Virtual Remote Desktop (KasmVNC edition)

What this skill is for

Use this when the workflow is:

• AI runs browser automation most of the time
• Captcha / risk-control / MFA appears
• User takes over remotely for 1-3 minutes
• AI continues automatically

Core commands

0) Install KasmVNC (one-time)

bash /home/ubuntu/.openclaw/workspace/skills/virtual-remote-desktop/scripts/install_kasmvnc.sh

This installer also prepares required runtime tools:

• fluxbox (lightweight desktop)
• xdotool + scrot (computer-use actions)
• xauth

1) Requirement-driven start (recommended)

Before starting, always confirm these user requirements:

• takeover device: 手机 or 电脑
• website render mode: 手机网页 or 桌面网页
• access mode: 本地隧道 (127.0.0.1) or 临时公网 (0.0.0.0)
• network quality: 弱网 / 普通 / 良好

bash /home/ubuntu/.openclaw/workspace/skills/virtual-remote-desktop/scripts/start_vrd_guided.sh

Preview config without starting:

bash /home/ubuntu/.openclaw/workspace/skills/virtual-remote-desktop/scripts/start_vrd_guided.sh --dry-run

1.1) Direct start (manual env)

bash /home/ubuntu/.openclaw/workspace/skills/virtual-remote-desktop/scripts/start_vrd.sh

Important env vars:

• AUTO_LAUNCH_URL (optional): open target page automatically
• KASM_BIND (default 127.0.0.1, safer)
• AUTO_STOP_IDLE_SECS (default 900)
• BROWSER_MOBILE_MODE=1 (launch browser with mobile emulation)
• BROWSER_DEVICE=iphone14pro|pixel7|ipad

AUTO_LAUNCH_URL="https://example.com/login" \
AUTO_STOP_IDLE_SECS=1200 \
bash /home/ubuntu/.openclaw/workspace/skills/virtual-remote-desktop/scripts/start_vrd.sh

Mobile-friendly VNC stream example (better phone takeover UX):

MOBILE_MODE=1 MOBILE_PRESET=phone \
AUTO_STOP_IDLE_SECS=900 \
bash /home/ubuntu/.openclaw/workspace/skills/virtual-remote-desktop/scripts/start_vrd.sh

Mobile stream options:

• MOBILE_MODE=1 enables mobile defaults
• MOBILE_PRESET=phone|tablet sets default resolution (960x540 / 1280x720)
• KASM_MAX_FPS can be lowered further (e.g. 18) on weak networks

AUTO_LAUNCH_URL="https://example.com" \
BROWSER_MOBILE_MODE=1 BROWSER_DEVICE=iphone14pro \
bash /home/ubuntu/.openclaw/workspace/skills/virtual-remote-desktop/scripts/start_vrd.sh

Notes:

• Browser mobile emulation changes UA + viewport + touch behavior.
• It is different from MOBILE_MODE (which optimizes VNC stream size).

2) Check status / health

bash /home/ubuntu/.openclaw/workspace/skills/virtual-remote-desktop/scripts/status_vrd.sh
bash /home/ubuntu/.openclaw/workspace/skills/virtual-remote-desktop/scripts/health_vrd.sh

3) Stop desktop

bash /home/ubuntu/.openclaw/workspace/skills/virtual-remote-desktop/scripts/stop_vrd.sh

AI automation actions (computer-use style)

All actions run on the active virtual display from pids.env.

# screenshot (base64)
bash scripts/action_screenshot.sh

# click / type / key / scroll
bash scripts/action_click.sh 500 420 left
bash scripts/action_type.sh "hello"
bash scripts/action_key.sh "ctrl+l"
bash scripts/action_scroll.sh down 4

# helpers
bash scripts/action_mouse_move.sh 800 300
bash scripts/action_cursor_position.sh
bash scripts/action_wait.sh 2

Recommended loop:

• action_screenshot.sh
• analyze
• action_click/type/key/...
• screenshot verify
• repeat

Best handoff pattern (AI ↔ User)

When captcha/risk-control appears:

• AI sends user the KasmVNC URL + username/password
• User manually solves challenge
• User replies “done”
• AI runs screenshot + validation step
• AI resumes automation

UX presets

Preset A: safe local tunnel (recommended)

• KASM_BIND=127.0.0.1
• user connects via SSH tunnel
• best for security

Preset B: temporary public takeover

• KASM_BIND=0.0.0.0
• short AUTO_STOP_IDLE_SECS (e.g. 300)
• use only for urgent remote intervention

Notes

• KasmVNC uses HTTPS + per-user auth (username/password).
• This skill stores runtime files in ~/.openclaw/vrd-data by default.
• Browser profile persistence is controlled by CHROME_PROFILE_DIR.
• If captcha frequency is high, keep one long-lived profile to reduce repeated challenges.