Inspiration
Coming home from a trip, the first question is always the same: "What happened while I was away?" Cloud camera apps answer it by uploading your footage to someone else's servers, and raw event lists bury the answer in dozens of clips. The UGREEN AI NAS already holds the cameras, the events, and the compute — so the answer should be produced at home, privately, in one page.
What it does
Away-Day Digest turns home-camera events into a one-page "while you were away" digest: what happened, who was seen, and the clips that prove it. The user gives a time range (default: last 7 days) and a camera scope (default: all) and confirms where to save the digest. The Skill gathers the period's events and evidence from the NAS surveillance system, matches faces and plates against the household's own on-NAS library, and writes a digest with counts by category, a timeline of notable events, known versus unknown people and vehicles, and what needs attention. Unknown stays unknown — the Skill never guesses an identity and never invents an event. If there are no events in the period, it says so and stops. Everything is processed on the NAS; no footage leaves the home.
How we built it
We started from the official Uliya Skill Development Kit (v0.7.7, capability catalog v2.2.0) and confirmed every capability we use exists in the catalog: ug-surveillance (cameras, event-group-counts, event-timeline, event-evidence, faces, face-detail, identity-resolve, vehicle-plates, vehicle-plate-detail) and ug-knowledge (save-text). The parent Skill only delegates — to the security-monitor-agent for evidence and identity work, and to the knowledge-agent for saving the digest — and reasons over the real results returned. The package is SKILL.md plus four Markdown references (digest format, identity rules, disclosure, test plan), written to the Skill template's five sections. We validated it with the official checker until it passed with zero errors and zero warnings, and we built a small Python standard-library harness that simulates the workflow end to end on invented sample data to prove the digest logic runs.
Challenges we ran into
The hardest part was honesty by design: a digest that guesses is worse than no digest, so the Skill had to define exactly what happens when evidence is missing, a count group fails, or a face has no library match — and stop instead of filling the gap. The second challenge was the checker's strict contract: exact section headings, single-quoted delegate commands with registered agents only, capability evidence in the exact "- -> " form, and no scripts, paths, or credentials anywhere in the package. The third was the 500-word proposal limit: saying what the Skill does, how it differs, and how it will be tested, in 498 words.
Accomplishments that we're proud of
A complete pre-test Skill package that passes the official checker with 0 errors and 0 warnings ("static-pass"). A differentiated concept: surveillance is the least-used NAS capability domain, and this Skill turns it into the most privacy-sensitive use case for local-only processing. A local simulation proving the workflow logic end to end, with the empty period edge case handled correctly. A 498-word proposal answering all 10 official questions.
What we learned
Local-first design changes every decision: when nothing may leave the home, the digest can only contain classifications and evidence references, never images — and that constraint makes the product clearer, not weaker. We also learned that a Skill is a contract, not code: its value is in the delegation boundaries, the empty-result rules, and the failure handling, all of which have to be written down before any NAS ever runs it.
What's next
If selected as a finalist, we will test on a real iDX6011 Pro during the testing day: a seven-day period with known and unknown visitors, a known visitor at the front door only, and an empty period as the edge case. After that, the natural next steps are a scheduled weekly digest, a morning-after summary, and an alert digest that only fires when an unknown subject is seen at night.
Built With
- ug-knowledge
- ug-surveillance
- ugos
- ugreen-ai-nas
- uliya-skill-kit
