Claude Code desktop app silently launches and immediately kills a Google Chrome process in the background

### Preflight Checklist

- [x] I have searched [existing issues](https://github.com/anthropics/claude-code/issues?q=is%3Aissue%20state%3Aopen%20label%3Abug) and this hasn't been reported yet
- [x] This is a single bug report (please file separate reports for different bugs)
- [x] I am using the latest version of Claude Code

### What's Wrong?

While working in a Claude Code session (no browser-automation tool was being called at the time), Google Chrome kept quitting/restarting unprompted, several times over about 30 minutes. This was disruptive enough that I couldn't reliably keep Chrome open alongside my session.

I confirmed via macOS unified log (log show) that this is not a crash (no entry in chrome://crashes, no ~/Library/Logs/DiagnosticReports entry, no OOM/jetsam kill) and not related to any Miro MCP tool use (that integration is pure HTTP/API, no browser involved).

The log shows Claude Code itself launching a fresh Chrome process and killing it ~3 seconds later, with no corresponding tool call on my side at that timestamp:

### What Should Happen?

I.e. Chrome PID 11952 was launched under Claude Code's TCC responsibility at 13:09:48 and reported terminated at 13:09:51 — about 3 seconds later — with no explicit browser-automation tool call from my side in that window.

This pattern repeated multiple times over the session (different PIDs each time), always with responsible = com.anthropic.claude-code. Disabling the "Claude → Google Chrome" toggle in System Settings → Privacy & Security → Automation did NOT stop it, which makes sense since that toggle only governs AppleEvents/osascript automation, not this process-launch/kill pattern.

Environment:

Claude Code desktop app version: 2.1.270
OS: macOS (Darwin 25.6.0)
Chrome version at the time: 153.0.8010.36
"Claude in Chrome" browser extension installed and previously used earlier in the session, but not actively invoked at the time of the incidents described above.
Impact: Chrome closes/restarts unexpectedly while the user is working in other Chrome tabs, losing tab state and interrupting work. chrome://crashes shows nothing, so from the user's side this looks like an unexplained, repeated force-quit.

Ask: Please investigate what background process/feature in the Claude Code desktop app is launching and then killing a Chrome instance every so often, and either stop doing this when not actively needed, or make it not affect a Chrome window the user is actively using (e.g. use an isolated/headless profile instead of the user's regular Chrome instance).

### Error Messages/Logs

```shell
13:09:48.772 tccd: AUTHREQ_ATTRIBUTION msgID=11952.1
responsible = { identifier: com.anthropic.claude-code, pid: 1802,
responsible_path: /Users//Library/Application Support/Claude/claude-code/2.1.270/claude.app/Contents/MacOS/claude }
requesting = { identifier: com.google.Chrome, pid: 11952,
binary_path: /Applications/Google Chrome.app/Contents/MacOS/Google Chrome }

13:09:51.529 runningboardd: [anon(501):11952] termination reported by proc_exit
```

### Steps to Reproduce

Have the "Claude in Chrome" browser extension installed and paired with a Claude Code desktop session (macOS).
During a long-running Claude Code session, use mcp__claude-in-chrome__* tools at least once (e.g. select_browser, tabs_context_mcp, navigate) to connect to the user's real Chrome browser.
Continue working in the same Claude Code session for an extended period (30+ minutes) WITHOUT calling any more claude-in-chrome tools, while keeping Google Chrome open and in normal use (browsing other tabs, etc.).
Observe that Chrome unexpectedly quits/restarts on its own several times during that period, with no corresponding tool call from the assistant at that moment.
To confirm it's not a normal crash, check chrome://crashes (empty) and ~/Library/Logs/DiagnosticReports/ (no new Chrome crash report).
To confirm the responsible process, run (while it's happening, or right after):
/usr/bin/log show --last 5m --style compact \
--predicate 'eventMessage CONTAINS "Google Chrome"' \
| grep -E "AUTHREQ_ATTRIBUTION|termination reported by proc_exit"
This shows a tccd AUTHREQ_ATTRIBUTION entry where responsible = com.anthropic.claude-code and requesting = com.google.Chrome, followed a few seconds later by runningboardd: ... termination reported by proc_exit for that same Chrome PID — i.e. Claude Code launched that Chrome process and it was killed within ~3 seconds, without any user- or assistant-initiated browser action at that timestamp.
Note: disabling System Settings → Privacy & Security → Automation → "Claude" → "Google Chrome" does NOT stop this — the behavior persists, which suggests it isn't going through AppleEvents/osascript but a direct process launch/kill from the app itself.
I was not able to find a single deliberate user action that triggers each individual occurrence — it appears to happen periodically/spontaneously in the background after the browser-automation feature has been used once in a session, rather than being tied to a specific reproducible click or command.

### Claude Model

None

### Is this a regression?

Yes, this worked in a previous version

### Last Working Version

2.1.272 (Claude Code)

### Claude Code Version

2.1.272 (Claude Code)

### Platform

Anthropic API

### Operating System

macOS

### Terminal/Shell

Terminal.app (macOS)

### Additional Information

_No response_