What problem does it solve? Deploying NVIDIA Video Search and Summarization (VSS) profiles involves dozens of interdependent decisions — profile selection, GPU sizing, LLM/VLM placement, credential gating, compose resolution, and readiness verification — where a single misstep surfaces as a 20-minute NIM cold-start failure. This Skill encodes the full deployment runbook so an agent executes the correct sequence every time. ## Core Features & Use Cases - Profile routing and sizing: Maps user intent to the base, alerts, lvs, search, warehouse, or edge profile, then applies per-profile GPU sizing tables and model placement rules (local, local_shared, remote). - Guarded deployment flow: Enforces credential validation, dry-run compose resolution, dangling depends_on normalization, NGC entitlement probes, and a human confirmation gate before docker compose up -d. - Readiness and debugging: Gates success on actual endpoint health checks rather than container start, with per-profile troubleshooting references for NIM probes, OOM tuning, and mode-specific failures. - Use Case: A user asks to deploy the VSS alerts profile on a 2-GPU RTX PRO 6000 host for real-time VLM alerting; the Skill stages perception models, writes env overrides, resolves and normalizes the compose file, deploys, and verifies the agent, RT-VLM, and alert-bridge endpoints. ## Quick Start Deploy the VSS base profile on this machine end-to-end using the vss-deploy-profile skill.
