The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external, untrusted data from Azure Activity Logs and monitoring alerts, creating a potential surface for indirect prompt injection.
Ingestion points: Azure Activity Logs, Resource Health events, and monitoring alert text are processed during the 'Live Fact Collection' and 'Service Branch' phases.
Boundary markers: The skill instructs the agent to separate 'facts' from 'interpretations' and mandates verification of service-specific claims using official Microsoft Learn URLs, providing logical boundaries for processed data.
Capability inventory: The skill utilizes Azure CLI (az) and PowerShell (Invoke-RestMethod) for data retrieval and writes reports to the local Troubleshooting/ directory.
Sanitization: While no explicit string escaping is defined, the workflow requires grounding all interpretations in official documentation to mitigate data poisoning.
[COMMAND_EXECUTION]: The instructions guide the agent to use Azure CLI and PowerShell for gathering diagnostic information.
Evidence: Multiple Azure CLI commands are referenced, including az vm get-instance-view, az aks show, and az resource show for resource inspection.
Evidence: references/service-branches.md includes guidance for using PowerShell's Invoke-RestMethod to query Log Analytics APIs.
[EXTERNAL_DOWNLOADS]: The skill references official Microsoft Learn documentation for behavior verification.
Evidence: The 'Workflow' and 'Official Grounding' sections require the use of Microsoft Learn URLs to verify observations, which are recognized as trusted official sources.

azure-troubleshooting