Inspiration

The knowledge of how a life runs (which lease, which insurer, which bills auto-renew, where the car papers are, which projects are mid-flight) is scattered across hundreds of PDFs, photos and exports. If the owner is sick, travelling or gone, nobody can reconstruct it. Writing it down by hand is a job nobody finishes, and it is stale the day it is written.

What it does

HANDOVER assembles, from files already on your NAS, the one document nobody writes: everything a trusted person would need to run your home, business or project if you suddenly couldn't. Every line is cited to its source file and quoted passage, and every gap is named: no warranty for the fridge, no signed copy of a contract, no contact for the landlord.

How it works

  1. Inventory (read-only): file-agent lists the folders you name and groups duplicates (ug-filesystem, ug-filemgr scan-dedupe).
  2. Extract: document-agent reads PDFs, email and text (ug-document extract); photo-agent runs OCR on photographed receipts and cards (ug-photo vision-ocr).
  3. Assemble: facts go into chapter templates for home, business or project. Conflicting documents (two leases, two end dates) are checked with ug-document reconcile and shown side by side, never silently resolved.
  4. Review: the draft book is shown in chat. You confirm, correct or redact.
  5. Save (approval required): knowledge-agent saves one Markdown file to a folder you choose (ug-knowledge save-text).
  6. Diff: re-running shows only what changed, including gaps closed.

Account and policy numbers are masked to the last four characters, credentials are never extracted, and nothing leaves the NAS.

How I built it

A Markdown-only Skill following the Uliya Skill Development Standard: SKILL.md plus reference files for chapter templates, extraction rules, gap rules, a finalist test plan and disclosures. The package passes the official uliya-skill-kit 0.7.3 checker with 0 errors and 0 warnings.

Challenges

Making "every fact cited" honest rather than decorative: every fact must carry a verbatim passage, long documents are chunked by page, and anything the Skill is unsure about becomes a gap instead of a guess.

What's next

If selected as a finalist: run the synthetic household and business test corpora on the assigned NAS, validate the approval flow and local-model behaviour on long documents, and record the demo.

Built With

  • markdown
  • ugos
  • ugreen-ai-nas
  • uliya-skill-kit
Share this project:

Updates

Submission history