screenshot

Pass

Audited by Gen Agent Trust Hub on Sep 9, 2026

Risk Level: SAFECOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill executes native OS commands such as screencapture, scrot, gnome-screenshot, and import through subprocess calls to perform screenshot captures. It also uses osascript on macOS to interact with running applications.
  • [DYNAMIC_EXECUTION]: The skill uses runtime compilation and execution for platform-specific tasks. The PowerShell script (take_screenshot.ps1) uses Add-Type to compile C# helper methods for Win32 API access, and the macOS scripts execute Swift code using the swift interpreter to manage permissions and window information.
  • [INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect injection through user-controlled parameters such as application names and file paths.
  • Ingestion points: User-provided strings for the --app, --window-name, and --path arguments in the Python and PowerShell scripts.
  • Boundary markers: Explicit boundary markers are not used for these inputs.
  • Capability inventory: The skill can execute system commands for screen capture and write files to the local file system.
  • Sanitization: The Python script includes logic to escape double quotes in application names before they are used in AppleScript commands, providing a basic level of protection.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 9, 2026, 03:52 AM
Security Audit — agent-trust-hub — screenshot