k8s-capi

Pass

Audited by Gen Agent Trust Hub on Nov 9, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDATA_EXFILTRATION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill processes data from Kubernetes API outputs, such as cluster resource names, machine statuses, and infrastructure events, which could be manipulated by a malicious actor in the managed environment to influence the agent's behavior. \n
  • Ingestion points: SKILL.md (via tools capi_list_clusters, capi_get_cluster, capi_list_machines, and kubectl_get)\n
  • Boundary markers: Absent; there are no instructions provided to the agent to distinguish between its own logic and potentially malicious instructions embedded in tool outputs.\n
  • Capability inventory: SKILL.md (via tools kubectl_apply and capi_scale_machinedeployment which allow cluster modification)\n
  • Sanitization: Absent; the skill does not specify any filtering, validation, or escaping of the cluster data before it is ingested into the prompt context.\n- [DATA_EXFILTRATION]: The skill includes the capi_get_cluster_kubeconfig tool, which retrieves full administrative credentials for workload clusters. Although this is consistent with its role in lifecycle management, it exposes sensitive access tokens and certificates within the agent's flow.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 9, 2026, 06:52 PM