computer-use

Warn

Audited by Gen Agent Trust Hub on Mar 29, 2026

Risk Level: MEDIUMCOMMAND_EXECUTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill makes extensive use of osascript, cliclick, and screencapture to perform actions on the host system. It generates and executes AppleScript and Bash commands dynamically to interact with third-party applications like WeChat and Feishu.
  • [DYNAMIC_EXECUTION]: The init.sh script dynamically compiles a Swift source file (scroll.swift) into an executable binary (/tmp/_cu_scroll) at runtime to handle mouse scroll events using CGEvent APIs.
  • [PRIVILEGE_ESCALATION]: The skill requires the user to manually grant 'Accessibility' and 'Screen Recording' permissions. These permissions grant the agent full control over the desktop environment, allowing it to capture sensitive information (passwords, private chats) or perform unauthorized actions if the model is misdirected.
  • [DATA_EXPOSURE]: The skill's primary function involves taking screenshots of the entire desktop and processing them via the AI model. This creates a high risk of accidental exposure of private information visible on the screen during the automation process.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Mar 29, 2026, 02:33 AM