Inspiration

After a burglary, fire or flood, insurers and police ask for proof of ownership and value. The evidence usually exists already: photos of the TV and the bike, receipts in a downloads folder. But over the years those files get copied across phones, drives, old sync folders and the NAS, so nobody knows which copy is the original, whether it has changed, or which valuables have no evidence at all. Our own storage drives are full of duplicates and a mess everywhere, and that is what this Skill sets out to fix.

What it does

The Skill turns those files into one claim-ready inventory. For each valuable item it lists the brand, model and serial number when they can be read from the photos (up to five items per photo), the receipt the user confirmed as a match, the original evidence file and its identical copies in the folders holding the picked files and in any folders the user names (an old phone backup, a family share), and a SHA-256 fingerprint of each photo and receipt. It ends with a gap list: items with no proof of purchase, receipts that match no photographed item, and printed warranty end dates. After a break-in, it can attach the home cameras' events, snapshots and clips for the time window the user gives. The saved inventory gets its own fingerprint for the user to write on a printed copy. Months later, a re-check run compares the saved inventory with the fingerprint the user kept, to show whether it was edited, fingerprints every listed file again, and reports what changed or disappeared, naming any identical copy that is still intact.

How it was built

It is an orchestration Skill for the Uliya Skill Kit 0.7.7 that uses only published NAS capabilities (catalog 2.2.0): photo search, vision, OCR and photo metadata through photo-agent; receipt extraction through document-agent; folder listing, file reads, SHA-256, stat and the duplicate scan through file-agent; the inventory save through knowledge-agent; and the camera list, event timeline and event evidence through security-monitor-agent. Each step passes only verified upstream results to the next. Photos are read only after the user picks them; originals and receipt matches are only proposals until the user confirms them. Every write (the inventory, an optional second copy, an optional re-check report, the optional album) goes through the platform's approval workflow. Source files are never changed and duplicate copies are never deleted. Serial numbers, receipts and camera clips are handled only by the NAS's own experts: the Skill names no outside service or model and has no upload step. Camera evidence is listed by event and file only; the Skill does not identify people or read plates. It needs no credentials or configuration.

Challenges

Serial numbers read by OCR can be wrong, so every reading is flagged for the user to check. Copies are found only in the folders scanned, and only byte-identical ones; resized or edited versions are not. Copying can reset file times, so the proposed original is a best guess the user confirms. A file cannot hold its own checksum, so the inventory's fingerprint goes in the report and onto the printed copy. An inventory saved on the same NAS does not survive the loss of the NAS, so the final report always asks for a printed or off-site copy. Empty searches and empty camera windows are reported as empty instead of broadened. One run covers 20 photos and 60 documents, so larger homes need several runs.

What's next

Nothing has run on a NAS yet; the package has passed only the kit's static checker. Next: run the two core cases and the edge case in references/validation-cases.md on a synthetic household set with planted duplicate copies, against set targets (at least 8 of 10 receipt matches, 3 of 4 serial numbers, all 3 planted copy groups, no false changes on re-check), then tune the photo-search wording and the matching rules on real libraries.

Built With

Share this project:

Updates

Submission history