Agent/subagent freezes indefinitely on Opus 4.6 — no timeout, no error, no recovery

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?

Environment

• Claude Code: latest version
• Model: Opus 4.6 (1M context)
• Plan: Max
• OS: Windows 11 Pro
• Connection: Stable WiFi (no drops observed)
• MCP servers active: 5 (Cloudflare, Gmail, Calendar, Chrome, Google Drive)
• Installed skills: 18 (12 community + 6 custom)

Problem

Subagents freeze mid-execution with no error, no timeout, and no recovery path.

Symptoms:

• Token counter stops completely
• Agent stops calling tools
• No error message displayed
• Session hangs indefinitely — only Ctrl+C recovers
• Happens most frequently after 2-3 successful agent runs in the same session
• Internet connection remains stable throughout (verified)

Reproduction

1. Start a session with multiple MCP servers connected and skills installed
2. Spawn agents for research tasks (WebSearch/WebFetch)
3. After 2-3 successful agent completions, subsequent agents freeze
4. Agent may complete 10-15 tool calls before freezing — it's not immediate

Observations

• Connection does NOT drop — this appears to be server-side
• Freezes increased significantly after upgrading to Opus 4.6
• Heavier system prompts (more tools/skills loaded) seem to correlate with more frequent freezes
• No way to set a working timeout — the documented 5-minute timeout does not appear to be implemented

Expected behavior

• Agents should timeout after a configurable period and return an error
• Failed agents should not block the session
• One agent's failure should not affect other agents

Related issues

• #26224 (Opus 4.6 hangs)
• #33949 (SSE streaming hangs)
• #4744 (missing timeout implementation)
• #6594 (subagent cascade failure)

What Should Happen?

Agents should either complete their task or fail with a clear error message and timeout. They should never hang silently for 10+ minutes with no feedback. Users need to know if an agent is thinking or frozen. A configurable timeout (e.g., 5 minutes) should automatically stop stuck agents and return control to the user.

Error Messages/Logs

No error messages — that's the problem. 
Agent freezes silently. No error, no timeout, no log output.
Only indication: token counter stops, "Processing..." stays on screen indefinitely.
Example: Agent completed 13 WebFetch tool calls successfully, then froze. 
Displayed "Processing..." for 11+ minutes, 298 tokens, zero activity. 
Only recovery: Ctrl+C.

Steps to Reproduce

1. Open Claude Code with multiple MCP servers connected (5+) and community skills installed (12+)
2. Start a conversation and spawn an agent for online research (WebSearch/WebFetch tasks)
3. Let the agent complete successfully (it will)
4. Spawn a second agent — usually completes fine
5. Spawn a third agent — freezes mid-execution
6. "Processing..." stays on screen, token counter stops, no tools called, no error
7. Wait 10+ minutes — nothing happens. Only Ctrl+C recovers

Claude Model

Opus

Is this a regression?

Yes, this worked in a previous version

Last Working Version

Unknown exact version. Agents worked reliably before Opus 4.6 release and before installing community skills (March 21, 2026).

Claude Code Version

2.1.63

Platform

Anthropic API

Operating System

Windows

Terminal/Shell

Windows Terminal

Additional Information

Related issues: #26224, #33949, #4744, #6594, #30122, #36469, #15945

Setup: 5 MCP servers (Cloudflare, Gmail, Calendar, Chrome, Google Drive), 18 skills, ~100 tools total in system prompt. Max plan, Opus 4.6 (1M context).

This is not a network issue — WiFi stays connected throughout. The freeze appears to be server-side: the API stops sending events while the connection remains alive.