The Agent Skills Directory

[Secure Configuration Defaults]: The skill emphasizes 'golden path' security configurations, such as defaulting to Autopilot mode, enabling private nodes, and disabling public control plane endpoints, which aligns with security best practices for GKE.
[Trusted Dependencies]: It references official and well-known client libraries (e.g., the official Python kubernetes client and the @kubernetes/client-node package) for programmatic interaction with cluster APIs.
[Tool Hierarchy and Auditing]: The skill defines a clear preference hierarchy, prioritizing structured MCP tools over raw shell commands. This approach facilitates better auditing and reduces the risks associated with manual shell intervention.
[Indirect Prompt Injection Surface]: The skill includes capabilities to retrieve cluster events and container logs (via get_k8s_logs and list_k8s_events).
Ingestion points: Untrusted data enters the agent context through cluster logs and event messages retrieved in references/cli-reference.md and references/mcp-usage.md.
Boundary markers: While the instructions focus on retrieval, the platform's underlying guardrails provide standard isolation for this data.
Capability inventory: The skill can perform cluster mutations including creating, updating, and deleting resources across all scripts.
Sanitization: The skill relies on the standard output formats of kubectl and GKE MCP tools, which are typical for administrative operations.
[Administrative Command Usage]: The inclusion of commands like kubectl exec and kubectl port-forward is documented as a fallback for specific administrative tasks. While these are high-privilege operations, they are presented as standard tools for Kubernetes management within the skill's intended scope.

gke-basics