home-sentinel is an AI Skill for UGREEN AI NAS that turns surveillance camera events into a plain-language security report for a period the user chooses.
Asked "what happened at home this week?", it reads earlier reports stored on the NAS, lists the cameras, pulls the event timeline, resolves identities, attaches timestamped evidence, flags anomalies against explicit rules, and saves the draft only after the user confirms.
The report contains:
- a chronological timeline of real events, with timestamps and cameras;
- an identity status for people and vehicles: known match, unrecognized, unresolved or not requested;
- timestamped evidence references for the key events, with their expiry time;
- anomalies flagged by explicit rules, each event-backed anomaly citing a real event identifier;
- coverage notes naming each camera and what happened with it;
- a comparison with earlier comparable reports read back from the NAS.
A parent SKILL.md runs five ordered steps (baseline, collection, identity, evidence, save) and delegates each one to a registered expert agent: ug-surveillance to security-monitor-agent, ug-filesystem to file-agent, ug-document to document-agent, ug-knowledge to knowledge-agent. The parent never calls cli.invoke directly and never forwards a placeholder: every identifier, timestamp, folder and filename comes from the user input or from a real previous result.
The Skill ships no model and calls no model API. It stores no credential and makes no external request. Reads need no approval; face and plate recognition require explicit consent; saving requires workflow approval plus user confirmation. The report keeps time, camera, topic, identity status and event id, and for Round 2 plate text, never an image.
Out of scope: live viewing, camera configuration, face enrolment, deleting or moving recordings.
Development status: the Markdown package (SKILL.md and four references under references/), the delegation prompts, the anomaly rules, the report template and the empty, partial and truncated handling are built, and the official Uliya checker reports zero errors and zero warnings. NAS execution is not built yet and is pending finalist access to an assigned UGREEN AI NAS.
Built With
- markdown
- ug-document
- ug-filesystem
- ug-knowledge
- ug-surveillance
- ugos
- ugreen-ai-nas
- uliya-skill