The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses kubectl and make to manage Kubernetes resources and execute local build targets. These commands are necessary for the skill's primary function of cluster configuration and model serving setup.
[INDIRECT_PROMPT_INJECTION]: The skill collects cluster state information, such as node labels and pod statuses, which is then used to generate hardware compatibility recommendations and deployment steps. This represents an attack surface where compromised cluster metadata could influence agent behavior.
Ingestion points: The skill queries the cluster using kubectl get nodes, kubectl get pods, and kubectl describe nodes in references/steps/step-1-verify.md and references/steps/step-4-provider.md.
Capability inventory: The skill can perform cluster-wide modifications using kubectl and execute local scripts using make.
Boundary markers: No delimiters or explicit warnings are used when processing cluster-derived data.
Sanitization: Cluster-derived strings are used directly in logic without explicit sanitization or validation against a strict schema.
[METADATA_POISONING]: The skill contains inconsistent author information, with SKILL.md attributing the work to Microsoft while SKILL.json lists Anthropic Marketplace 3p as the author.

airunway-aks-setup