The Agent Skills Directory

Command Execution: The skill utilizes kubectl commands to inspect node labels, taints, and status. This is an expected diagnostic procedure for troubleshooting GKE node disruptions and does not involve arbitrary command execution.- Indirect Prompt Injection: The skill is designed to ingest and interpret external data such as GKE logs and event descriptions.
Ingestion points: Data enters the context via query_logs calls and kubectl describe outputs.
Boundary markers: The skill does not define specific delimiters for separating log content from instructions.
Capability inventory: The agent uses read-only commands for cluster inspection and PromQL for monitoring metrics.
Sanitization: There is no explicit requirement to sanitize or filter log content before interpretation.
Context: Processing untrusted external data (like logs) is a standard requirement for observability tools, though it creates a theoretical surface where malicious strings in logs could attempt to influence agent behavior.

gke-ai-troubleshooting-handle-disruption-gpu-tpu