mcp__workspace__bash persistently unusable after first command: all subsequent calls return "process already running" (Windows + Cowork)

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?

github-issue-cowork-bash.md

What Should Happen?

When you call mcp__workspace__bash, the command should execute in the session's Linux sandbox and return its output within the given timeout_ms. Once the command completes (or times out), the process slot should be released so the next bash call can run immediately — no leftover oneshot-xxx ID, no already running rejection, no cascading blockage. Every bash call in a session should be independent, regardless of whether a previous one timed out.

Error Messages/Logs

// First call — timeout
mcp__workspace__bash(command="echo hello", timeout_ms=10000)
→ bash failed on resume, create, and re-resume.
  resume: Command timed out after 10000ms

// Second call — slot already taken by orphaned oneshot
mcp__workspace__bash(command="whoami", timeout_ms=15000)
→ bash failed on resume, create, and re-resume.
  resume: Command timed out after 15000ms;
  create: RPC error -1: process with name "confident-brave-feynman"
  already running (id: oneshot-fb9a9f34-b0a2-4281-b94b-85e000ba6129)

// Third call — permanently blocked
mcp__workspace__bash(command="echo ok", timeout_ms=15000)
→ RPC error -1: process with name "confident-brave-feynman"
  already running (id: oneshot-fb9a9f34-b0a2-4281-b94b-85e000ba6129)

// Same pattern after local VM cleanup + restart — new ID, same failure
mcp__workspace__bash(command="echo ok", timeout_ms=15000)
→ bash failed on resume, create, and re-resume.
  resume: Command timed out after 15000ms;
  create: RPC error -1: process with name "confident-brave-feynman"
  already running (id: oneshot-1bf2f78e-216c-4b91-8a8c-2a033988e0e6)

Steps to Reproduce

1. Open Claude Desktop (Windows 11) and start a fresh Cowork session.

1. Call mcp__workspace__bash with any trivial command:

command: "echo hello"
timeout_ms: 10000

1. Observe: the call either times out ("Workspace still starting") or, rarely, succeeds on the very first attempt.

1. Call mcp__workspace__bash a second time with any trivial command:

command: "whoami"
timeout_ms: 15000

1. Observe: the call fails with:

RPC error -1: process with name "confident-brave-feynman" already running (id: oneshot-xxx)

1. Call mcp__workspace__bash a third time with any command.

1. Observe: every subsequent call now returns the same "already running" error for the remainder of the session, regardless of timeout_ms or command content.

1. Run the recommended local cleanup (stop vmwp/cowork-svc, delete local-agent-mode-sessions, restart CoworkVMService), then start a new session and repeat steps 1–7.

1. Observe: the pattern reproduces identically — a new oneshot-xxx ID is generated, but the first call still times out and the second still hits "already running."

Claude Model

None

Is this a regression?

Yes, this worked in a previous version

Last Working Version

_No response_

Claude Code Version

2.1.170

Platform

Other

Operating System

Windows

Terminal/Shell

PowerShell

Additional Information

_No response_