Browser Gemini Search

Search the web using Google Gemini via OpenClaw-controlled Chrome with remote debugging enabled and an approved user profile.

Security Scan

The skill's requests and runtime instructions match its stated purpose (controlling the user's Chrome to use Gemini); it is coherent but has expected privacy and configuration risks you should consider before enabling.

Purpose & Capability

The name/description (use Chrome remote-debugging to drive Gemini) aligns with the instructions: all runtime steps are browser control calls and guidance to start Chrome with --remote-debugging-port. There are no unrelated required binaries, env vars, or credentials.

Instruction Scope

Instructions stay within the stated task (connect to Chrome, find/open Gemini, type query, read response). However the SKILL.md also tells the user to add '*.google.com' to browser.ssrfPolicy in openclaw.json if SSRF blocks Gemini — this is a configuration change that broadens an allowlist and should be applied deliberately. The skill also instructs running Chrome with remote debugging and to use an existing user profile, which implicitly gives the agent access to any logged-in sessions in that profile.

Install Mechanism

Instruction-only skill with no install spec and no code files; nothing is downloaded or written to disk by the skill itself.

Credentials

The skill requests no environment variables or credentials, which is appropriate. But it requires controlling the user's Chrome profile (profile="user"), meaning it can access cookies, authenticated sessions, and other browser state — a high-privilege local capability that is proportionate to the stated purpose but sensitive in practice.

Persistence & Privilege

always:false and no install actions mean the skill does not demand permanent presence or system-wide changes. The skill can be invoked autonomously by the agent (platform default); combine that with browser control if you want to limit autonomous behavior.

Assessment

This skill appears to do what it claims, but it controls your running Chrome and will operate in whatever profile you point it at — so it can read cookies, sessions, and other sensitive browser state. Before installing or using it: (1) Prefer creating and using a dedicated Chrome profile with minimal logins for this skill instead of your main account. (2) Only enable Chrome's --remote-debugging-port when you trust the local environment and disable it when finished; the debugging port can expose powerful control if other local processes can reach it. (3) Review any suggested edits to openclaw.json (adding *.google.com to an SSRF allowlist) carefully — widening allowlists increases SSRF risk. (4) If you want tighter control, disable autonomous invocation for this skill or require manual confirmation before it runs. (5) Avoid using this on shared or untrusted machines. If you need help reducing risk, consider running Gemini queries in an isolated, logged-out browser profile or use an API-based integration that doesn't rely on controlling a logged-in browser session.

Use OpenClaw's browser tool to control the user's Chrome and search Gemini.

Workflow

1. Ensure browser is connected

   • Run browser(action="start", profile="user", target="host")
   • If attachOnly error or timeout: Chrome is not running with debugging port
     • Ask user to run: & "C:\Program Files\Google\Chrome\Application\chrome.exe" --remote-debugging-port=9222
     • Then retry connection

2. Find or open Gemini tab

   • Run browser(action="tabs", profile="user", target="host") to list open tabs
   • Look for existing Gemini tab (URL contains gemini.google.com)
   • If found: browser(action="focus", targetId="<id>", profile="user", target="host")
   • If not found: open new tab via browser(action="navigate", url="https://gemini.google.com", target="host")

3. Wait for page load

   • Run browser(action="snapshot", profile="user", target="host") to verify page is ready
   • Look for the input textbox (usually textbox with placeholder like "Ask Gemini" or "输入双子座的提示")

4. Type the search query

   • Use browser(action="act", kind="type", ref="<textbox_ref>", text="<user's search query>", profile="user", target="host")
   • Then browser(action="act", kind="click", ref="<send_button_ref>", profile="user", target="host") to send

5. Read Gemini's response

   • Wait 5-10 seconds for response to generate
   • Run browser(action="snapshot", profile="user", target="host") to read the answer
   • Present the answer to the user

Quick reference

# Step 1: connect
browser(action="start", profile="user", target="host")

# Step 2: find tab or navigate
browser(action="tabs", profile="user", target="host")
browser(action="focus", targetId="11", profile="user", target="host")  # if found
browser(action="navigate", url="https://gemini.google.com", target="host")  # if not found

# Step 3 & 4: type and send
browser(action="act", kind="type", ref="1_1236", text="search query here", profile="user", target="host")
browser(action="act", kind="click", ref="2_2", profile="user", target="host")  # send button

# Step 5: read response
browser(action="snapshot", profile="user", target="host")

Common issues

• "Chrome MCP existing-session attach timed out": Chrome debugging port not enabled. User must restart Chrome with --remote-debugging-port=9222.
• SSRF blocked URL: The Gemini domain must be in browser.ssrfPolicy.hostnameAllowlist in openclaw.json. Add if missing: *.google.com
• Tab focus fails: Use correct targetId from tabs output
• Input ref changes: Re-run snapshot to get fresh refs after page navigation