The Agent Skills Directory

Command Execution: The skill utilizes kubectl and az (Azure CLI) to retrieve status, events, and configuration from Kubernetes and Azure environments. These tools are used for read-only diagnostic purposes, following the skill's guidelines for host authorization and user consent for any mutations.
Indirect Prompt Injection: The skill processes outputs from external commands, such as kubectl get and az show. Ingesting data from external environments represents a potential surface for indirect instructions. The skill includes guidance to capture exact outputs and state which reads were performed, which helps mitigate accidental misinterpretation.
Data Exposure Considerations: Diagnostic commands like az account show retrieve subscription and tenant identifiers. These are handled as part of the environment 'binding' process to ensure troubleshooting occurs in the correct cloud context.
Trusted External References: The skill provides links to official documentation from Microsoft Learn and PyTorch to support its diagnostic logic. These references originate from well-known and established technical organizations.

aks-gpu-inference