The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests and processes untrusted data from the Kubernetes cluster, including event messages, application logs, and GitOps status fields. This data could contain malicious instructions intended to mislead the agent or manipulate its reporting.
Ingestion points: references/kubernetes-core.md (cluster events via kubectl get events), references/monitoring-logs.md (application logs via kubectl logs), and references/helm-gitops.md (GitOps controller status messages).
Boundary markers: The skill instructions emphasize context confirmation and output capping but do not specify clear delimiters to separate ingested data from agent instructions.
Capability inventory: The skill has access to kubectl, helm, bash, jq, openssl, dig, and ssh to perform diagnostics.
Sanitization: The skill employs head, tail, and jq filters to limit the volume of data returned to the agent context, reducing the likelihood of processing large-scale malicious payloads.
[COMMAND_EXECUTION]: The skill relies on local shell execution of diagnostic tools including kubectl, helm, openssl, dig, and jq. While the instructions explicitly forbid mutation commands (e.g., apply, delete, patch), the execution capability remains a surface that could be targeted if the agent is compromised via prompt injection.

kubernetes-health