The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses subprocess.run and PowerShell commands to execute the Azure CLI (az). This is used to obtain OAuth2 access tokens and fetch environment data from the Azure Resource Manager (ARM) APIs. These operations are limited to authenticated Azure management endpoints.
[DYNAMIC_EXECUTION]: The instructions in SKILL.md direct the agent to populate a specific DATA section within the scripts/generate_pptx.py file before execution. This is a controlled template-filling mechanism designed to inject the retrieved Azure data into the PowerPoint generation logic.
[INDIRECT_PROMPT_INJECTION]: The skill processes external data retrieved from Azure Advisor and Cost Management APIs, which could theoretically contain malicious strings if an attacker controls resource names or tags in the target subscription.
Ingestion points: Azure Advisor recommendation descriptions and resource metadata fetched via JSON APIs.
Boundary markers: The skill uses structured templates (Markdown and PPTX) that separate data from formatting instructions, reducing the risk of accidental command interpretation.
Capability inventory: Subprocess execution for Azure CLI, local file system writes for report generation, and network access to management.azure.com.
Sanitization: Data is processed using standard JSON parsers and the python-pptx library, which treats the inputs as static text elements for the report.

azure-advisor-report