The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes destructive commands such as helm uninstall and kubectl delete to remove operator releases, custom resources, and namespaces. These operations are the primary intended function of the skill and are secured by instructions requiring the agent to summarize impacts and obtain user approval before execution.
[REMOTE_CODE_EXECUTION]: Support for managing remote Kubernetes clusters via SSH is provided. The instructions emphasize transparency, requiring the agent to display all commands and wait for explicit approval before running them on the remote host.
[PROMPT_INJECTION]: The skill creates an indirect prompt injection surface by ingesting and processing output from Kubernetes and Helm commands which could contain untrusted data from the cluster environment.
Ingestion points: Output from kubectl get, kubectl describe, and helm status commands in SKILL.md and scripts/validate-nim-operator-uninstall.sh.
Boundary markers: No explicit delimiter markers are used for command output, although the skill mandates human-in-the-loop confirmation for any subsequent actions.
Capability inventory: High-privilege destructive capabilities including helm uninstall and kubectl delete (SKILL.md).
Sanitization: No evidence of automated data sanitization or validation before processing command outputs.

nim-operator-uninstall