The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides instructions and examples for simulating user input (mouse clicks, keyboard typing) and managing application windows using pyautogui, pywinauto, and system-level AppleScript (via subprocess on macOS). These capabilities are necessary for its primary purpose of OS automation.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted external data by capturing screenshots for visual recognition and OCR, reading from the system clipboard, and importing data from local files (e.g., Excel spreadsheets). This data ingestion surface could potentially be exploited if external content contains malicious instructions intended for the agent, though no active exploitation was found.
[SAFE]: All identified dependencies are well-known, standard libraries for automation and image processing. The credential examples provided in the documentation (e.g., 'TestPassword123') are clearly marked as placeholders for demonstration and testing purposes.

os-use