The Agent Skills Directory

[COMMAND_EXECUTION]: The skill is designed to execute various kubectl commands, providing extensive control over a Kubernetes cluster. This includes the ability to modify cluster states, such as scaling deployments, restarting services, and applying new manifests via the kubectl apply command.
[DATA_EXFILTRATION]: The skill interacts with the KUBECONFIG environment variable and accesses the default kubeconfig path (~/.kube/config), which typically contains sensitive cluster access credentials. Furthermore, commands such as kubectl get -o yaml and kubectl logs can retrieve and display potentially sensitive information, including Secrets, ConfigMaps, and application logs.
[PROMPT_INJECTION]: The skill demonstrates a vulnerability surface for indirect prompt injection through the processing of data from the Kubernetes environment.
Ingestion points: Data enters the agent's context through output from kubectl logs, kubectl get, and kubectl describe commands, which fetch logs and resource specifications directly from the cluster.
Boundary markers: The skill does not define specific delimiters or instructional boundaries to separate command output from the agent's own internal logic.
Capability inventory: The skill includes high-impact capabilities such as arbitrary command execution (via kubectl), port forwarding, and the ability to modify infrastructure components.
Sanitization: No validation or sanitization of the data retrieved from the cluster is performed, allowing potentially malicious instructions embedded in logs or resource metadata to influence agent behavior.

kubectl