The Agent Skills Directory

[COMMAND_EXECUTION]: The Python and Bash scripts (scripts/take_screenshot.py, scripts/ensure_macos_permissions.sh) execute system-level commands such as screencapture, scrot, gnome-screenshot, and osascript to perform screen captures and permission management.
[COMMAND_EXECUTION]: The PowerShell script (scripts/take_screenshot.ps1) utilizes Add-Type to dynamically compile and execute C# code for interfacing with the Windows API (GetForegroundWindow, GetWindowRect), which is a form of dynamic execution for system interaction.
[PROMPT_INJECTION]: The skill exposes the agent to indirect prompt injection (Category 8) by facilitating the ingestion of untrusted visual data from the operating system environment.
Ingestion points: Screenshots captured and saved to the local filesystem via the included scripts.
Boundary markers: None present; the agent lacks the ability to distinguish between legitimate UI elements and malicious text instructions embedded within captured images.
Capability inventory: The agent can execute arbitrary shell commands and perform filesystem operations via the skill's utility scripts.
Sanitization: No sanitization, OCR filtering, or content validation is applied to the visual data before it is processed by the agent.

screenshot