gke-ai-troubleshooting-tpu-partition-overlap
Pass
Audited by Gen Agent Trust Hub on Oct 15, 2026
Risk Level: SAFE
Full Analysis
- Read-Only Diagnostic Focus: The skill explicitly contains a 'Read-only rule' that restricts the agent to diagnostic commands. It prohibits draining, deleting, or re-creating nodes, ensuring that all destructive actions must be performed manually by a human user through official GKE documentation links.
- Standard Tool Utilization: The skill relies on well-known administrative tools such as
gcloud,kubectl, and standard Monitoring (PromQL) queries. These are standard interfaces for managing Google Cloud resources and do not involve suspicious or obfuscated execution patterns. - Vendor-Owned Resources: All URLs and documentation references point to official Google Cloud (
cloud.google.com) or Google Kubernetes Engine repositories and documentation, aligning with the expected vendor context for a Google-authored skill. - Indirect Prompt Injection Surface: While the skill processes user-supplied parameters like
{project_id}and{slice_name}to form diagnostic commands, the scope is limited to read-only status checks and predefined metric queries, minimizing the risk associated with untrusted data ingestion.
Audit Metadata