VSS Alert Management
Operate NVIDIA VSS real-time alerts, subscriptions, incident queries, Slack notifications, and camera onboarding.
The document discloses data flows across VSS, Alert Bridge, VIOS, Slack, NGC credentials, and external services, and requires redeploy confirmation, two-step rule deletion, sanitized errors, and Agent-mediated routing. However, the webhook defaults to binding 0.0.0.0, credentials are injected through environment or configuration, mode switching tears down monitoring, and rollback is not defined, so full marks are not justified.
Routing precedence, mode detection, API parameters, abnormal cases, and user-facing error handling are fairly detailed. However, key workflows depend on external skills, deployment state, services, and scripts not included in the supplied evidence. Static review cannot reproduce execution, and no committed CI workflow with key-path test coverage is shown, so the score is capped and reduced.
The intended audience, alert monitoring, rules, queries, Slack operations, and CV/VLM boundaries are clearly described, including ambiguity handling. Fit is nevertheless narrow because it requires a specific NVIDIA VSS deployment, Docker, GPU capacity, NGC credentials, and reachable services; Chinese-language interaction and mainland-China network compatibility are not addressed.
The skill has clear sections for purpose, prerequisites, limitations, troubleshooting, version, author, Apache-2.0 license, GitHub provenance, examples, and cross-skill links. It lacks a clear changelog, explicit maintenance commitment, update procedure, and locked dependency specification, and several referenced files and scripts are absent from the supplied material.
The workflows directly cover alert creation, querying, rule management, camera onboarding, and Slack notifications, with usable curl examples, routing rules, and output requirements. Actual value still depends heavily on deployment state, external microservices, and unprovided implementation files, so static evidence cannot support a higher score than the allowed ceiling.
Version metadata, a benchmark report, and multiple evaluation manifests provide some audit trail. The benchmark is still an in-repository claim, with no verifiable CI workflow, test implementation, execution logs, or independent corroboration demonstrating reproduction of the key paths.
- Do not switch between CV and VLM modes without confirmation; switching tears down the deployment and loses active monitoring state.
- Slack and NGC credentials are sensitive and should not be exposed in logs, command output, or unprotected .env files; the webhook's default 0.0.0.0 binding should be restricted in production.
- Rule deletion requires confirmation, but deployment teardown and webhook stopping lack a unified rollback mechanism.
- Core functionality depends on NVIDIA VSS, Docker, GPU capacity, NGC, and reachable external services; the material does not establish stable operation on mainland-China networks or with Chinese-language requests.
What does this skill do, and when should you use it?
vss-manage-alerts is for the alerts profile of NVIDIA Video Search and Summarization. It routes requests across CV verification and VLM real-time deployments for monitoring, Alert Bridge subscriptions, incident lookup, Slack notifications, camera onboarding, and verifier-prompt customization. It requires a reachable VSS deployment, VIOS sensors, curl, jq, and Docker. Persistent subscriptions and Slack operations require VLM real-time mode, while switching modes tears down the current deployment and stops active monitoring.
Checks Docker containers or generated.env to identify the deployed alerts mode; sends monitoring and incident-query requests through the VSS Agent POST /generate endpoint; guides creation, listing, and deletion of persistent Alert Bridge rules; guides Slack webhook setup, status checks, and test messages through alert-notify; registers RTSP/IP cameras in VIOS; handles CV static-alert onboarding and verifier-prompt customization; and uses curl, jq, Docker, referenced playbooks, and named helper scripts.
- A VSS operator needs to start or stop a one-time real-time monitoring session for a named sensor without a detection condition.
- A system administrator needs to create, inspect, or delete a persistent alert rule for a specific sensor and condition in VLM real-time mode.
- An incident responder needs alerts for a sensor, time range, or detection type queried from the deployed system.
- A platform administrator needs to configure, check, or test Slack alert notifications in VLM real-time mode.
- A video-system administrator needs to register an RTSP or IP camera in VIOS so the alerts pipeline can use it.
- A CV-mode operator needs to onboard static alerts or customize verifier prompts.
What are this skill's strengths and limitations?
- Covers monitoring, persistent subscriptions, incident queries, Slack notifications, and camera onboarding in one alert-operations workflow.
- Clearly separates CV verification and VLM real-time capabilities and defines refusal and redeployment behavior for incompatible requests.
- Requires live querying for incident questions instead of answering from memory.
- Includes troubleshooting paths for connection failures, NGC authentication errors, and GPU memory problems.
- It is not standalone; it requires a reachable VSS alerts profile and matching microservices.
- Alert Bridge subscriptions and Slack notifications are unavailable in CV mode, and switching modes stops current monitoring.
- NGC-hosted models and NIMs may involve rate limits, GPU-memory requirements, and license terms.
- The supplied source provides no independent test results or cross-platform execution evidence.
How do you install this skill?
Install the individual skill with the NVIDIA skills CLI:
npx skills add nvidia/skills --skill vss-manage-alerts --yes
The README does not specify the resulting installation directory; the CLI prompts for the skill and destination. Reload the agent's skills after installation.
How do you use this skill?
Ensure an alerts profile is deployed and reachable at $HOST_IP, and provide $NGC_CLI_API_KEY, $NVIDIA_API_KEY, curl, jq, and Docker. Example prompts include "Start real-time alert for sensor warehouse_sample", "Show me recent alerts for sensor warehouse_sample", and "Set up Slack notifications". The skill detects the deployment mode before routing the request. Subscription and Slack requests require VLM real-time mode, while incident queries are sent through the VSS Agent /generate endpoint. Cameras must already be registered in VIOS.