vss-manage-alerts

vss-manage-alerts is an NVIDIA operational skill for driving the alert workflows of the Video Search and Summarization (VSS) blueprint. It solves the problem of operating a deployed VSS alerts pipeline end to end: real-time monitoring, Alert-Bridge subscription CRUD, Slack/webhook incident notifications, incident and alert queries, camera onboarding, and VLM-verifier prompt customization. It routes by the deployed mode (CV verification vs VLM real-time) combined with user intent, first probing the running containers to detect which mode is live before executing any workflow.

Capability-wise it relies on curl, jq, and Docker on the caller, uses NGC credentials supplied through environment variables for image pulls, and ships helper scripts under scripts/alert-notify/ — a FastAPI multi-backend webhook server that receives VSS incident alerts and fans them out to Slack and/or an OpenClaw dashboard backend. Slack integration is configured through standard environment variables (a bot OAuth token with chat:write scope and a channel ID) documented in an .env example, not hardcoded secrets. The skill includes routing tables, step-by-step workflows, runnable eval manifests, and troubleshooting for connection, authentication, and GPU out-of-memory failures.

It targets developers and operators running an NVIDIA VSS deployment who need to manage alerting and notifications. It performs deployment and container operations and orchestrates notifications, so it is an operational tool rather than a read-only one, but it handles credentials via official environment-variable conventions with no exfiltration, gates deployment actions behind user confirmation or a trusted auto-deploy flag, and NVSkills-Eval reported a 100% Security score.