The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill suggests a command curl -s http://10.0.20.10:5000/v2/_catalog | python3 -m json.tool. Piping data from a network source directly into an interpreter is a significant security risk that can lead to remote code execution if the source is compromised.- [COMMAND_EXECUTION]: The skill includes instructions and scripts that perform automated, high-impact actions such as kubectl delete pods -A and terraform apply -auto-approve. These commands modify or destroy infrastructure resources without required human-in-the-loop verification.- [EXTERNAL_DOWNLOADS]: The skill performs network requests to an internal registry service at 10.0.20.10 to retrieve container catalogs.- [PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection by ingesting untrusted data from the cluster.
Ingestion points: kubectl logs, kubectl describe, and kubectl get events as defined in SKILL.md.
Boundary markers: No markers or instructions are present to prevent the agent from following instructions embedded in cluster logs or metadata.
Capability inventory: The skill has access to bash, kubectl delete, ssh, and terraform apply.
Sanitization: No sanitization of ingested cluster data is performed.- [COMMAND_EXECUTION]: The skill provides instructions for accessing cluster nodes with administrative privileges via ssh root@<node-ip>, which presents a risk of unauthorized or accidental system-wide changes.

cluster-health