Inspiration
After a break-in, a fire or a breakdown, insurers ask for proof of ownership and value - and most households discover the gap at claim time. 59% of consumers keep no inventory of their possessions; of those who do, 48% lack receipts and 27% lack photos (US NAIC). The photos and the receipts already exist, scattered across the same NAS. I wanted a Skill that links them, privately, and tells me what I could actually claim.
What it does
Home Proof finds the belongings in your photo library using the NAS vision models, extracts product, brand, model, serial number, merchant, date, amount and warranty wording from your stored receipts, invoices and warranty documents - every field quoted from its source - and links the two sides fail-closed: an exact shared serial number or a quoted brand and model match links automatically, anything weaker waits for your confirmation. The result is a warranty-aware home inventory grouped LINKED, POSSIBLE_MATCH, PROOF_MISSING and DOC_ONLY, with warranties bucketed open, expiring, expired, unstated or needs-review. On confirmation it saves an insurance-ready note and organizes proof-linked photos into an album. Nothing leaves the device. No status is a guarantee that an insurer or manufacturer will accept a claim.
Why it is different
Home inventory apps exist, but they depend on manual entry - which is why most people never finish one. Cloud AI tools make you upload family photos and financial documents to someone else's server. Home Proof builds the inventory automatically from what already lives on the NAS, entirely locally, and it is evidence-bound: it never invents a product, a price, a serial number or a match to make the inventory look complete.
How I built it
The Skill is an orchestration SKILL.md built with the official UGOS Skill Creation Tool (uliya-skill-kit 0.7.3), delegating to the registered file, document, photo and knowledge experts over supported CLI capabilities: file search and hash dedupe locate and clean candidate documents; document extraction builds quoted purchase records; photo object search, vision descriptions and vision OCR find belongings and read label text; knowledge save-text stores the confirmed inventory; album creation organizes proof-linked photos. A documented match-strength ladder and fail-closed warranty rules keep every claim traceable, and all writes are approval-gated. The package ships a designed end-to-end worked example and a Finalist validation plan as references. It was drafted with an AI assistant under my direction, and that use is disclosed in the package; it passes the official static checker with zero diagnostics.
Challenges
Real evidence is messy: torn receipts, photos without readable labels, the same purchase in three files, weak pairs that almost match. The design decision was to fail closed at every step - unresolved dates go to needs-review, weak matches wait for the owner, and unstated warranty terms are never guessed.
