The Agent Skills Directory

[COMMAND_EXECUTION]: The skill extensively uses PowerShell to execute Azure CLI (az) commands for scanning existing resources, verifying login status, and performing deployments. It also executes bundled Python scripts for diagram generation and Node.js for image conversion.
[DYNAMIC_EXECUTION]: In scripts/cli.py, the skill dynamically constructs a JavaScript code string to be executed via node -e for converting HTML diagrams to PNG images. This code utilizes puppeteer to capture screenshots. While this is a dynamic execution pattern, it is localized to a specific utility function and uses hardcoded templates for the script logic.
[INDIRECT_PROMPT_INJECTION]: The skill exhibits a surface for indirect prompt injection as it ingests untrusted data from existing Azure resource metadata and external Microsoft documentation which is then processed by the LLM.
Ingestion points: Azure resource descriptions from az CLI results (phase0-scanner.md) and technical specifications fetched from Microsoft Docs via web_fetch (bicep-generator.md).
Boundary markers: None. The data is interpolated directly into instructions for the agent to generate Bicep templates and diagrams.
Capability inventory: The skill has access to powershell for shell command execution, file system writes for project outputs, and subprocess execution (python, node, az).
Sanitization: There is no explicit sanitization or filtering of the ingested resource metadata before it is presented to the LLM or used in code generation.

azure-architecture-autopilot