The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/agent.py and documentation in references/api-reference.md use subprocess.run to execute system commands such as lsusb, usbguard, and powershell. These commands are used to audit the status of security policies and connected hardware. The commands are constructed using static strings or predefined list arguments, which prevents command injection from user-supplied input.\n- [PROMPT_INJECTION]: The utility scripts/process.py parses CSV files containing endpoint logs to generate USB activity reports. This ingestion of external data represents an indirect prompt injection surface where malicious content embedded in the log fields (e.g., a device name containing instructions) could potentially influence the agent's reasoning when interpreting the report.\n
Ingestion points: Data is read from CSV files in scripts/process.py via csv.DictReader.\n
Boundary markers: No explicit delimiters or instructions to ignore embedded commands are included in the processing logic.\n
Capability inventory: The skill contains scripts with the ability to perform file system writes and execute system commands.\n
Sanitization: The script performs no sanitization or escaping of the ingested CSV field values before including them in the final JSON report.

implementing-usb-device-control-policy