Inspiration
After a burglary, fire or flood, insurers ask for proof of what you owned and what it cost. Almost nobody keeps a home inventory. The proof usually exists, scattered across a photo library and a downloads folder of receipts, but rebuilding it under stress takes days, and missing proof means smaller payouts.
What it does
The Skill has two modes.
- Inventory mode asks the NAS photo expert to describe item photos and read serial labels and capture dates, and the document expert to extract merchant, date, price and warranty from receipts. It matches the two with transparent rules (strong, medium, weak, unproven) and lists the gaps you can fix today.
- Claim mode pulls the affected items from the saved inventory, finds extra photo proof and, with your consent, adds camera events from the incident window. It copies everything into one claim folder (originals are never moved) and drafts the claim letter from those facts only.
How we built it
A parent Skill orchestrates five registered experts (photo-agent, document-agent, file-agent, knowledge-agent, security-monitor-agent) through 11 delegations, handles all five delegation states, and makes every write wait for approval. Matching is deterministic and explained in the report, so every row cites the file it came from. Camera search runs only for theft, inside the incident window, with consent, and never identifies unknown people.
Challenges we ran into
Designing matching rules that are honest about uncertainty: a receipt line is used for one item only, ties go to the user, and items without proof are reported, never invented.
Accomplishments that we're proud of
Everything stays on the NAS: serial numbers, receipts, room photos and camera footage never leave the user's own hardware, and every field in the inventory cites the file it came from.
What we learned
The value is in joining sources no single app connects (photos, documents and cameras), and in being explicit about how strong each piece of proof is.
What's next for Home Inventory Claim Pack
Tuning on real expert output on the testing day, and replacement-value estimates as an optional, clearly separated column.
AI disclosure
Claude (Anthropic) via Claude Code helped research the capability catalog and draft the Skill and proposal; the entrant chose the idea and reviewed and edited the content. Full disclosure is in references/disclosures.md inside the Skill package.
Built With
- ai-nas
- markdown
- ugos
- ugreen
- uliya-skills
