Virtual Desktop Browser

Launch Chromium in non-headless mode inside Xvfb virtual display (fixed 1200x720x24) and automate with human-like mouse/keyboard/screenshot operations. Use f...

This skill is coherent with its stated purpose but has real operational impact: it requires installing OS packages (apt-get) and pip dependencies and will spawn Xvfb and a real Chromium process and then simulate mouse/keyboard input. Before installing/running, consider: 1) install and run inside an isolated environment (container or dedicated VM) so simulated input and browser sessions cannot interact with sensitive local apps; 2) be aware the agent can type/click arbitrarily in that virtual browser—do not pass secrets or session tokens into pages the skill will visit; 3) confirm you are comfortable running apt-get/pip on the host; 4) you may want to review or sandbox the skill code (skill.py) and the templates/images you supply for matching. If you need higher assurance, test in a disposable VM first.

Type: OpenClaw Skill Name: virtual-desktop-browser Version: 1.0.0 The skill provides powerful GUI automation capabilities by running Chromium in a virtual Xvfb display and controlling it via PyAutoGUI. While the implementation in `skill.py` aligns with the stated purpose of automating bot-resistant websites, it introduces high-risk behaviors such as simulated keyboard/mouse input, screen capture (Base64 exfiltration potential), and window management. Notably, it launches Chromium with the `--no-sandbox` flag, which weakens browser security. These capabilities are risky and could be abused for unauthorized actions if the agent is misdirected, though no explicit malicious intent or backdoors were found.