Heirloom

Inspiration

A friend's father died suddenly. His family found a bank account, a crypto wallet with a lost seed phrase, and insurance papers inside an email nobody could open. Some assets were never recovered. Nobody knew what existed, where it lived or who should receive it.

Family papers are also the files people least want to upload to a cloud AI. A private NAS is the natural home for this job.

What it does

Heirloom is a Skill for UGREEN AI NAS that turns scattered family records into something a relative can follow without guessing.

• Keep mode: the owner catalogs chosen folders. Each document, scan or voice memo gets a source-linked card with its location, a short summary and any renewal or expiry dates.
• Gaps report: which important papers were found, not found in the folders shared, or unclear.
• Differing accounts: when relatives describe the same event differently, Heirloom shows each version side by side with its source and never picks a winner.
• Inherit mode: a relative or executor sees what exists, what is sensitive and what needs action, with a handover packet for each named person.
• Reruns: later runs handle only new or changed items.

Heirloom never moves, edits, tags or deletes an original. Sensitive numbers such as account, ID and policy numbers are recorded only as a category and a location, never copied. Every file is saved only after the user approves a preview.

How I built it

Heirloom is a Markdown Skill written to the UGOS Skill Development Standard. Its SKILL.md delegates, in order, to four registered experts: file, document, photo and knowledge. Each step is mapped to an action in the UGOS capability catalog: file listing and fingerprints, document extraction and cross-checks, text and dates from scans, and speaker-separated transcription of voice memos. Each delegated result is handled as completed, failed, cancelled, waiting for the user, or waiting for approval. The package passes the UGOS static checker with zero errors and zero warnings. I used Claude for research and drafting, and reviewed and edited everything myself.

Challenges

• No code allowed. Every safety rule had to live in the wording of the workflow.
• Sensitive data. The experts read the originals, so Heirloom asks them only for categories and locations, and says plainly that it cannot control what an expert holds while it works.
• Access control. A Skill cannot lock files or verify a death. The Inherit mode release checklist is a human step, and the Skill says so.
• Fitting the standard. My first draft used the wrong section names and capability format. The checker taught me what the platform expects.

What I learned

Designing for a platform with approval states changes how you write a workflow. A result counts only once the expert confirms it, never because the Skill planned it.

What's next

Heirloom has not yet run on a NAS. If I become a finalist, I will test OCR accuracy on scans, transcription of accented speech, cross-item linking, approval behavior and run time on a full folder, using an invented synthetic family dataset.

Built With

• claude
• markdown
• ugos
• ugos-skill-creation-tool
• ugreen-ai-nas
• uliya