desktop-gui-inspect

Pass

Audited by Gen Agent Trust Hub on Jul 30, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTIONDATA_EXFILTRATION
Full Analysis
  • [COMMAND_EXECUTION]: The skill invokes local bash scripts such as screenshot.sh, click.sh, and type_text.sh to control a virtual X11 desktop environment (DISPLAY=:99). These scripts are located in the user's home directory under ~/.openclaw/skills/computer-use/scripts/.
  • [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted visual data from third-party desktop applications. Ingestion points: Desktop screenshots captured via ./scripts/screenshot.sh. Boundary markers: The 'Golden Rule' provides instructional constraints against committing changes, but there is no technical enforcement. Capability inventory: Broad input simulation capabilities including clicking, typing, and dragging. Sanitization: No technical sanitization or validation of visual content is performed.
  • [DATA_EXFILTRATION]: The skill captures screenshots of the desktop and returns them as base64-encoded PNG strings. This facilitates the movement of visual data from the virtual environment to the agent's output.
  • [PRIVILEGE_ESCALATION]: Skill documentation instructs the user to manually apply chmod +x to script files if the automatic installer fails to set executable permissions, which involves manual modification of file system permissions to enable script execution.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 30, 2026, 04:29 PM
Security Audit — agent-trust-hub — desktop-gui-inspect