Home Inventory Claim Pack

Inspiration

After a break-in, flood or fire, the insurer's first question is "can you prove you owned it, and what it cost?" Most households do have that proof, in phone photos, saved receipts and camera footage, but it is scattered, and people end up hunting for it under stress, days later. A UGREEN AI NAS already holds all three kinds of evidence in one private place, so it is the right home for a Skill that keeps the proof ready.

What it does

Home Inventory Claim Pack has two modes.

• Inventory mode ("make a list of my valuables for insurance"): finds photos of valuables, identifies brand, model and serial with the vision model and OCR, finds receipts, invoices and order screenshots, extracts merchant, date and price, matches each item to its receipt, records when it was last photographed, and can quote your own policy PDF to flag items above a single-item limit. After approval it tags the files, creates a Home Inventory album and saves a priced inventory with a proof to-do list.
• Incident mode ("we were burgled last night"): loads the saved inventory, reads camera events for the incident window, drafts a factual claim letter and, after approval, copies the proof into one dated Claim folder with an incident timeline and an evidence manifest recording a SHA-256 fingerprint for every file, so an insurer can verify nothing was altered.

It never invents prices, serials or policy terms, and never identifies people in footage.

How we built it

A pure orchestration SKILL.md following the Uliya Skill Development Standard, delegating to photo-agent, file-agent, document-agent, knowledge-agent and security-monitor-agent. Each delegation carries the real upstream values so experts need no shared context, and catalog limits are respected (200-file reconcile batches, 200-event timelines). References define receipt matching, policy flags, claim pack layout and edge cases. It passes the official UGOS Skill Creation Tool checker with zero errors and zero warnings.

Challenges

Honest receipt matching (match, possible match, no proof found) without letting the model guess; making every write (tags, album, folder, copies, saved files) wait for confirmation and platform approval; and keeping evidence trustworthy, which led to the SHA-256 manifest.

What's next

Validate on a real iDX6011 Pro with synthetic photos, receipts, a sample policy and a staged camera incident, with pass/fail criteria for matching accuracy, manifest verification and zero invented values.

Built With

• ai-nas
• computer-vision
• llm
• markdown
• ocr
• skill
• ugos
• ugreen
• uliya