Stealthy Auto Browse

stealthy-auto-browse

Setup Required

This skill requires a running stealthy-auto-browse container. Set the STEALTHY_AUTO_BROWSE_URL environment variable to connect.

1. Run the container:

docker run -d -p 8080:8080 -p 5900:5900 psyb0t/stealthy-auto-browse

2. Configure OpenClaw (~/.openclaw/openclaw.json):

{
  "skills": {
    "entries": {
      "stealthy-auto-browse": {
        "env": {
          "STEALTHY_AUTO_BROWSE_URL": "http://localhost:8080"
        }
      }
    }
  }
}

Or set the environment variable directly:

export STEALTHY_AUTO_BROWSE_URL=http://localhost:8080

3. Verify: Visit http://localhost:5900 to see the browser via VNC.

Control a headless browser that evades bot detection. Uses Camoufox (custom Firefox) with OS-level mouse/keyboard input via PyAutoGUI - completely undetectable by behavioral analysis.

Key Concepts

Why this exists: Standard browser automation (Selenium, Playwright, Puppeteer) is detectable via CDP (Chrome DevTools Protocol) fingerprinting and synthetic event detection. This browser:

• Uses Firefox (no CDP exposure)
• Generates consistent fingerprints via browserforge
• Executes mouse/keyboard at OS level, not via JavaScript injection

Two input modes:

1. Playwright methods (click, fill, type) - Uses CSS selectors, faster but detectable
2. System methods (system_click, system_type, mouse_move) - OS-level input, undetectable but needs coordinates

Workflow for undetectable interaction:

1. Navigate to page with goto
2. Get elements with get_interactive_elements
3. Find target element's x, y coordinates
4. Click using system_click with those coordinates
5. Type using system_type

Environment

Set STEALTHY_AUTO_BROWSE_URL to the browser API endpoint (e.g., http://localhost:8080).

API Reference

All commands are POST requests to $STEALTHY_AUTO_BROWSE_URL/ with JSON body {"action": "<name>", ...params}.

Navigation

goto - Navigate to URL

{
  "action": "goto",
  "url": "https://example.com",
  "wait_until": "domcontentloaded"
}

wait_until: domcontentloaded (default), load, networkidle

Undetectable Input (Use These!)

system_click - Move mouse and click at coordinates (UNDETECTABLE)

{ "action": "system_click", "x": 500, "y": 300, "duration": 0.3 }

duration: Optional mouse movement duration in seconds

mouse_move - Move mouse to coordinates without clicking

{ "action": "mouse_move", "x": 500, "y": 300, "duration": 0.5 }

mouse_click - Click at current position or specific coordinates

{"action": "mouse_click"}
{"action": "mouse_click", "x": 500, "y": 300}

system_type - Type text using OS keyboard (UNDETECTABLE)

{ "action": "system_type", "text": "hello world", "interval": 0.08 }

interval: Delay between keystrokes in seconds (default: 0.08)

send_key - Send special key (enter, tab, escape, etc.)

{"action": "send_key", "key": "enter"}
{"action": "send_key", "key": "tab"}
{"action": "send_key", "key": "escape"}

scroll - Scroll page

{"action": "scroll", "amount": -3}
{"action": "scroll", "amount": 5, "x": 500, "y": 300}

amount: Negative = scroll down, positive = scroll up

Playwright Input (Detectable - Use Only When Necessary)

click - Click element by CSS selector (detectable)

{"action": "click", "selector": "#submit-btn"}
{"action": "click", "selector": "button.login"}

fill - Fill input field (clears first)

{
  "action": "fill",
  "selector": "input[name='email']",
  "value": "user@example.com"
}

type - Type into element character by character

{ "action": "type", "selector": "#search", "text": "query", "delay": 0.05 }

Page Inspection

get_interactive_elements - Get all clickable elements with coordinates (ESSENTIAL!)

{ "action": "get_interactive_elements", "visible_only": true }

Returns array of elements:

{
  "elements": [
    {
      "i": 0,
      "tag": "button",
      "text": "Sign In",
      "selector": "#login-btn",
      "x": 500,
      "y": 300,
      "w": 100,
      "h": 40,
      "visible": true
    }
  ]
}

Use x and y coordinates with system_click for undetectable clicks!

get_text - Get page text content

{ "action": "get_text" }

get_html - Get full page HTML

{ "action": "get_html" }

eval - Execute JavaScript in page context

{"action": "eval", "expression": "document.title"}
{"action": "eval", "expression": "window.scrollY"}

Screenshots

GET /screenshot/browser - Browser viewport PNG GET /screenshot/desktop - Full desktop PNG

Use curl or fetch:

curl $STEALTHY_AUTO_BROWSE_URL/screenshot/browser -o screenshot.png

State & Utility

GET /state - Current browser state (url, title, window_offset)

GET /health - Health check (returns "ok")

ping - Check connection, returns current URL

{ "action": "ping" }

close - Shut down the browser and server

{ "action": "close" }

calibrate - Recalculate window offset for coordinate translation

{ "action": "calibrate" }

Resolution Control

set_resolution - Change display resolution

{ "action": "set_resolution", "width": 1920, "height": 1080 }

Note: Width < 450 requires USE_VIEWPORT=true environment variable

get_resolution - Get current resolution

{ "action": "get_resolution" }

reset_resolution - Reset to original resolution

{ "action": "reset_resolution" }

Fullscreen

enter_fullscreen / exit_fullscreen

{"action": "enter_fullscreen"}
{"action": "exit_fullscreen"}

Complete Workflow Example

Here's how to login to a site undetectably:

# 1. Navigate to login page
curl -X POST $STEALTHY_AUTO_BROWSE_URL -H "Content-Type: application/json" \
  -d '{"action": "goto", "url": "https://example.com/login"}'

# 2. Get all interactive elements
curl -X POST $STEALTHY_AUTO_BROWSE_URL -H "Content-Type: application/json" \
  -d '{"action": "get_interactive_elements"}'
# Response shows email input at x:400, y:200 and password at x:400, y:260

# 3. Click email field (undetectable)
curl -X POST $STEALTHY_AUTO_BROWSE_URL -H "Content-Type: application/json" \
  -d '{"action": "system_click", "x": 400, "y": 200}'

# 4. Type email (undetectable)
curl -X POST $STEALTHY_AUTO_BROWSE_URL -H "Content-Type: application/json" \
  -d '{"action": "system_type", "text": "user@example.com"}'

# 5. Click password field
curl -X POST $STEALTHY_AUTO_BROWSE_URL -H "Content-Type: application/json" \
  -d '{"action": "system_click", "x": 400, "y": 260}'

# 6. Type password
curl -X POST $STEALTHY_AUTO_BROWSE_URL -H "Content-Type: application/json" \
  -d '{"action": "system_type", "text": "secretpassword"}'

# 7. Find and click submit button (from get_interactive_elements, e.g., x:400, y:320)
curl -X POST $STEALTHY_AUTO_BROWSE_URL -H "Content-Type: application/json" \
  -d '{"action": "system_click", "x": 400, "y": 320}'

Response Format

All POST responses follow this format:

{
  "success": true,
  "timestamp": 1234567890.123,
  "data": { ... },
  "error": "message if failed"
}

Tips

1. Always use get_interactive_elements first to find coordinates for clicking
2. Prefer system_click and system_type over Playwright methods for stealth
3. Add small delays between actions to appear more human-like
4. Use calibrate if clicks seem offset after page changes
5. Check screenshots to debug what the browser sees
6. Match timezone - The container's TZ must match the IP's location for fingerprint consistency

VNC Access

Connect to port 5900 via noVNC web viewer to see the browser in action.