Inspiration
Homeowners with UGREEN cameras never watch hours of footage. They only learn something happened after a break-in, a delivery dispute, or a stranger at the door, because reviewing every clip is too slow and boring.
What it does
Home Watch Digest turns a week of your UGREEN AI NAS camera events into a short, plain-language digest — what was normal, what was unusual, and where the evidence is — saved on the NAS only after you approve.
The Skill counts and timelines the week's events, flags the unusual ones — unknown people, night motion, unfamiliar vehicles — resolves known faces and plates from the NAS library, and writes a dated summary with per-event camera, time, and evidence links.
How it works
You ask for the digest, or it runs weekly. The Skill confirms the time range and cameras, pulls the week's events, builds the digest, shows it to you, and — only with your approval — saves it on the NAS.
How we built it
Markdown-only orchestration per the Uliya Skill Development Standard (catalog 2.2.0, kit 0.7.7). The parent Skill delegates self-contained tasks to security-monitor-agent and knowledge-agent, passing only verified results between steps. All surveillance reads need no approval; saving needs workflow approval. No scripts, no external services; the NAS local model summarizes; footage and faces never leave the NAS.
UGREEN AI NAS features used: ug-surveillance (cameras, event-group-counts, event-timeline, event-evidence, identity-resolve) in steps 2-5 to list cameras, count and timeline events, fetch evidence, and resolve identities; ug-knowledge save-text in step 7 to save the approved digest.
What makes it different
Camera apps show endless clip lists and generic motion alerts; cloud summaries send your footage off-site. This Skill is local-first, evidence-linked, and honest about uncertainty: it reports "unknown person" instead of guessing, and it never moves footage off the NAS.
What we built so far
The full Uliya package — SKILL.md plus digest-format, test-plan, and disclosure references — checked with the official kit checker (0 errors, 0 warnings). Nothing in the workflow is missing on paper. Still needs NAS testing: real event-timeline output shapes, identity-resolve behaviour on unknown faces, and save-text paths on the assigned NAS.
Testing plan if finalist
Core 1: a normal week (known family faces, daytime motion) produces a digest that says "nothing unusual" with correct counts. Core 2: a planted night-time unknown-person event is flagged "unusual" with camera, time, and evidence link. Edge: a camera with zero events reports "no events recorded" and stops — no invented content.
Known limits
It reads only events the NAS already recorded — it cannot see what cameras missed; identity resolution is limited to the NAS face and plate library, so strangers stay "unknown"; digest quality depends on camera placement and motion settings. No outside dependencies; no third-party services.
Disclosures
Third-party components and licenses: none — the package is Markdown only and uses only UGREEN's own capability catalog; no third-party code is bundled. AI tools used: Muse (Meta AI) assisted with research and drafting; the entrant reviewed and approved the package.
Built With
- ai
- nas
- ug-knowledge
- ug-surveillance
- ugreen