[Bug] Claude Code hangs indefinitely during processing with no error output

Found 3 possible duplicate issues:

1. https://github.com/anthropics/claude-code/issues/13224
2. https://github.com/anthropics/claude-code/issues/13198
3. https://github.com/anthropics/claude-code/issues/13188

This issue will be automatically closed as a duplicate in 3 days.

• If your issue is a duplicate, please close it and 👍 the existing issue instead
• To prevent auto-closure, add a comment or 👎 this comment

🤖 Generated with Claude Code

Yes. Claude Code no longer works/is down. I have switched to Codex until this is fixed.

Yes i have the same issue. Back to codex.

Same here in macos v2.0.60, v2.0.59, v1.0.126
Claude Code CLI

Same here on MacOS (Apple Silicon). It's primarily running parallel instances (I usually use 2/3 max) and they're just hanging indefinitely. I've checked my usage and I'm no where near it.

Guys, is it happening only when running inside VS code for you?
I am thinking that it might be a problem with vs code integration

Guys, is it happening only when running inside VS code for you? I am thinking that it might be a problem with vs code integration

No, it's happening for me in my Terminal (Ghostty) and in Cursor.

This is happening very consistently for me - Only able to do one terminal at a time. Happens even without Opus (only Sonnet), seems like a bug the CLI itself since Close/Resume fixes it

Also happening for me in Kitty on Arch Linux, Opus and Sonnet

I'm having the same issue, with Bedrock, in Alacritty. Its taking about 3 minutes to prompt for permission to run any shell command

Same exact thing and it is killing me. I'm just gonna downgrade to .59 until fixed. Mine did not occur until .65 -- 2.0.65 of course. Get this baby fixed.. I'm in a tech war and BEAD is coming. I am David fight goliaths and I am getting somewhere here. Lets get this fixed yea?

-----EDIT--- I logged back in... /resume... and it finished the todo list because it did not need my permissions at that point. This may have been a network... Working on my Hotspot... Will revert tonight once home and plugged directly into home network to see what happens. Will report back then. ----

Bug Description
Claude Code constantly stops working. Creating… (esc to interrupt · 13m 41s · ↓ 630 tokens)

Environment Info

Platform: darwin
Terminal: Apple_Terminal
Version: 2.0.69
Feedback ID: 67a35e63-f9a8-48c4-b247-8da2eae25c6c
Errors

This issue has been inactive for 30 days. If the issue is still occurring, please comment to let us know. Otherwise, this issue will be automatically closed in 30 days for housekeeping purposes.

It is still happening. Last time on 2026-02-06 (the last time I started claude), with 2.1.17 (Claude Code).

This is continuously an issue. I can see it on multiple workstations in different countries at the same time, both mac os and windows. This is obviously a server side issue. if the server is overloaded, just tell us, maybe even show a queue. Don't make the UI pretend its actually doing something.

Reproduction via subagent dispatch + remote-control

Adding a specific reproduction angle: this hang behavior is triggered reliably when spawning subagents via the Task tool during a /remote-control session. The interaction pattern:

1. Parent session runs a workflow that dispatches subagents via Task tool (e.g., parallel task execution)
2. One or more subagents enter the silent hang state (no output, no tool calls, no timeout)
3. The parent thread is blocked awaiting Task.prompt() which has no timeout wrapper
4. The /remote-control session has no programmatic equivalent of the Esc interrupt
5. All subsequent prompts queue but never process — session shows "In progress" indefinitely

Why this is worse than the interactive case: In an interactive terminal, Esc breaks the hang. In remote-control / headless / SDK contexts, there is no recovery path short of kill -SIGINT. This blocks CI/CD, unattended workflows, and remote control use cases entirely.

Suggested fix (matching #28482): The Task tool's SessionPrompt.prompt() call needs an inactivity timeout — if no output or tool call is produced within N seconds, auto-cancel the subagent and yield control. A heartbeat/watchdog mechanism would also help distinguish "thinking" from "hung."

Environment: Claude Code v2.1.76, Linux (WSL2), Opus 4.6 with 1M context.

I have exactly the same problem. Is there any debug information you could need?
Environment: Claude Code v2.1.81, Linux, Opus 4.6