k8s-node-troubleshooting
Pass
Audited by Gen Agent Trust Hub on Nov 8, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill retrieves and processes data from the Kubernetes API, including node conditions, event logs, and pod metadata. These are untrusted external inputs that could be manipulated by an attacker with cluster access to influence the agent's behavior.
- Ingestion points: Data enters the agent context through commands like
kubectl get,kubectl describe, andkubectl eventsacross several files including SKILL.md and reference files in the references/ directory. - Boundary markers: While the skill lacks explicit technical delimiters for data interpolation, it instructs the agent to 'read first' and obtain human approval before acting on findings.
- Capability inventory: The skill can execute state-altering commands such as
kubectl uncordon,kubectl taint, andkubectl scaleas documented in SKILL.md, references/taints.md, and references/capacity.md. - Sanitization: The instructions include a strict policy to 'Never read Secrets' and to avoid repeating any sensitive information like passwords or tokens encountered in command outputs.
- [COMMAND_EXECUTION]: The skill makes extensive use of the
kubectlCLI to manage and investigate cluster resources. This includes operations that change the operational state of the cluster, such as uncordoning nodes or scaling deployments. The risk associated with these commands is mitigated by a mandatory human-in-the-loop requirement, where the agent is instructed to ask for permission before execution.
Audit Metadata