Inspiration
I am a private individual with some freelance paperwork. Receipts, policies and warranties pile up in downloads and scan folders. When an appliance fails or a policy renews, I cannot find the paper or the date, and I miss notice windows. A chatbot answers once and forgets. I wanted a durable register that stays on my own NAS.
What it does
Paper Trail turns invoices, receipts, warranty cards and insurance papers scattered across NAS folders into one dated register. It reads only folders I name, extracts issuer, date, amount, validity and notice terms with the page and quote as evidence, and tells me what expires or renews soon. It saves the register on the NAS, applies status tags visible in the UGOS file manager, writes a change log, and confirms and logs every file change. Nothing is guessed, and nothing is deleted.
How I built it
The Skill is a Markdown-only package: SKILL.md plus four references. The parent Skill delegates to file-agent, document-agent, photo-agent and knowledge-agent, passing only verified results downstream. It uses UGREEN AI NAS catalog 2.0.0 actions: file-workflow get-user-dir; ug-filesystem ls, find and stat; ug-filemgr scan-dedupe and tag actions; ug-document metadata, extract, qa and reconcile; ug-photo vision-ocr; and ug-knowledge save-text.
Challenges I ran into
The catalog has no scheduler, so the expiry digest runs on demand. Extraction accuracy depends on the NAS document expert; handwritten or poor scans may be unreadable, so every fact carries its page and quote and computed dates are flagged. Approvals are explicit: the Skill waits at waiting_user and waiting_approval, and it never bypasses platform gates.
Accomplishments that I'm proud of
A complete round-1 Skill package, reviewed by hand against the kit's format rules and catalog 2.0.0. The flow handles completed, failed, cancelled, waiting_user, waiting_approval and empty results.
What I learned
Local-first matters for financial documents. A durable, evidence-backed register is more useful than a one-off answer, and tags in the file manager make the state visible where I already work.
What's next
Test Paper Trail on a UGREEN AI NAS: extraction accuracy on real scans, vision-ocr quality, save-text naming, tag and copy approvals, and the file types covered by find. Later, add scheduled reminders when the platform supports them.
Built With
- ai-agents
- document-ai
- markdown
- ocr
- ugos
- ugreen-ai-nas
