The Agent Skills Directory

Command Execution: The skill utilizes diagnostic commands, including gcloud and kubectl, to retrieve cluster status, pod events, and storage configuration. These operations are essential for the skill's primary purpose of diagnosing persistent storage failures.
Telemetry and Attribution: The skill includes instructions to prefix gcloud commands with an environment variable and append a specific string to User-Agent headers for usage attribution. This is a common mechanism for tracking the performance and usage of vendor-provided tools and does not involve the collection of sensitive user data.
Indirect Prompt Injection Surface: The skill ingests untrusted data from external sources such as cluster events, pod descriptions, and container logs. While this presents a potential surface for indirect prompt injection, the skill mitigates this by enforcing a read-only diagnostics boundary and recommending GitOps-based corrections rather than directly modifying cluster state.
Ingestion points: Data enters the agent context through commands like kubectl describe pod, kubectl get events, and kubectl describe node (SKILL.md).
Boundary markers: The skill explicitly instructs the agent to maintain a "read-only diagnostics boundary" and uses a "Dry-Run Fallback" mode (SKILL.md).
Capability inventory: Capability to execute commands inside pods (kubectl exec) and retrieve cluster credentials (gcloud container clusters get-credentials) is documented (SKILL.md).
Sanitization: The skill relies on specific diagnostic flags and focuses on read-only operations to minimize risks associated with processing external content.

gke-storage-troubleshooting