[BUG] Cowork: "Failed to create bridge sockets after 5 attempts" — all sessions crash with exit code 1, persists after full reinstall (macOS M4 Pro)

Preflight Checklist

• [x] I have searched existing issues and this hasn't been reported yet
• [ ] - [x] This is a single bug report
• [ ] - [x] I am using the latest version of Claude Desktop

Environment

• Claude Desktop version: 1.1.4328
• - Claude Code SDK: 2.1.111
• - - macOS: Sequoia, Darwin 25.3.0
• - - - Hardware: Apple M4 Pro, 48GB RAM

What's Wrong?

Every Cowork session — new and resumed — crashes instantly with Claude Code process exited with code 1. The VM boots successfully, network connects, API is reachable, OAuth token is valid. The crash happens ~1 second after spawn, consistently.

Issue started: April 15, 2026, coinciding with the Anthropic platform outage. Has not recovered since service was restored.

Key error from cowork_vm_node.log

[Spawn:vm] Spawn succeeded in 48ms
[vm-stderr] Error: Failed to create bridge sockets after 5 attempts
[Process] Exited, code=1, signal=null, duration=972ms, oom=false

From coworkd.log:

[process:bdb01aea] wrote srt-settings (domains=23 parentProxy=false) to /etc/srt-settings/bdb01aea.json
[process:bdb01aea] started PID 4009
[process:bdb01aea] direct child exited: code=1 signal=<nil>

What Should Happen?

Cowork sessions should start and run successfully as they did prior to April 15, 2026.

Steps to Reproduce

1. Open Claude Desktop
2. 2. Navigate to Cowork
3. 3. Start any new session (or attempt to resume existing)
4. 4. Session immediately fails with exit code 1

Steps Already Tried — None Resolved It

• Force quit and full restart / reboot
• - Cleared IndexedDB and Session Storage
• - - Cleared local-agent-mode-sessions and claude-code-sessions
• - - - Deleted vm_bundles/claudevm.bundle (forced full ~10GB re-download)
• - - - - Deleted claude-code-vm directory
• - - - - - Started brand new sessions (confirmed is_resume: false in logs)
• - - - - - - Full uninstall + reinstall from claude.ai/download

All attempts produce identical error. VM infrastructure (boot, network, API reachability) is healthy — failure is specific to bridge socket creation inside the VM.