skill and run the checker; this is disclosed in the package and the PDF.

Challenges

Keeping every step grounded in the previous step's real result, so the Skill stops and says so when nothing is found instead of filling gaps. The starter guide suggests a README, but the checker accepts only SKILL.md at the top level, so the user guide lives in references/.

What's next

Run it on a real UGREEN AI NAS: extraction quality on real receipts and photos, reconcile output, batch sizes, copy-then-rename and the approval prompts.

Built With

  • ugos
Share this project:

Updates

Submission history