Claim Dossier: UGREEN Hackathon 2026 Round 1
Entrant: Joseph Gaither (individual). Track: Automate Your Workflow.
1. One sentence: What does the Skill do? After a break-in, leak, fire or storm, it turns the household's own NAS data into an insurance claim dossier that proves what happened, what was owned and what it cost.
2. Who is the target user and problem? Homeowners and renters filing a claim. Insurers ask for proof of ownership and value per item, and most people spend days digging through receipts, email exports and photos while stressed. Unproven items get denied or underpaid.
3. What is your proposed solution? The user lists the lost or damaged items and the incident window. The Skill searches their cameras, receipts, invoices and photos on the NAS, matches evidence to each item, scores it proven, partial or no evidence found, fingerprints every evidence file, and lists exactly what is still missing.
4. What is the Core User Flow? Say "I need to file a claim for a break-in last night", give the window and item list, review the dossier in chat, then optionally approve copying the evidence into one claim folder with the dossier saved beside it.
5. What substantive difference does your idea offer compared with existing tools or conventional AI solutions? Home inventory apps need you to catalog everything before the disaster. Cloud AI needs you to upload private receipts and camera footage. Claim Dossier works after the fact on data already on the NAS, crosses four data types in one run, never estimates value, and adds SHA-256 fingerprints so an adjuster can trust the files.
6. Which UGREEN AI NAS capabilities will be used, and at which step in the workflow? Step 1 ug-surveillance (cameras, event-timeline, event-evidence). Step 2 ug-document (extract with receipt and invoice schemas, reconcile). Step 3 ug-photo (search-picture, vision-vlm, meta, vision-ocr). Step 4 ug-filesystem (stat, sha256sum). Step 6, approved only: ug-filesystem (mkdir, cpmv copy) and ug-knowledge (save-text).
7. What was your technical approach? A Markdown orchestration Skill that delegates to five registered experts in a fixed order, passing only real upstream results forward. Steps 1 to 5 are read-only. Inputs: incident type, window, item list, approved scope. Writes need user confirmation plus platform approval, copy never move, never overwrite. Models: the NAS local model through the photo and document experts. No data leaves the NAS. All five delegation states are handled.
8. What is the current development status? Implemented: SKILL.md with five delegation steps, scoring rules, approval gates and state handling, plus three references (dossier format, validation plan, limits). It passes the official uliya-skill-kit 0.7.3 static checker with zero errors and zero warnings. Pending NAS validation: real expert outputs, match quality and timing.
9. What is your final validation plan? Core 1: break-in with seeded camera event and three fully documented items, expect all proven with correct totals. Core 2: water leak without cameras, expect one proven, two partial, accurate gap list, nothing written. Edge: conflicting receipt prices, an item with no data, and a file deleted mid-run, expect both prices reported, no substitution, and the unreadable path reported.
10. Are there any known limitations of your Project? Proof depends on what is stored on the NAS; receipts that live only in email are reported as gaps. Vague item names match poorly. Value comes only from documents, with no depreciation estimate. Camera retention limits the timeline. Untested on device until the finalist day.
Third-party components: none. External services and data flows: none; all processing uses NAS experts.
Built With
- markdown
- skill.md
- ugreen-ai-nas
