Chrome Extension Relay Helper - Mac

Attach the OpenClaw Browser Relay Chrome extension to a live tab so the browser tool (profile="chrome") works. Use this skill before any browser automation t...

Security Scan

The skill's code and instructions are consistent with its stated purpose (attaching a Chrome relay extension on macOS); it performs intrusive but explained actions (killing Chrome, patching Chrome preferences, taking a debug screenshot) which are justified by the described workflow.

Purpose & Capability

The name/description match the included script and instructions: the skill locates and clicks the OpenClaw Relay extension icon via the macOS accessibility tree so the browser(profile="chrome") relay can be attached. Required tools (Peekaboo) and the node Accessibility permission are explained and reasonably related to the task.

Instruction Scope

The instructions and script do several system-level things that are explicitly documented: kill Chrome (pkill -9), patch ~/Library/Application Support/Google/Chrome/Default/Preferences to suppress restore dialogs, open and maximize Chrome, inspect the accessibility tree, click the extension icon, and save a debug screenshot to ~/.openclaw/media. All of these are within the skill's stated goal, but they are intrusive operations that affect the user's Chrome state and could cause data loss if unsaved tabs/windows exist.

Install Mechanism

This is an instruction-only skill with a shell script; no install spec is run by the skill itself. The README suggests installing Peekaboo via a Homebrew tap (steipete/tap/peekaboo). Using a third-party Homebrew tap is an explicit user instruction and not executed automatically by the skill; users should audit that tap before installing, but the install suggestion is coherent with the documented dependency.

Credentials

The skill requests no secrets or environment variables. It does require granting Accessibility permission to the node binary (to allow Peekaboo to send UI events), and it reads/writes the Chrome Preferences JSON and writes diagnostic screenshots to ~/.openclaw/media. Those accesses are proportionate to the task but touch sensitive browser state (Preferences) and user files, so users should be aware and back up Chrome state if needed.

Persistence & Privilege

The skill is not always-enabled and does not request persistent platform privileges. It does not modify other skills or system configurations beyond the documented per-user Chrome Preferences patch and writing to ~/.openclaw/media; those changes are self-contained and described in the SKILL.md.

Assessment

This skill appears to do exactly what it says, but it performs intrusive actions that you should accept knowingly: it will kill any running Chrome processes (pkill -9), modify your Chrome Default Preferences file to suppress restore prompts, and write debug screenshots to ~/.openclaw/media. Before using it: (1) back up your Chrome profile (~/Library/Application Support/Google/Chrome) if you have unsaved tabs or important state, (2) inspect the brew tap/package before installing Peekaboo, (3) confirm you are comfortable granting Accessibility permission to your node binary (required so Peekaboo can send clicks), (4) ensure the OpenClaw extension is loaded and pinned at ~/.openclaw/browser/chrome-extension as documented, and (5) run the script manually the first time so you can observe its behavior (it is not auto-installed). If any of these operations are unacceptable (killing Chrome or modifying preferences), do not install or run the skill.

OpenClaw Chrome Relay Helper - Mac

Automates attaching the OpenClaw Browser Relay Chrome extension to a live tab on macOS. Once attached, the browser(profile="chrome") tool works — you can navigate, snapshot, click, and scrape using your real Chrome session.

macOS only. Requires Peekaboo (macOS UI automation CLI).

Quickstart

bash <skill-dir>/scripts/attach.sh

Outputs one of:

• ALREADY_ATTACHED — already connected, nothing to do
• ATTACHED — freshly connected, ready to use
• FAILED: <reason> — check ~/.openclaw/media/relay-attach-fail.png for a debug screenshot

Then navigate and automate:

browser(action="navigate", profile="chrome", targetUrl="https://example.com")
browser(action="snapshot", profile="chrome", compact=True)  # read page content

Typical wall time: ~29s on a clean launch.

