[BUG] Cowork desktop session crashes/fails on resume; 13GB VM bundle + "Argument list too long" on resume spawn

Preflight Checklist

• [x] I have searched existing issues 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?

My Cowork desktop session consistently crashes / fails to resume. The session represents months of work and I'd like to recover it rather than start a new project. Logs point to an oversized Cowork VM image and a resume-time spawn error.

What Should Happen?

Cowork project should be able to restart the session or continue from summary. Instead, the application crashes and project is unable to be used.

Error Messages/Logs

| Source | Signature | Count |
|---|---|---|
| `cowork_vm_node.log` | `OSError: [Errno 7] Argument list too long: 'node'` (on a `claude --resume …` spawn with a very long `--allowedTools`/`--tools` argument list) | 1 |
| `cowork_vm_node.log` | `PermissionError: [Errno 13] Permission denied: '/tmp/cp.js'` | 1 |
| `cowork_vm_node.log` | `[VM:network] Not connected, will show error in 30000ms if not resolved` | 17 |
| `main.log` | `Sentry caught: { type: 'Unknown', value: 'No message', stack: undefined }` | 8 |
| `main.log` | `oauth failed (not_logged_in): cache cleared mid-flow` | 8 |
| `main.log` | `marketplace_migration.paths_unavailable: No active account/org for marketplace operations` | repeated |

There is also an older macOS "disk writes" resource-violation report for the app (~2.1 GB of file-backed memory dirtied over ~10.6 hrs, exceeding the daily write limit; heaviest stack `uv_cancel → write` in the Electron Framework) — consistent with the very large VM image being written heavily.

Steps to Reproduce

Steps to reproduce

1. Open the desktop app and resume an existing, long-running Cowork session.
2. The session crashes / fails to come back.

Restarting the app and rebooting the machine do not help. The app is on the latest version.

Claude Model

Opus

Is this a regression?

Yes, this worked in a previous version

Last Working Version

_No response_

Claude Code Version

1.8555.0

Platform

Anthropic API

Operating System

macOS

Terminal/Shell

Terminal.app (macOS)

Additional Information

Environment

• Claude for Desktop: 1.8555.0 (current/latest). An earlier macOS crash report was logged under 1.7196.0.
• macOS 26.3 (build 25D125), Apple Silicon (arm64), 16 GB RAM
• Free disk: ~84 GB of 228 GB (not a disk-space issue)

What the data shows

The desktop store ~/Library/Application Support/Claude is ~15 GB, dominated by:

• vm_bundles/claudevm.bundle — ~13 GB (rootfs.img ~10 GB + rootfs.img.zst ~2 GB)
• local-agent-mode-sessions/<id> — ~1.6 GB

GitHub diagnosis confirms root cause — unbounded --resume arg list hitting ARG_MAX (E2BIG) + uncompacted 13GB rootfs.img.

Likely culprit

The Argument list too long error when spawning the resumed claude process stands out: the resume command line includes a very large --allowedTools/--tools list, which can exceed the OS argument-length limit (E2BIG) and prevent the session from resuming.

What I've already tried

• Restarted the app
• Rebooted the machine
• Confirmed I'm on the latest desktop version

Impact / request

This session contains months of work. I'd like (a) a fix so the session resumes, and (b) guidance on safely recovering or migrating the existing session/VM state without starting over. Happy to provide full logs privately (~/Library/Logs/Claude/main.log, coworkd.log, cowork_vm_node.log).