The Agent Skills Directory

[Indirect Prompt Injection] (HIGH): The skill defines patterns for interacting with external data (Kubernetes resource names) using shell commands.
Ingestion points: The agent is instructed to use placeholders like <pod-name> and <namespace> which are typically sourced from untrusted user input or cluster state.
Boundary markers: None present in the instructions to prevent the agent from obeying instructions embedded in resource names or log outputs.
Capability inventory: Includes kubectl exec, kubectl logs, and kubectl run, providing a direct path from untrusted input to shell execution.
Sanitization: No sanitization or validation of the input variables is described or required.
[Command Execution] (HIGH): The troubleshooting section explicitly directs the use of kubectl exec -it <pod-name> -- /bin/sh and kubectl run debug --image=nicolaka/netshoot. These commands provide interactive shell access and allow the deployment of new containers, which are high-privilege actions that could be abused to bypass security controls or exfiltrate data from the cluster environment.

kubernetes-operations