Inspiration
My phone's camera roll is full of receipts, warranty cards and bills I photographed "for later." Later never comes: when something breaks, I scroll through thousands of family photos hoping to find the right one. Cloud tools mean uploading financial paperwork I'd rather keep private, and a NAS already holds the library, so the fix belongs there.
What it does
Paperwork Rescue is a UGREEN AI NAS Skill. Tell it a date range ("rescue October's receipts") and it:
- Finds photos that look like paperwork and reads vendor, date and total.
- Drops anything already archived, so monthly re-runs only handle new photos.
- Shows a preview table. Low-confidence items go to a "needs review" group.
- After approval, tags the photos and copies them into
Paperwork/year/type. Originals are never moved or deleted. - Saves a Markdown index, so a receipt can be looked up later.
Photos of identity documents and payment cards are counted but never read out or recorded.
How I built it
The Skill is pure Markdown orchestration (no scripts) following the UGOS Skill Development Standard. A parent workflow delegates in order to the photo, file and knowledge experts, and each step uses only the previous step's verified results. I drafted it with Claude and validated it with the official Skill Creation Tool checker, which reports zero errors and zero warnings.
Challenges
- Capability limits: my first ideas (bank-statement reconciliation, scheduled scans) didn't match the capability catalog, so I redesigned around what UGOS actually supports.
- Safety design: handling sensitive photos, partial failures and duplicate copies without silently guessing.
- No NAS yet: it has passed static checks only. Recognition accuracy on real phone photos and the approval flow still need testing on a real NAS.
What I learned
Design from the real capability list, make every write approval-gated, and treat "no result" as a valid answer instead of filling gaps.
What's next
Test on the NAS if selected, tune recognition for blurred and non-English receipts, and add scanned PDFs.
Built With
- agent-orchestration
- claude
- document-extraction
- file-tagging
- markdown
- ocr
- photo-library
- ugos
- ugos-skill-creation-tool
- ugreen-ai-nas
- uliya-skill-standard
- vision-language-model
