Inspiration
Indonesian families and small businesses download invoices, receipts and payment screenshots from email and WhatsApp onto a shared NAS, and many cannot tell a real invoice from a phishing copy. One urgent fake bill can cause a costly mistaken payment.
What it does
Safe Inbox Guardian checks a NAS inbox folder for phishing-style files, explains each risk with evidence, and quarantines confirmed risky files only after user approval. It hashes and dedupes files, extracts sender, payee, amounts, dates and links, compares them with related contracts and receipts, reads scanned images with OCR, and produces a High, Medium and Low risk report. Nothing moves or is deleted without approval.
How we built it
We built it as a Uliya Skill for UGREEN AI NAS, following the official Uliya Skill Development Standard. The SKILL.md orchestrates five delegations across four expert agents: file, document, photo and knowledge agents, using only capabilities from the official UGREEN capability catalog v2.2.0. A local Python simulator tested the risk logic on 6 sample files, and all 6 matched the expected rating.
Challenges we ran into
Round 1 provides no NAS access, so we could not test on a real UGREEN AI NAS yet. We designed the skill to be honest about that: low-confidence OCR results stay Medium risk at most, and nothing is claimed as NAS-tested.
Accomplishments that we're proud of
The skill package passes the official Uliya static checker with 0 errors and 0 warnings, and the Round 1 proposal answers all 10 questions in 459 words.
What we learned
A defensive NAS skill needs approval gates and evidence more than automation. Reporting uncertainty honestly is part of the product, not a weakness.
What's next for Safe Inbox Guardian
If selected as a finalist, validate the skill on the assigned UGREEN AI NAS, record the under-3-minute demo video, and refine the risk report from real NAS test results.
