The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/agent.py uses the subprocess.run function to execute Azure CLI (az) commands. The arguments are passed as a list rather than a shell string, and the subcommands are predefined, which effectively prevents command injection vulnerabilities.
[INDIRECT_PROMPT_INJECTION]: The skill processes data from external sources that could theoretically contain malicious instructions designed to influence an LLM.
Ingestion points: Data is ingested from Azure security alerts and assessments via the az security alert list and az security assessment list commands in scripts/agent.py.
Boundary markers: The script does not utilize specific delimiters or instructions to isolate the fetched security metadata from the agent's control context.
Capability inventory: The script includes the capability to write security reports to the local file system using the generate_posture_report function.
Sanitization: The script performs no sanitization on text fields retrieved from Azure, such as alert names or assessment descriptions, before including them in the console output or JSON report.

securing-azure-with-microsoft-defender