ClaimReady

Your NAS already holds the proof. ClaimReady turns photos, receipts, your policy and camera events into a home inventory and, after a theft or flood, a cited insurance claim pack, locally.

Comment

Inspiration

After a theft, fire or flood, insurers ask for proof of ownership: receipts, serial numbers, photos. Almost nobody keeps a home inventory. The proof already exists, but it's scattered across photo backups, PDFs and camera archives, and claims are prepared under stress against short reporting deadlines. A UGREEN AI NAS is the one place where all of that data lives together, privately.

What it does

ClaimReady is a UGOS Skill with two modes, plus a lookup for questions about the inventory.

Inventory.

  • Finds valuables in your photos and reads serial labels with OCR.
  • Extracts receipts and invoices.
  • Links each item to its proof on a proof ladder: strong (serial match), medium (a unique brand/model or type match), or "needs review".
  • Fingerprints every evidence file with SHA-256.
  • Produces a gap list: valuables with no proof yet, each with one concrete fix, before anything happens.

Claim pack. After an incident:

  • Pulls surveillance events in the confirmed window.
  • Answers six fixed questions from your policy with clause citations: cover, deductible, limits, reporting deadline, required documents, exclusions.
  • Asks you to confirm facts that only you know, such as forced entry.
  • Re-verifies the evidence hashes.
  • Drafts a factual claim narrative.
  • Copies everything into a claim folder, with an index that shows the deadline.

How we built it

The Skill is a Markdown-only orchestration Skill (SKILL.md plus 5 references), built with the official UGOS Skill Creation Tool (uliya-skill-kit 0.7.7). The parent Skill only delegates, with self-contained instructions, to five registered experts:

  • photo-agent: structured photo search, VLM, OCR
  • document-agent: receipt/invoice extraction, policy QA with citations, writing
  • file-agent: checksums, tags, copies
  • knowledge-agent: saving the ledger and the claim index
  • security-monitor-agent: camera events and evidence

How the Skill behaves:

  • It stays read-only until the user chooses which writes to make, and every write waits for platform approval.
  • Files are copied, never moved.
  • It handles completed, failed, cancelled, waiting_user and waiting_approval.
  • It never broadens the scope after an empty result.
  • It reports partial work honestly.
  • It uses the built-in local model only and calls no external service.

Challenges we ran into

Mapping every step to what the capability catalog really supports. For example:

  • create-album only accepts picture IDs the user named, or a whole search result, so we dropped the album and group the evidence with a tag instead.
  • Photo search returns 20 previews per search.
  • Document extraction processes a bounded number of files, so the gap list is worded "no proof among the processed documents".

We simulated 10 branch scenarios against the catalog in four rounds, until every one gave an unambiguous, safe next action.

What we learned

On a NAS, the value isn't one more chat assistant. It's joining private silos (photos, documents, cameras, policy) that a cloud AI should never see together, and making the result verifiable.

What's next

Validation on the iDX6011 Pro with our test dataset:

  • 16 CC0 item photos, 3 serial labels, 12 synthetic receipts (including a duplicate and a blurry scan), and a synthetic policy.
  • Expected result: 9 items proven (3 by serial), 1 to review, 6 gaps.
  • A garage theft case to test the claim pack.

Built With

Share this project:

Updates

Submission history