desktop-control
Pass
Audited by Gen Agent Trust Hub on Mar 1, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The script
scripts/screen.pyexecutesosascriptviasubprocess.runto retrieve frontmost application metadata, window titles, and the accessibility UI tree.\n- [COMMAND_EXECUTION]: The scriptscripts/keyboard.pyexecutes the systempbcopycommand to handle non-ASCII text input via the clipboard.\n- [COMMAND_EXECUTION]: All bundled scripts (mouse.py,keyboard.py,screen.py) contain logic to executepip installfor thepyautoguilibrary if it is not present on the host system.\n- [EXTERNAL_DOWNLOADS]: The skill automatically fetches thepyautoguipackage from the Python Package Index (PyPI) at runtime if the dependency is missing.\n- [DATA_EXFILTRATION]: Thescripts/screen.pyutility allows the agent to take screenshots (captureaction) and scrape the accessibility tree (read-uiaction), exposing potentially sensitive data visible on the user's screen.\n- [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it ingests untrusted UI data that could contain malicious instructions for the agent.\n - Ingestion points:
scripts/screen.pyreads active window titles, UI labels, and textual content from the accessibility tree.\n - Boundary markers: No delimiters or instructions are used to separate screen-scraped data from the agent's internal logic.\n
- Capability inventory: The skill can simulate arbitrary mouse and keyboard events, write files (screenshots), and execute system commands via
subprocess.\n - Sanitization: Content provided to the typing and clicking actions is used directly without sanitization or escaping of potentially malicious character sequences.
Audit Metadata