win-mouse-native

Warn

Audited by Gen Agent Trust Hub on Mar 29, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The wrapper script win-mouse.cmd executes PowerShell with the -ExecutionPolicy Bypass parameter. This allows the skill to circumvent Windows security restrictions that would otherwise prevent the execution of unsigned or local scripts.\n- [COMMAND_EXECUTION]: The script scripts/win-mouse.ps1 utilizes the Add-Type command to dynamically compile C# code into the running session. This provides the agent with direct access to native Windows API functions in user32.dll (SetCursorPos, SendInput), which are used to simulate mouse movements and clicks.\n- [PROMPT_INJECTION]: The skill establishes an indirect prompt injection surface by ingesting user-provided coordinates and actions into high-privilege system calls.\n
  • Ingestion points: User-supplied values passed as command-line arguments to win-mouse.cmd are directly utilized by the PowerShell script via the $args array.\n
  • Boundary markers: None. The skill does not implement delimiters or instructions to ignore embedded commands within the input data.\n
  • Capability inventory: The skill possesses the capability to compile and execute native code (Add-Type) and perform hardware-level interactions (SetCursorPos, SendInput).\n
  • Sanitization: The script performs basic type casting to integers (e.g., [int]$A) for coordinates, which mitigates some forms of injection but does not restrict the potential for the agent to be manipulated into performing unintended UI actions.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Mar 29, 2026, 09:21 PM
Security Audit — agent-trust-hub — win-mouse-native