[BUG] [Cowork] mcp__workspace__bash sandbox VM still fails to start — 2-month persistent outage (ref #76374)

### 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?

The isolated Linux VM (mcp__workspace__bash) used by Claude Cowork still fails to start across ALL Cowork sessions (existing and newly created). This was first reported as #76374 on July 9, 2026; that issue was auto-closed by github-actions as "not planned" for inactivity on Sept 3, 2026 without ever being fixed. As of September 2026 the outage is still ongoing — roughly two months of continuous failure.

Error messages observed (alternating):
- "Workspace unavailable. The isolated Linux environment failed to start (VM connection timeout after 60 seconds)."
- - "Workspace still starting. The isolated Linux environment is booting in the background (usually 10-30 seconds). Try again shortly."
The VM never comes up. File tools (Read/Write/Edit) and Desktop Commander MCP work normally — only mcp__workspace__bash is affected. Retries and brand-new sessions do not help.

Anthropic email support was contacted again (Sept 4, 2026); the AI agent confirmed no active incident on status.anthropic.com and had no engineering ETA. The issue persists regardless.

Note: Cowork-specific (not Claude Code CLI), filed here as it appears to share infrastructure. Continuation of #76374.

### What Should Happen?

The mcp__workspace__bash sandbox VM should start within 10-30 seconds and be usable for running bash commands in the isolated Linux environment, as it did before July 9, 2026.

### Error Messages/Logs

```shell
Workspace unavailable. The isolated Linux environment failed to start (VM connection timeout after 60 seconds).
Workspace still starting. The isolated Linux environment is booting in the background (usually 10-30 seconds). Try again shortly.
```

### Steps to Reproduce

1. Open any Claude Cowork session (new or existing).
2. 2. Run any mcp__workspace__bash command (e.g. echo "test").
3. 3. Observe the VM timeout after 60 seconds.
4. 4. Retry multiple times — the error persists.
5. 5. Create a brand-new Cowork session — the same error occurs immediately.

### Claude Model

None

### Is this a regression?

Yes, this worked in a previous version

### Last Working Version

Worked normally until July 9, 2026 (Cowork sandbox)

### Claude Code Version

N/A — Claude Cowork desktop app (not Claude Code CLI), Windows 11 ARM64

### Platform

Anthropic API

### Operating System

Windows

### Terminal/Shell

PowerShell

### Additional Information

Continuation of #76374 (auto-closed as "not planned" by github-actions on Sept 3, 2026 for inactivity — not fixed). An Anthropic email support ticket is also open (conversation ID 215475016975291); the AI agent confirmed no active status-page incident and had no engineering ETA. Only mcp__workspace__bash is affected — all other tools (Read/Write/Edit, Desktop Commander MCP) work. Current workaround: Desktop Commander MCP + local Kali WSL2.

The payload names no repository files or tests; start by reproducing in a Claude Cowork session with an mcp__workspace__bash command such as echo "test". The user-visible done condition is that the isolated Linux VM starts within 10-30 seconds and the 60-second timeout messages no longer appear. This likely needs maintainer access to Cowork sandbox infrastructure rather than a small public-repo patch.

由索引模型根据 Issue 内容生成。