Use airunway-aks-setup for initial GPU node-pool, controller, provider, and
model setup. Consider the optional aks-gpu-inference skill only for a Day-2
incident affecting an existing AKS GPU or inference deployment.
Read the host-provided available-skill inventory, or use an approved
read-only host capability query. Do not require a CLI on a skill-only remote
host and do not infer availability from an prefix.
If aks-gpu-inference is available, invoke it through the host-native skill
mechanism.
If it is absent and would help, explain the focused GPU/inference diagnostic
benefit and ask before installation. Use only the installation path
supported by the current host after approval.
If installation is declined or unsupported, continue the base setup/error
workflow and reason from supplied evidence. Do not claim that the optional
skill ran.
Installation or invocation is not approval to mutate the cluster. Preserve the
separate confirmation gates for install, deployment, and remediation actions.