k8s-istio-troubleshooting

Pass

Audited by Gen Agent Trust Hub on Nov 8, 2026

Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill uses the kubectl CLI to perform system inspection and administrative tasks.
  • Commands include kubectl get, kubectl describe, kubectl logs, and kubectl rollout restart.
  • Potentially disruptive operations, such as restarting workloads, are gated by explicit instructions requiring the agent to obtain user consent before execution.
  • [COMMAND_EXECUTION]: The skill utilizes kubectl exec to interact with the Envoy proxy's administrative API (pilot-agent request GET) inside running containers.
  • These operations are restricted to specific diagnostic endpoints (clusters, stats) and are limited to HTTP GET requests.
  • Execution is placed under an "Ask first" directive, ensuring the agent requests user permission before the command is run.
  • [INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it ingests and interprets untrusted data from the Kubernetes cluster environment.
  • Ingestion points: The agent reads container logs (kubectl logs), Kubernetes resource definitions in YAML format (kubectl get -o yaml), and output from proxy admin API requests.
  • Boundary markers: Absent. The instructions do not specify the use of delimiters or boundary markers to separate untrusted data from the agent's instructions.
  • Capability inventory: The skill can read cluster state, retrieve logs, execute commands inside pods (with user consent), and trigger workload rollouts (with user consent).
  • Sanitization: Absent. There are no instructions to sanitize or validate the content of the logs or resource metadata before the agent processes them.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 8, 2026, 10:58 PM