[Windows desktop] Remote Control sessions exit ~4s after spawn since bundled CLI 2.1.308 #105658

Summary

On Windows, every Remote Control session started from the Claude desktop app fails with "could not start Claude Code in that folder", whatever folder is used. It has failed this way since the bundled CLI was updated to 2.1.308. The last successful session ran shortly before that update.

Environment

• Claude desktop app 4.29.0.0 (MSIX), Windows 11
• Bundled Claude Code CLI 2.1.309. The failures began with 2.1.308.
• Standalone CLI 2.1.311 is also installed. Normal headless runs (claude -p ...) work fine with both versions.

What happens

1. Start Remote Control for any registered folder.
2. The app spawns claude.exe --output-format stream-json --verbose --input-format stream-json --permission-prompt-tool=stdio --await-initialize ... (command line about 3.5K chars).
3. The CLI runs rg --no-config --files --hidden <folder>, then exits with code 0 after 3.7–3.8 s.
4. In the session record (%APPDATA%\Claude\claude-code-sessions\...\local_*.json), createdAt == lastActivityAt and remoteControlSpawn is set. No transcript is written.

What I checked

• I ran the same claude.exe with the same stream-json / --await-initialize arguments by hand. It stays alive while stdin is open and exits 0 as soon as stdin is closed. So the app appears to close stdin during initialization.
• The cause is not the size of the remote MCP config. I removed 6 connectors and restarted the app, which cut remoteMcpServersConfig from about 905K to about 417K chars. That is smaller than in sessions that used to work, and the failure is unchanged.
• Restarting the app does not help.
• The desktop app logs (%APPDATA%\Claude\logs\main.log) have not been written since August, so I can't see the error on the app side.

Expected

The Remote Control session starts and stays connected, as it did before 2.1.308.