The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes administrative Kubernetes commands to perform diagnostics.
It executes kubectl exec to query internal Prometheus metrics via wget on localhost:9090.
It provides commands to modify namespace labels (kubectl label), including an emergency procedure to disable network policy enforcement.
It accesses Kubernetes configuration files stored at ~/.kube/<cluster>.yaml for cluster authentication.
[PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection by processing untrusted data from the cluster environment.
Ingestion points: Retrieves potentially untrusted content via kubectl logs, kubectl get events, and hubble observe.
Boundary markers: Does not utilize explicit delimiters or markers to isolate log or event data from the agent's instructions.
Capability inventory: Features administrative capabilities including resource labeling, container execution (kubectl exec), and Flux lifecycle management.
Sanitization: Does not implement sanitization or filtering for data retrieved from Kubernetes logs or event streams.

sre