windows-diagnostics
Warn
Audited by Gen Agent Trust Hub on Aug 6, 2026
Risk Level: MEDIUMCOMMAND_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill instructions direct the agent to run PowerShell with the
-ExecutionPolicy Bypassflag, which intentionally circumvents system security configurations designed to restrict the execution of unauthorized scripts. - [DATA_EXFILTRATION]: The collection script gathers extensive system and user metadata, including file names and sizes from private user directories such as Desktop, Documents, Downloads, and Videos, which may contain sensitive information.
- [DATA_EXFILTRATION]: Comprehensive system state information, including active network connections, process owners, local user accounts, and startup registry entries, is written to a world-readable JSON file in the system's temporary directory (
%TEMP%\windows_diagnostics.json). - [PROMPT_INJECTION]: The skill processes untrusted system data (such as event log messages, registry values, and package names) and presents them to the agent to drive follow-up actions like software removal. This creates a surface for indirect prompt injection where malicious system metadata could influence agent behavior.
- [DATA_EXFILTRATION]: The script enumerates all installed software and AppX packages, including publisher information and install dates, providing a detailed profile of the user's software environment.
- [COMMAND_EXECUTION]: The skill implements background jobs to perform wide-ranging system queries via CIM and WMI, accessing low-level hardware and operating system details that are typically restricted for security reasons.
Audit Metadata