What problem does it solve? Onboarding researchers onto a shared GPU Kubernetes cluster usually means ad-hoc tickets, inconsistent quotas, and tenants who can interfere with each other or consume unbounded GPU budget. This Skill stamps out complete, bounded tenant environments in minutes so every researcher gets fair GPU access with proper isolation. ## Core Features & Use Cases - Namespace bundle templating: Generates Namespace, ResourceQuota (including requests.nvidia.com/gpu), LimitRange, default-deny NetworkPolicy, Pod Security labels, and cost-chargeback labels as one declarative bundle. - RBAC role patterns: Provides namespace-scoped Role/RoleBinding templates for researchers, team leads, and CI service accounts following CIS benchmarks (no wildcards, no cluster-admin). - JupyterHub GPU workbench: Configures Zero-to-JupyterHub profile lists mapping CPU-only, time-sliced GPU, dedicated GPU, and MIG-slice profiles with idle culling. - Tenancy escalation guidance: Defines explicit criteria for when a tenant should move from namespace-as-a-service to vCluster or a dedicated cluster. - Use Case: A platform engineer needs to onboard a vision-lab team of five researchers with a 4-GPU quota, isolated networking, and notebook access. The Skill produces the full YAML bundle, RBAC bindings, and JupyterHub profile config, committed to Git for GitOps delivery. ## Quick Start Ask the agent to onboard a new research team onto the shared GPU cluster with a specific GPU quota, storage need, and list of users, and it will generate the complete namespace bundle, RBAC, and JupyterHub configuration.
