attacking-eks-gke-aks
Warn
Audited by Gen Agent Trust Hub on Jul 30, 2026
Risk Level: MEDIUMCOMMAND_EXECUTIONCREDENTIALS_UNSAFEDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
- [CREDENTIALS_UNSAFE]: The skill documents methods to extract and use sensitive credentials. It includes commands to read service account tokens at
/var/run/secrets/kubernetes.io/serviceaccount/tokenand node-level Kubelet client certificates at/var/lib/kubelet/pki/kubelet-client-current.pem. - [COMMAND_EXECUTION]: The skill provides templates for privilege escalation and host escape. It includes a YAML manifest for a privileged pod using
hostNetwork: true,hostPID: true, and ahostPathmount of the root directory, enabling full control over the underlying node. - [DATA_EXFILTRATION]: The skill covers querying cloud metadata services (IMDS) to retrieve IAM roles and security tokens from AWS, GCP, and Azure environments.
- [PROMPT_INJECTION]: The skill possesses an indirect prompt injection attack surface as it instructs the agent to process untrusted output from cluster enumeration commands (e.g.,
kubectl get secrets) and tool results without explicit sanitization or boundary markers to separate external data from its primary instructions.
Audit Metadata