virtual-remote-desktop

KasmVNC-based virtual desktop for headless Linux with AI-first automation and human handoff. Use when most steps are automated but a user must manually inter...

What to consider before installing: - The code will attempt to download and install a KasmVNC .deb from GitHub and uses sudo/apt — only run on machines you control or in an isolated VM. - The installer uses an embedded Python snippet that imports 'requests' but the script does not ensure the module exists; the install may fail unexpectedly. - The install may add your user to the 'ssl-cert' group so the service can use TLS keys — this grants access to system TLS private key files and is a privilege change. - The skill reads/writes browser profile and cookie files (CHROME_PROFILE_DIR/Default/Cookies) and stores runtime secrets under ~/.openclaw/vrd-data; if you keep real login profiles there you may expose session cookies to the remote desktop/human handoff. - Metadata understates actual requirements (binaries/env vars). Expect to need vncserver/vncpasswd, python3, wget, curl, ss/sg, xdotool, scrot, sudo. Recommendations: review scripts line-by-line, run initially in a disposable VM/container, verify the downloaded .deb URL before installing, do not point CHROME_PROFILE_DIR at sensitive live profiles, and ensure you trust the source. If you need help hardening the flow (e.g., avoid adding ssl-cert group, require manual review of downloaded asset), ask and I can suggest changes.

Type: OpenClaw Skill Name: virtual-remote-desktop Version: 1.0.3 The skill provides a KasmVNC-based virtual desktop for AI automation, which inherently involves high-privilege operations like running a browser, interacting with a GUI (`xdotool`), and network exposure. While the `SKILL.md` and scripts do not show explicit malicious intent (e.g., data exfiltration, backdoors), the `start_vrd.sh` script allows binding the VNC server to `0.0.0.0` (`KASM_BIND=0.0.0.0`) for 'temporary public takeover', which is a significant security risk if not properly secured or if the `AUTO_STOP_IDLE_SECS` is bypassed. Additionally, the script's `chrome_sandbox_flags` function may launch Chrome with `--no-sandbox` under certain conditions, increasing the attack surface for browser exploits. These capabilities, while documented as part of the skill's functionality, represent substantial vulnerabilities that could be exploited by a compromised agent or environment, thus classifying it as suspicious rather than benign.