vss-setup-video-analytics-api

This skill deploys the `vss-video-analytics-api` REST service (a Node.js API from NVIDIA's VSS blueprint) as a standalone Docker container rather than as part of the full warehouse stack. It solves the narrow problem of running just the analytics REST API pointed at a chosen JSON config, a data-log bind mount, and an operator's own Elasticsearch and optional Kafka, without bringing up the entire blueprint. A minimal bring-up sets the apps and data directories, creates the data-log path, runs `docker compose up` for the single service, and probes `/livez`.

The skill routes to detailed references for the full deploy workflow and a field-by-field JSON configuration guide covering the server port (default 8081), Elasticsearch connection and retries, Kafka brokers (settable to an empty array to run broker-less), and application tuning knobs such as POST body size and AMR retention. It documents three config sources (image-baked default, service-shipped config, and a custom bind-mounted config) and includes troubleshooting for connection-refused, NGC 401/403 auth failures, and GPU out-of-memory conditions.

Target users are operators and agents who want the REST API alone, want to supply their own config, or want to point the API at a different Elasticsearch/Kafka. The skill is notably security-conscious about the `NGC_CLI_API_KEY`: it warns the key is a long-lived credential, tells users never to commit it, paste it into chat, or store it in `/tmp`, and recommends reading it interactively or from a secret manager and writing derived `.env` files with restrictive permissions.