What problem does it solve? Turning a recorded video clip into a concise narrative summary with timestamped events normally requires manual review or custom VLM prompting; this Skill routes the request to the NVIDIA VSS long-video-summarization (LVS) microservice with human-in-the-loop scenario/event targeting, or falls back to a direct VLM call when the service is unavailable. ## Core Features & Use Cases - Availability-based routing: Probes /v1/ready on the LVS service (port 38111) and chooses the LVS microservice path or the VLM /v1/chat/completions fallback automatically. - HITL-gated summarization: Collects scenario, events, and optional objects_of_interest from the user before calling POST /v1/summarize, with a sanctioned defaults opt-in for autonomous runs. - Verbatim output rendering: Returns the service's video_summary and timestamped events without paraphrasing, including a fallback warning when only the VLM path was used. - Use Case: An operator asks to summarize an uploaded warehouse video with scenario 'warehouse monitoring' and events like 'boxes falling' and 'forklift stuck'; the Skill fetches the clip URL via VIOS, calls the LVS service once, and renders the summary plus event list. ## Quick Start Ask the agent to summarize a recorded video clip, providing a scenario and the events you want surfaced, and it will probe the LVS service and produce the summary.
