The Agent Skills Directory

[COMMAND_EXECUTION]: The skill facilitates extensive system and cluster diagnostics through standard CLI tools.
Executes openshell, docker, podman, kubectl, and helm to inspect gateway status and compute driver health.
Uses systemctl with sudo to manage NVIDIA CDI refresh services, which is consistent with administrative troubleshooting tasks.
Employs common utilities like rg, stat, journalctl, and openssl for inspecting configuration files, sockets, and certificates.
[EXTERNAL_DOWNLOADS]: The skill references official vendor-owned container images.
Instructs the agent to run ghcr.io/nvidia/openshell/sandbox:latest for version verification.
The container registry path is verified as a legitimate resource belonging to the skill's author (nvidia).
[INDIRECT_PROMPT_INJECTION]: The skill accesses external system data that could theoretically be manipulated to influence agent behavior.
Ingestion points: Processes data from container logs, system journals, and gateway configuration files (gateway.toml, ConfigMaps, Secrets).
Boundary markers: None explicitly implemented; the skill assumes a trusted administrative environment.
Capability inventory: The agent is granted access to high-privilege orchestration tools and local system diagnostics.
Sanitization: Content is read directly for troubleshooting purposes; no specific sanitization of log output is performed before analysis.

debug-openshell-cluster