Inspiration

Every household has a drawer of paper and a folder of PDFs it never opens again. The moment you need one (a fridge that stops cooling, an insurance renewal, a bill dispute) you are digging through files one by one, and the warranty has often already run out. A NAS already holds all of this paperwork. We wanted it to know what it holds and speak up before something expires.

What it does

Paper Trail is a Skill for UGREEN AI NAS. Point it at a folder and it:

finds every PDF, Office document and photo of a paper document;
builds a register: type, vendor, date, amount and any warranty, contract or renewal end date;
lists what ends in the next 90 days first;
after you confirm, tags each file, optionally files them into year and type folders, and saves the register next to them;
answers questions like "When does the fridge warranty end?" with the answer and the source file.

Nothing is deleted or overwritten, and documents never leave the NAS.

How we built it

Paper Trail is a Markdown-only orchestration Skill built with UGREEN's Skill Creation Tool (uliya-skill-kit 0.7.3). It delegates each step to a registered NAS expert:

Step	Expert	Capability
Find documents and photos	file-agent	`ug-filesystem find`
Extract fields	document-agent	`ug-document extract` (receipt, invoice, contract schemas)
Read photographed paperwork	photo-agent	`ug-photo vision-ocr`, `vision-vlm`
Tag and file	file-agent	`ug-filemgr tag-add-file`, `ug-filesystem mkdir`, `cpmv`
Save the register	knowledge-agent	`ug-knowledge save-text`
Answer questions	document-agent	`ug-document qa`

Each step passes only the real result of the step before it, every write waits for user confirmation and platform approval, and all five delegation states (completed, failed, cancelled, waiting_user, waiting_approval) are handled. The package passes UGREEN's static checker with 0 errors and 0 warnings.

Challenges we ran into

Our first idea did not fit the platform. We started with a photo culling Skill, then found that the capability catalog has no image quality scoring and Skills cannot write sidecar files or ship scripts. We switched to an idea that every step of the catalog supports.
Photos are not documents. ug-document extract works on files, so photographed receipts go through vision-ocr and vision-vlm instead, and those records are marked for review.
The template and the checker disagreed. The kit's template shows one format for capability evidence, while the checker requires another. Reading the checker's rules got us to a clean pass.

Accomplishments that we're proud of

A complete workflow from trigger to saved register, with no placeholders.
Every step maps to a real capability in the catalog.
Safe by design: confirm before writing, never overwrite, never delete, nothing leaves the NAS.

What we learned

A good NAS Skill is mostly about orchestration: choosing the right expert for each step, writing self-contained instructions, and saying exactly what happens when a result is empty or an approval is pending. Checking the capability catalog before writing a single step saved us from building something the device cannot run.

What's next for Paper Trail

Test extraction accuracy and speed on a real UGREEN AI NAS with a 500-document folder.
Spending summaries by vendor and category per year.
Scheduled runs that pick up new documents and remind you before a warranty or renewal ends.