k8s-storage-troubleshooting
Pass
Audited by Gen Agent Trust Hub on Nov 8, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONCREDENTIALS_UNSAFE
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from the Kubernetes environment (logs, events, resource descriptions) which could contain malicious instructions designed to influence the agent's behavior.
- Ingestion points: The skill reads from kubectl logs, kubectl events, and kubectl describe across all markdown files.
- Boundary markers: It lacks explicit technical delimiters for tool output but provides high-level instructions to the agent to 'Read first' and 'ask before changing'.
- Capability inventory: The skill can perform modifications via kubectl scale, kubectl patch, and kubectl apply.
- Sanitization: Includes instructions to avoid repeating sensitive data like passwords or tokens found in output.
- [COMMAND_EXECUTION]: The skill uses kubectl to perform administrative tasks and interact with running workloads.
- Usage: Includes kubectl exec to run commands within user pods and kubectl scale to modify deployment replicas.
- Safety Gates: High-impact commands are properly sequestered under 'Ask first' or 'Proposed fix' sections, explicitly preventing the agent from running them without manual user intervention.
- [CREDENTIALS_UNSAFE]: The skill demonstrates proactive security awareness regarding Kubernetes secrets.
- Practice: Explicitly warns the agent against reading Secret resources or repeating sensitive values (passwords, tokens, keys) in its responses.
Audit Metadata