The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill enables the agent to read unvalidated data from the Kubernetes cluster, such as container logs and cluster events. This creates a surface where a compromised application or an attacker with cluster access could emit malicious instructions in logs to influence the agent's behavior.
Ingestion points: The skill uses kubectl logs, kubectl get events, and kubectl describe pod to fetch data from the cluster environment in SKILL.md and references/details.md.
Boundary markers: No boundary markers or specific "ignore instructions" delimiters are provided to the agent when fetching these logs or events.
Capability inventory: The skill includes high-privilege write and execution capabilities, including kubectl apply, kubectl exec, kubectl edit, kubectl delete, and kubectl scale.
Sanitization: There is no evidence of sanitization, filtering, or validation performed on the fetched cluster data before it is presented to the agent context.

kubernetes-ops