The Agent Skills Directory

[COMMAND_EXECUTION]: The take_screenshot.py script executes various system binaries, including screencapture on macOS and scrot or import on Linux, using the subprocess module to perform screen captures.
[PRIVILEGE_ESCALATION]: The skill documentation recommends running the Windows helper script take_screenshot.ps1 with the -ExecutionPolicy Bypass flag, which circumvents system-level PowerShell script execution restrictions.
[DYNAMIC_EXECUTION]: The skill utilizes dynamic execution for OS integration. take_screenshot.py executes Swift scripts (macos_permissions.swift, macos_window_info.swift) via the swift interpreter. Additionally, take_screenshot.ps1 uses the Add-Type cmdlet to compile and load C# code at runtime to access native Windows API methods for window management and screen capture.
[INDIRECT_PROMPT_INJECTION]: As the skill captures and displays visual system information to the agent, it creates a surface for indirect prompt injection. Malicious instructions displayed on a user's screen (e.g., in a browser or text editor) could be interpreted by the AI as valid commands. \n * Ingestion points: Screen captures generated by take_screenshot.py and take_screenshot.ps1 are ingested into the agent's context. \n * Boundary markers: There are no explicit markers or warnings instructing the AI to ignore content found within the captured images. \n * Capability inventory: The skill has access to shell execution and file system writes via its capture scripts. \n * Sanitization: The Python script performs minimal sanitization, such as rudimentary quote escaping for application names, before passing them to system commands.

screenshot