The Agent Skills Directory

[COMMAND_EXECUTION]: The win-mouse.cmd script executes PowerShell with the -ExecutionPolicy Bypass flag, which ignores local script execution restrictions to run the underlying automation logic.
[COMMAND_EXECUTION]: The scripts/win-mouse.ps1 script utilizes the Add-Type cmdlet to compile C# source code at runtime. This dynamic compilation is required to access the user32.dll functions SetCursorPos and SendInput, which enable the skill's mouse control capabilities.
[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface as it is designed to execute mouse actions based on potentially untrusted coordinate data provided by the user or extracted from the environment.
Ingestion points: Untrusted coordinates are ingested via user instructions as described in SKILL.md.
Boundary markers: The skill includes explicit security guardrails in SKILL.md and SAFETY.md that instruct the agent to use checkpoints for long sequences and verify authorization for remote desktop or VM sessions.
Capability inventory: The script scripts/win-mouse.ps1 provides full mouse interaction capabilities (moving, clicking, dragging) via direct Win32 API calls.
Sanitization: While input arguments are cast to integers, there is no validation to ensure coordinates are within safe bounds or targeting intended UI elements.

win-mouse-native-hardened