k8s-autoscaling-troubleshooting
Pass
Audited by Gen Agent Trust Hub on Nov 10, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [SAFE]: The skill follows least-privilege principles by focusing on read-only investigation and explicitly warning the agent against accessing sensitive information like Kubernetes Secrets or dumping objects with potentially inline credentials.
- [INDIRECT_PROMPT_INJECTION]: The skill interacts with external cluster data, creating a surface for indirect prompt injection from log or event content.
- Ingestion points: Usage of
kubectl logsandkubectl get eventsinSKILL.mdandreferences/metrics-keda-vpa.mdto analyze cluster state. - Boundary markers: Absent; the skill relies on interpreting specific status reasons and messages within the output.
- Capability inventory: Cluster modification tools include
kubectl patch,kubectl set resources, andkubectl rollout restartas described inSKILL.mdandreferences/metrics-keda-vpa.md. - Sanitization: The skill mitigates risks by using specific
jsonpathselectors to avoid processing entire object payloads and by explicitly blacklistingSecretsand the use of-o yamlon sensitive objects. - [COMMAND_EXECUTION]: The skill generates shell commands for the agent to execute against a Kubernetes cluster.
- Evidence: Investigation commands are read-only. All modification commands (e.g.,
kubectl patch) are explicitly categorized as requiring user approval or manual execution, ensuring human-in-the-loop oversight.
Audit Metadata