Prerequisites

1. Peekaboo (macOS UI automation CLI)

brew install steipete/tap/peekaboo

Peekaboo reads Chrome's accessibility tree to find the extension icon by description — no pixel coordinates needed.

2. Accessibility permission for node

Go to System Settings → Privacy & Security → Accessibility and add your node binary. Without this, Peekaboo cannot send click events.

Find your node path with: which node

3. openclaw.json browser profile

Add this to your ~/.openclaw/openclaw.json:

"browser": {
  "profiles": {
    "chrome": {
      "cdpUrl": "http://127.0.0.1:18792",
      "driver": "extension",
      "color": "#FF5A36"
    }
  }
}

Restart the gateway after editing: openclaw gateway restart

4. Extension loaded and pinned in Chrome

The OpenClaw Browser Relay extension must be loaded as an unpacked extension in Chrome. It's included with OpenClaw at:

~/.openclaw/browser/chrome-extension

Load it via chrome://extensions → Developer mode ON → Load unpacked.

The extension must also be pinned to the toolbar. The script finds the icon via Chrome's accessibility tree, which only exposes toolbar-pinned extensions — not icons hidden inside the Extensions panel. To pin: click the puzzle-piece icon → click the pin icon next to "OpenClaw Browser Relay".

How it works

The script finds the extension icon using Chrome's accessibility tree — not pixel coordinates. The icon's description changes based on state:

• Detached: "OpenClaw Browser Relay (click to attach/detach)"
• Attached: "OpenClaw Browser Relay: attached (click to detach)"

Peekaboo scans for a pop-up button element whose description starts with "OpenClaw Browser Relay", determines state, and clicks to attach if needed. Retries up to 8× (every 2s) to handle slow Chrome startup.

Window maximize is required before scanning — Chrome's toolbar icons are not visible in the accessibility tree on a small or default-sized window.

What the script does (step by step)

1. Fast path — if Chrome is running and badge already shows "attached", exits immediately (~2s)
2. Kill any running Chrome instance
3. Patch ~/Library/Application Support/Google/Chrome/Default/Preferences to suppress the "Restore Pages?" dialog on relaunch
4. Open Chrome to https://info.cern.ch/ — the world's first website, a 428-byte static HTML file with zero anti-bot tech, JS, cookies, or Cloudflare
5. Maximize the window via Peekaboo (required for toolbar visibility)
6. Scan accessibility tree for the extension icon (retries up to 8×, 2s apart)
7. Click the icon to attach
8. Verify state changed to "attached" before returning

Known pitfalls

Approach	Why it doesn't work
Hardcoded pixel coordinates	Breaks at any screen size other than what they were measured on
AppleScript keystroke with Ctrl+Shift	Modifier keys are silently dropped — only the bare key fires
Chrome extension keyboard shortcut	The extension manifest has no commands — shortcuts don't trigger attach
Vision model to locate icon	Not reliable enough for toolbar UI at any resolution
Skipping window maximize	Toolbar icons don't appear in the accessibility tree on a small window
Extension not pinned to toolbar	Unpinned extensions are hidden inside the Extensions panel — not visible in the accessibility tree
Wrong profile name or port in config	browser(profile="chrome") requires the profile named exactly chrome pointing to port 18792

Token efficiency tips

# ✅ Use snapshot for reading page content (~3k tokens)
browser(action="snapshot", profile="chrome", compact=True)

# ❌ Avoid screenshot + vision for UI element detection
# → Unreliable for toolbar/coordinate identification
# → 10–50x more expensive than snapshot

Integration pattern

Any skill that needs Chrome relay should call this first:

# 1. Attach
bash <path-to-chrome-relay>/scripts/attach.sh

# 2. Navigate
browser(action="navigate", profile="chrome", targetUrl="https://target.com")

# 3. Automate
browser(action="snapshot", profile="chrome", compact=True)
browser(action="act", profile="chrome", request={kind: "click", ref: "..."})