Home Contents Inventory: a UGREEN AI NAS Skill

Entrant: Nathan Melia (individual, United Kingdom) · Track: Understand Your Life · Skill package: home-contents-inventory.zip

1. What does the Skill do?

It turns the photos and receipts already on your NAS into an insurance-ready list of your belongings, showing which items have proof of purchase and which do not.

2. Who is the target user and problem?

Households and small landlords. After a burglary, fire or flood, insurers ask for proof of ownership and value, but receipts and photos sit in separate folders and nobody builds the list until after the loss.

3. What is your proposed solution?

A Skill that reads a photo folder and a documents folder the user names, identifies items in the photos, matches each to a receipt, invoice or warranty, and produces a reviewed inventory with gaps highlighted.

4. What is the Core User Flow?

1. User: "Make a contents list for my insurance from /Photos/Home and /Documents/Receipts."
2. Skill describes items, extracts purchase records and matches them.
3. User sees a draft table: item, photo, document, date, price, warranty, confidence, plus an "unmatched" list.
4. User corrects and confirms.
5. Skill saves the inventory as Markdown and, if the user agrees, creates an evidence album and tags the receipts.

5. What substantive difference does your idea offer?

Cloud inventory apps need manual entry and upload of private documents. Generic chatbots read one file at a time. This Skill cross-references two kinds of private data, images and financial documents, entirely on the NAS. It also reports what is missing rather than inventing values.

6. Which UGREEN AI NAS capabilities are used, and where?

• Step 1: ug-filesystem (ls/find) lists photo and document paths.
• Step 2: ug-photo (vision-vlm, vision-ocr) describes items and reads model labels.
• Step 3: ug-document (extract) pulls merchant, date, price and warranty.
• Step 4: ug-document (reconcile) matches items to records.
• Steps 6–8, after confirmation only: ug-knowledge (save-text), ug-photo (create-album) and ug-filemgr (tag-add-file).

7. What was your technical approach?

A Markdown orchestrator following the Uliya standard. It delegates to file-agent, photo-agent, document-agent and knowledge-agent and passes only real upstream results downstream. Inputs are the two folder paths, an optional filter, and a save location. Reads need no approval. The save, album and tag steps are write actions, gated behind explicit user confirmation and the platform approval flow. It uses the NAS's on-device models, with no external services, so financial and household data never leave the device. All five delegation states are handled; partial results are kept.

8. What is the current development status?

• Implemented: the complete SKILL.md and disclosures. It passes the bundled checker (check-skill.mjs, static-pass, 0 errors, 0 warnings).
• Not yet implemented: tuning of match wording and confidence rules.
• Pending NAS validation: expert behaviour, recognition quality and approval prompts.

9. What is your final validation plan?

• Core test 1: 15 synthetic item photos and 10 synthetic receipts. Measure correct matches and the unmatched list.
• Core test 2: confirm the save produces one file at the given path and the album contains exactly the confirmed photos.
• Edge case: an empty or unreadable documents folder. The Skill must say so, offer a photo-only list, and write nothing.

10. Are there any known limitations?

Recognition depends on photo quality and on-device model accuracy. Generic items such as "chair" match poorly, so matches carry confidence levels and need review. The Skill does not value items or contact insurers. It depends on UGOS expert agents and catalog version 2.0.0.

---

Third-party components and licences: none. The package is Markdown only and uses UGREEN-provided expert agents and on-device models. AI tools: an AI assistant was used to read the developer kit, check the Skill and proofread.

Built With

• markdown