The Agent Skills Directory

Command Execution: The skill references gcloud and kubectl as fallback methods for cluster management tasks. While these are standard tools for Kubernetes administration, they involve the execution of shell commands which should be performed in a secure, authenticated environment.
Indirect Prompt Injection: The skill identifies a surface for indirect prompt injection because it ingests cluster configuration data which could theoretically be manipulated to influence the agent's behavior.
Ingestion points: Data is ingested through cluster and resource description tools such as get_cluster and get_k8s_resource (SKILL.md).
Boundary markers: No explicit delimiters or instructions are provided to the agent to disregard potential instructions embedded within the ingested Kubernetes resource data.
Capability inventory: The skill utilizes the apply_k8s_manifest tool (SKILL.md), which permits modifying the cluster state based on processed information.
Sanitization: There are no explicit validation or sanitization steps mentioned for the data retrieved from the cluster APIs before processing.

gke-reliability