k8s-loki-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
kubectlto inspect the cluster state, retrieve logs, and query Loki's internal API via the service proxy. This includes commands likekubectl logs,kubectl get, andkubectl get --raw. These are necessary for the skill's diagnostic purpose. - [INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it processes external, untrusted data from the Kubernetes environment.
- Ingestion points: The skill ingests untrusted data from application logs (
kubectl logs) and Loki configuration files (kubectl get configmap) as seen in SKILL.md and references/ingestion.md. - Boundary markers: The instructions do not define strict delimiters for the ingested data, although they use placeholders like
my-namespacefor scope targeting. - Capability inventory: The skill has the capability to propose and execute state-changing operations like
kubectl applyandkubectl rollout restart(SKILL.md). - Sanitization: The skill includes defensive instructions, such as advising the agent to "read only the parts you need," never repeat seen keys/tokens, and prefer masked API endpoints like
/config?mode=diffto protect sensitive information.
Audit Metadata