The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill facilitates the analysis of untrusted data from Azure resource logs and Kubernetes events, creating a surface for indirect prompt injection. Ingestion points: Operations in SKILL.md and troubleshooting/aks/pod-failures.md read output from tools such as az monitor and kubectl logs. Boundary markers: None are specified for the processed data. Capability inventory: Diagnostic tools including kubectl run, kubectl debug, and az rest are present in reference files. Sanitization: No sanitization of external content is specified. This risk is inherent to the troubleshooting use case and is managed by the skill's focus on read-only diagnostics and clear human-in-the-loop warnings.
[COMMAND_EXECUTION]: The skill provides instructions for running diagnostic commands such as kubectl debug node and az rest to inspect Azure resources. These commands are necessary for the skill's primary function and are documented with clear warnings regarding their impact. No evidence of unauthorized or malicious execution was found.

azure-diagnostics