Inspiration
In most households one person knows where the passports, insurance policies and certificates are. Everyone else finds out they don't on the worst possible day. These are also the last papers anyone wants to upload to a cloud AI, which makes a private NAS the right place to solve it.
What it does
Family Emergency Binder is an AI Skill for UGREEN AI NAS. You point it at the folders where household papers live. It identifies each record, pulls out the holder, issuer, expiry date and emergency phone number, and saves one Markdown binder with:
- an expiry watchlist (expired, expiring soon, no date found)
- records grouped by category, each linked to where the original is
- checklist records that were not found in the scanned folders
- files that could not be read
Document and policy numbers are masked to their last four characters, so the binder points to the originals without copying them. A Refresh mode re-reads only new or changed files and reports what changed since the last binder.
How I built it
The Skill is a Markdown-only package written to the Uliya Skill Development Standard with the official UGOS Skill Creation Tool. SKILL.md orchestrates four registered experts through delegation: file-agent lists the scope and tags sources, photo-agent locates and reads photographed documents (opt-in), document-agent classifies and extracts facts, and knowledge-agent saves the binder with read-back verification. Both writes go through platform approval.
Challenges
Without a NAS in Round 1, every step had to be justified against the public capability catalog instead of by running it. The hardest design question was what to do when a result is empty or partial: the Skill never guesses a date, never widens the folder scope, and never reports a save it cannot verify.
What I'm proud of
The binder reports absence as well as presence, and it is safe to keep: masked identifiers, no passwords, no external sharing. The package passes the official static checker with no errors or warnings.
What I learned
How UGOS Skills separate orchestration from execution, and how much of a reliable agent workflow is deciding what happens when a step does not complete.
What's next
Run the validation plan on a real UGREEN AI NAS: twelve synthetic records, two core test cases and one edge case, then tune batch sizes and the photo search step from real results.
AI tools used
Claude Code (Anthropic) proposed the concept and name, wrote the Skill package and drafted the proposal text, using the official UGOS Skill Creation Tool. The disclosures are in the Skill package and the proposal PDF.
Built With
- claude-code
- markdown
- ugos
- ugreen-ai-nas
- uliya-skill-kit
