skills/googlecloudplatform/gke-mcp/gke-ai-troubleshooting-handle-disruption-gpu-tpu/Gen Agent Trust Hub
gke-ai-troubleshooting-handle-disruption-gpu-tpu
Warn
Audited by Gen Agent Trust Hub on Jul 30, 2026
Risk Level: MEDIUMPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [PROMPT_INJECTION]: The skill contains a directive in SKILL.md (Step 0) that explicitly instructs the agent to bypass human-in-the-loop verification. It mandates that if specific project context is provided, the agent must immediately invoke tools without asking for confirmation or presenting an execution plan. This overrides platform-level safety features designed to ensure user oversight.- [COMMAND_EXECUTION]: The workflow involves running kubectl commands to inspect node labels (Step 1). While relevant to diagnosing GKE issues, this grants the agent direct interaction with the cluster's administrative interface.- [PROMPT_INJECTION]: Indirect Prompt Injection vulnerability assessment:
- Ingestion points: Data enters the agent context through query_logs (GKE logs) and query_prometheus (monitoring metrics) tools.
- Boundary markers: No delimiters or instructions to ignore embedded instructions are present in the processing logic.
- Capability inventory: The agent has capabilities to execute kubectl commands and call diagnostic tools (get_k8s_resource, query_prometheus, query_logs).
- Sanitization: No sanitization or validation of the external log/metric content is performed before processing.
Audit Metadata