The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes various system and cluster management commands to perform diagnostics on the environment.
Evidence: Uses the openshell CLI, kubectl, helm, docker, podman, and mise for status checks and log retrieval.
[DATA_EXPOSURE]: The skill accesses diagnostic data, including container logs, Kubernetes secret names, and local configuration files, which is necessary to identify deployment failures.
Evidence:
Inspects Kubernetes secrets: kubectl -n openshell get secret openshell-server-tls ...
Retrieves Helm deployment values: helm -n openshell get values openshell
References local configuration paths: ~/.config/openshell/gateways/<name>/mtls/
[EXTERNAL_DOWNLOADS]: The skill references container images from the vendor's official registry.
Evidence: References ghcr.io/nvidia/openshell/supervisor:latest for sandbox diagnostic checks.
[INDIRECT_PROMPT_INJECTION]: The skill processes data from logs and cluster events which serves as a surface for indirect prompt injection if those sources contain malicious instructions.
Ingestion points: Reads output from docker logs, kubectl logs, and kubectl get events into the agent context.
Boundary markers: Absent. The instructions do not explicitly tell the agent to ignore embedded instructions within log outputs.
Capability inventory: High. The skill has the ability to execute cluster and container management commands (kubectl, docker, helm).
Sanitization: Absent. Raw tool output is passed directly to the agent's context for interpretation.

debug-openshell-cluster