vss-manage-alerts

Manage VSS alert subscriptions, real-time monitoring, Slack notifications, and incident queries.

2|Updated Aug 20, 2026
One-click install
npx skills add https://github.com/atomicrajat/industry_safety_monitoring_system --skill vss-manage-alerts-atomicrajat
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: vss-manage-alerts
Source: https://github.com/atomicrajat/industry_safety_monitoring_system/tree/main/.claude/skills/vss-manage-alerts
Command: npx skills add https://github.com/atomicrajat/industry_safety_monitoring_system --skill vss-manage-alerts-atomicrajat

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires fastapi, uvicorn, slack_sdk, httpx, and includes scripts (resource) and references (resource) components.

What problem does it solve? Operating an NVIDIA VSS alerts deployment requires knowing which mode is live (CV verification vs VLM real-time), which endpoint serves which intent, and how to safely create, list, stop, and query alert rules without mixing up incidents and subscriptions. This Skill routes every alert-related request to the correct workflow and API. ## Core Features & Use Cases - Mode-aware routing: Detects the deployed alerts mode (CV verification or VLM real-time) and routes requests to the correct workflow, refusing unsupported operations with a redeploy hint. - Alert subscription management: Creates, lists, and stops persistent Alert Bridge realtime rules with sensor resolution via VIOS/VST and a mandatory confirmation gate before deletion. - Slack notifications and incident queries: Starts and manages the alert-notify webhook server for Slack delivery, and answers "what happened" questions through the VSS Agent /generate endpoint. - Use Case: A user says "Watch warehouse_sample for anyone without a safety vest" — the Skill resolves the sensor to its RTSP stream, derives a ppe_vest_violation tag, and POSTs a canonical rule to Alert Bridge, returning the rule UUID. ## Quick Start Ask the assistant to set up a realtime alert on a named camera for a specific detection condition, such as monitoring warehouse_sample for PPE violations.

Frequently Asked Questions about vss-manage-alerts

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I create a realtime alert rule on a camera in NVIDIA VSS?

In VLM real-time mode, name the sensor and detection condition; the Skill resolves the sensor ID and RTSP stream via the VST API, derives a snake_case alert_type tag, and POSTs the rule to Alert Bridge at /api/v1/realtime with chunk_duration 30 and a yes/no system prompt.

What is the difference between CV verification mode and VLM real-time mode in VSS alerts?

CV verification mode runs the static Grounding DINO pipeline with VLM verdict verification, while VLM real-time mode runs only alert-bridge and rtvi-vlm for dynamic prompt-based alerts. Subscription CRUD and Slack notifications are only available in VLM real-time mode.

How do I stop or delete an alert rule in VSS?

Say "stop the <type> alert on <sensor>" and the Skill finds the matching rule via GET /api/v1/realtime, then asks a yes/no confirmation with the rule ID. Only after you confirm does it issue the DELETE request — the confirmation gate always applies.

Can I send VSS incident alerts to Slack?

Yes, in VLM real-time mode the Skill starts the alert-notify webhook server on port 9090, which receives incidents and posts rich messages to a Slack channel using a bot token and channel ID. It supports health checks, status, and test notifications.

Why does the Skill refuse to create an alert subscription on my deployment?

Alert subscriptions and Slack notifications require VLM real-time mode; on a CV verification deployment the Skill refuses and suggests redeploying with /vss-deploy-profile -p alerts -m real-time. It never auto-redeploys because a mode switch tears down running monitoring.

How do I query past incidents or alerts in VSS?

Ask a natural-language question like "any alerts today?" and the Skill sends it to the VSS Agent /generate endpoint, which dispatches to the incident store. It never answers from memory, and an empty result is reported as zero incidents rather than an error.