Edge DOM Browser Skill

This skill defines the architecture, connection mechanics, supported commands, and troubleshooting guides for utilizing the Edge DOM browser (Edge DOM Bridge) in Antigravity.

I. Architecture & Setup

The Edge DOM browser allows direct programmatic interaction with the user's personal Microsoft Edge instance at the DOM level using a 3-piece bridge system:

Edge Extension: An unpacked developer extension (global_workflows/edge_dom_bridge_extension/content.js) injected into every page.
- Location: global_workflows/edge_dom_bridge_extension/
- Extension ID: cdkodipddhcdfjgophblpnkcgepookpb
Native Messaging Host: A python script (global_workflows/edge_dom_bridge_extension/native_host.py) automatically spawned by Microsoft Edge.
TCP Socket Server: Listens on loopback address 127.0.0.1:18900 where scripts connect to send JSON-RPC commands.

[!CAUTION] FOREGROUND REQUIREMENT: 🚨 The DOM Bridge only reads the ACTIVE TAB. If automation requires switching tabs using Win32 keyboard events (like Ctrl+Tab), Edge MUST be in the foreground. When executing these actions, you MUST put a bold message with red emojis (e.g., 🚨 WARNING: BRING EDGE TO FRONT 🚨) in your response so the user focuses the Edge window.

II. Connection Snippet

To send commands over the bridge, run a Python script via the run_command tool connecting to port 18900:

import socket
import json

def edge_dom(command):
    sock = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
    sock.settimeout(15)
    sock.connect(("127.0.0.1", 18900))
    sock.sendall(json.dumps(command).encode("utf-8") + b"\n")
    response = b""
    while b"\n" not in response:
        chunk = sock.recv(65536)
        if not chunk: break
        response += chunk
    sock.close()
    return json.loads(response.decode("utf-8").strip())

# Example action: Retrieve active page metadata
result = edge_dom({"action": "get_page_info"})
print(json.dumps(result, indent=2))

III. Supported Commands

{"action": "get_page_info"} — Returns the URL, title, and current scroll position of the active tab.
{"action": "get_all_text"} — Returns all visible text on the page.
{"action": "get_text", "selector": "CSS_SELECTOR"} — Extracts text from a specific CSS selector.
{"action": "get_elements"} — Returns a semantic element map with interactive tags (e.g., @e1, @e2).
{"action": "click", "selector": "CSS_SELECTOR"} — Click element by CSS selector.
{"action": "click_tag", "tag": "@e5"} — Click element by its semantic bridge tag.
{"action": "type", "selector": "CSS_SELECTOR", "text": "value"} — Type into an input element.
{"action": "scroll", "amount": 500} — Scroll down by pixels.
{"action": "get_value", "selector": "CSS_SELECTOR"} — Get value of an input field.
{"action": "eval_selector", "selector": "CSS_SELECTOR"} — Evaluate element selector and return match count.

IV. Troubleshooting & Fallbacks

1. ConnectionRefusedError (WinError 10061)

If the socket connection is refused, verify the following:

Is the Extension Enabled? Check edge://extensions and verify that the unpacked "Antigravity DOM Bridge" is ON. Edge restarts can sometimes silently disable unpacked extensions.
Refresh the Tab: The content script must be injected. Ask the user or programmatic navigation to refresh the active tab.
Manifest ID Match: Check allowed_origins in the native messaging manifest (global_workflows/edge_dom_bridge_extension/antigravity_dom_bridge.json). The extension ID MUST match cdkodipddhcdfjgophblpnkcgepookpb.
Testing Tool: Run python global_workflows/edge_dom_bridge_extension/test_bridge.py to diagnose the connection.

2. HTTPS Only

The DOM Bridge works exclusively on https:// websites. It will time out or crash on localhost, chrome://, or edge:// URLs.

3. Shadow DOM Limitations

For pages rendering elements inside shadow DOM trees (such as Google Search Console), standard DOM commands will fail to query elements.

Fallback: Switch to the Visible Browser (Playwright physical coordinates) or vision-based Point and Click direct clicks, which bypass DOM-traversal limitations.

Edge DOM Skill