The Agent Skills Directory

Indirect Prompt Injection (HIGH): The skill creates a high-risk attack surface by consuming untrusted data from the Kubernetes cluster while possessing administrative capabilities.
Ingestion points: The tools get_pod_logs, get_events, and describe_pod (referenced in SKILL.md) ingest free-text data from the cluster environment.
Boundary markers: There are no explicit instructions or delimiters mentioned to separate untrusted data from agent instructions.
Capability inventory: The skill includes high-privilege tools such as delete_pod (with force=True), rollback_deployment, and rollback_helm_release (referenced in SKILL.md).
Sanitization: No evidence of sanitization or filtering of log/event content before processing.
Privilege Escalation (HIGH): The skill explicitly provides tools for destructive and state-altering operations in a production environment.
Evidence: delete_pod(name, namespace, grace_period=0, force=True) in SKILL.md provides direct capability for resource destruction.
Evidence: rollback_deployment and rollback_helm_release allow the agent to modify the cluster's desired state without additional confirmation layers.

k8s-incident