The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides numerous shell commands for system administration and cluster debugging, including kubectl, docker, podman, helm, and openshell CLI. These tools are used to inspect the health and configuration of the cluster environment.
[INDIRECT_PROMPT_INJECTION]: The skill facilitates the ingestion and analysis of untrusted data from the local environment, such as cluster logs, container output, and configuration files. This creates a surface where malicious instructions embedded in logs or files (log poisoning) could attempt to influence the agent's actions.
Ingestion points: Gateway process logs, container logs (docker logs, podman logs), Kubernetes events (kubectl get events), and local configuration files (/etc/openshell/gateway.toml).
Boundary markers: The instructions do not explicitly define delimiters or include specific warnings for the agent to ignore embedded instructions within the logs being processed.
Capability inventory: The skill employs powerful cluster management utilities including kubectl, docker, podman, helm, and the openshell CLI, which are necessary for the debugging use case.
Sanitization: No explicit sanitization or filtering of the ingested log or configuration content is mentioned in the diagnostic workflow.
[EXTERNAL_DOWNLOADS]: The skill references official vendor documentation (docs.nvidia.com) and container images hosted on a public container registry (ghcr.io/nvidia/). These resources are from the skill's authoring organization.
[PRIVILEGE_ESCALATION]: The workflow describes operations that require elevated privileges, such as using sudo to manage systemd services or regenerate CDI specifications. This behavior is consistent with the intended role of debugging system-level deployments.

debug-openshell-cluster