The Agent Skills Directory

Diagnostic Command Execution: The skill utilizes kubectl exec to perform connectivity tests from within user Pods to the GKE metadata server. This is a standard diagnostic practice for verifying Workload Identity functionality, though it involves executing commands within the workload environment.
Human-in-the-loop Remediation: The skill identifies root causes and generates remediation commands or manifest changes. It correctly maintains a security boundary by instructing the agent to propose these changes to a human operator for review and manual application rather than executing modifications automatically.
Indirect Prompt Injection Surface: The diagnostic workflow involves reading container logs (kubectl logs), which constitutes an ingestion point for external, potentially untrusted data. The skill mitigates risks by operating in a diagnostic, non-interactive mode and routing findings to a human for final evaluation.
Usage Attribution and Telemetry: The skill includes instructions to apply attribution tags to gcloud commands and HTTP headers. This represents a standard vendor telemetry practice for usage tracking and does not involve the collection or exfiltration of sensitive user data.

gke-workload-identity