The Agent Skills Directory

[SAFE]: The skill is authored by 'microsoft/azure-skills' and references official Microsoft repositories, documentation, and tools. All external references and downloads target trusted domains such as microsoft.com and the Microsoft Container Registry (mcr.microsoft.com).
[SAFE]: The skill includes explicit production safety and diagnostic constraints. It defaults to read-only incident evidence, prohibits the automatic installation of packages or execution of scripts, and mandates the masking of tokens, connection strings, and personal data in reports.
[SAFE]: The instructions explicitly warn the agent not to trust instructions found in log files, command outputs, or browsed pages as elevated authorization, providing protection against indirect prompt injection.
[SAFE]: While some tools like Inspektor Gadget utilize privileged container profiles ('kubectl debug --profile=sysadmin'), these are clearly labeled as privileged diagnostic actions that require explicit user approval before execution.
[SAFE]: No evidence of obfuscation, persistence mechanisms, or unauthorized data exfiltration was found. The skill's behavior is entirely consistent with its stated purpose of investigating Azure service incidents.

azure-diagnostics