computer-use

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (high risk: 0.80). The skill's SKILL.md workflow and "Opening Applications"/VNC sections show the agent taking screenshots and "Analyze — Identify UI elements" (and installing/launching Google Chrome and exposing the desktop via noVNC), which means the agent is expected to read and act on arbitrary web or other third‑party content rendered in the desktop/VNC session (untrusted public pages), enabling indirect prompt injection.

MEDIUM W013: Attempt to modify system services in skill instructions.

• Attempt to modify system services in skill instructions detected (high risk: 0.90). The skill explicitly instructs installing system services, running sudo commands, copying files into /etc/systemd/system, and modifying/masking /usr/bin/xfdesktop (and installing packages via dpkg/apt), all of which change system state and require elevated privileges.