computer-control
Pass
Audited by Gen Agent Trust Hub on Sep 17, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill captures screenshots of the desktop environment to guide its actions. This creates a surface for indirect prompt injection if the agent encounters malicious text or UI elements on a webpage or within an application. 1. Ingestion points: The screenshot action captures the full display for model processing in SKILL.md. 2. Boundary markers: There are no explicit instructions or delimiters mentioned to help the model distinguish between task instructions and content found on the screen. 3. Capability inventory: The skill can simulate mouse clicks, typing, and key presses via xdotool and scrot across multiple files. 4. Sanitization: No sanitization of visual content is performed before it is sent to the model.
- [COMMAND_EXECUTION]: The skill uses external system utilities (xdotool, scrot, xclip) to perform desktop actions and capture screenshots, as described in the architecture section of SKILL.md.
Audit Metadata