Inspiration
When something breaks or a household needs its purchase records, the evidence is usually scattered across receipts, phone photos and folders. A generic AI summary can confidently connect the wrong item. HomeProof makes the evidence and uncertainty visible.
What it does
HomeProof is an initial UGREEN AI NAS Skill that organizes purchase records into a reviewable household inventory. It links evidence, keeps ambiguous matches separate, highlights missing documents and optionally saves an approved Markdown report. It does not establish ownership, warranty eligibility, insurance coverage or current value.
How we built it
New Markdown-only workflow authored on 9 November 2026 against the official UGOS Skill Creation Tool 0.7.3 and capability catalog 2.0.0. Registered file, document, photo and knowledge experts handle scoped file discovery, extraction, label OCR and approved saving. No external API or custom executable is bundled.
OpenAI Codex assisted with authoring the workflow, documentation and proposal. The official starter kit is a development resource and is not redistributed; its supplied archive contains no separate LICENSE file. The new contribution is the evidence-matching, provenance, ambiguity and report-review workflow.
Challenges
Distinguishing item prices from receipt totals, avoiding duplicate purchases from repeated scans, keeping serial ambiguity visible, and reconciling an interrupted save without producing duplicate outputs.
Accomplishments
The initial package passes the official static checker with zero errors and warnings. It includes explicit boundaries, registered expert delegation, source-change handling, report approval and finalist validation scenarios.
What is next
If selected, install and test on the assigned UGREEN NAS, establish the supported local model route, measure extraction and matching outcomes using synthetic fixtures, verify save reconciliation, and record the required real-device demonstration.
Development status
Round 1 authoring package only. No target-device installation, runtime execution or live deployment is claimed.
