claude-code - 💡(How to fix) Fix [BUG] Cowork tasks crash with exit code 1 immediately after spawn on Desktop 1.3036.0 (macOS 26.3.1a, Apple M3) — model-independent, VM healthy [3 comments, 4 participants]

Preflight Checklist

• I have searched existing issues and this hasn't been reported yet
• This is a single bug report (please file separate reports for different bugs)
• I am using the latest version of Claude Code

What's Wrong?

Every new Cowork task crashes within ~1 second of the Claude Code binary spawning inside the VM, displaying "Something went wrong — Claude Code process exited with code 1" in the task panel.

The VM side is fully healthy: startup steps complete, API reachability confirmed, SDK 2.1.111 installs successfully, and OAuth approves. The crash happens only after the binary is spawned. Reproduced on every task I've attempted since updating to 1.3036.0, including trivial prompts like "what time is it".

Regular Chat works fine — only Cowork is affected.

What Should Happen?

A Cowork task should accept a prompt and produce a response. The spawned Claude Code process should remain healthy (cycle_health: 'healthy', had_first_response: true) rather than exiting with code 1 before producing any output.

Error Messages/Logs

Process spawned id=69bb5ce1-93f0-457d-bb13-6d249de73f23 name=happy-youthful-euler command=/usr/local/bin/claude
  session_id: 'local_a7d68912-dba4-460a-b051-3475200c0caf',
  vm_instance_id: '860ca488-9b1b-4e79-a09c-ee2cc3498b05',
  model: 'claude-opus-4-7[1m]',
  cycle_health: 'unhealthy',
  had_first_response: false,
  seconds_to_outcome: 1,
  is_resume: false,
  unhealthy_reason: 'system_error',
  error_message: 'Claude Code process exited with code 1',
  error_category: 'process_crashed',
  transcript_size_bytes: undefined

Preceding VM startup (all healthy):
[VM] VM startup step: install_ca_certificates completed
[VM] API reachability: REACHABLE
[VM] SDK installed successfully: version=2.1.111
[VM] OAuth token approved successfully

Steps to Reproduce

1. Update Claude Desktop to 1.3036.0 (cf521c) on macOS 26.3.1(a), Apple Silicon (M3)
2. Open the Cowork tab
3. Click "New task"
4. Enter any prompt (e.g., "what time is it")
5. Observe: VM startup completes, Progress panel shows 2 steps complete, then error banner appears: "Something went wrong — Claude Code process exited with code 1"

Reproduces 100% of attempts. Tested with models: Opus 4.7 (claude-opus-4-7[1m]), Opus 4.6, and Sonnet 4.6 — all crash identically within ~1 second of spawn.

Claude Model

None

Is this a regression?

Yes, this worked in a previous version

Last Working Version

Unknown — I know Cowork was working in recent builds prior to 1.3036.0, but I don't have the exact last-working version number.

Claude Code Version

Claude Code SDK 2.1.111 (per VM log: "SDK installed successfully: version=2.1.111"). Cannot run claude --version directly — binary runs inside Cowork VM at /usr/local/bin/claude, not accessible from host shell.

Platform

Anthropic API

Operating System

macOS

Terminal/Shell

Terminal.app (macOS)

Additional Information

Environment

• Claude Desktop: 1.3036.0 (cf521c), built 2026-04-16T16:17:33Z
• macOS: Tahoe 26.3.1 (a)
• Hardware: iMac 24-inch (2023), Apple M3, 24 GB RAM
• Plan: Max 20x

What I've ruled out

• Not macOS 26.3 vsock/VM networking (possibly-related issues #23830, #32225, #47644) — VM networking is confirmed healthy in logs
• Not model-specific — reproduced on Opus 4.7, Opus 4.6, and Sonnet 4.6
• Not OAuth/auth — token approves successfully before the spawn
• Not SDK install failure — SDK 2.1.111 installs cleanly
• Not user workspace/config — fresh "New task" from empty prompt reproduces it
• Regular Chat in the same Desktop app works fine

Signature The combination of had_first_response: false, seconds_to_outcome: 1, and error_category: 'process_crashed' suggests the binary is crashing on startup inside the VM, before it accepts any input.

Happy to provide full log tails or Console.app output if helpful.