k8s-velero-troubleshooting
Pass
Audited by Gen Agent Trust Hub on Nov 8, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill represents a surface for indirect prompt injection as it ingests untrusted data from the Kubernetes API and container logs (e.g.,
kubectl get,kubectl logs). - Ingestion points: Data enters the agent's context through the output of
kubectl get(object statuses, failure reasons, messages) andkubectl logsfrom the Velero server and node-agent pods (SKILL.md, steps 1, 2, 4, 5). - Boundary markers: The skill does not use specific delimiters or 'ignore embedded instructions' warnings for the ingested data, but it restricts investigation to specific fields using jsonpath.
- Capability inventory: The skill possesses write capabilities through
kubectl patch,kubectl annotate, andkubectl apply(SKILL.md, 'Proposed fix' section; references/volume-data.md, 'Proposed fix' section). - Sanitization: No explicit sanitization or filtering of external content is defined, relying on the agent's interpretation of the structured Kubernetes output.
- [COMMAND_EXECUTION]: The skill uses
kubectlcommands for investigation and remediation. The remediation commands are clearly marked with 'Proposed fix' or 'Ask first' headers to ensure human oversight before execution.
Audit Metadata