[BUG] Claude Code CLI hangs indefinitely after sending result event in stream-json mode

Bug Description

Claude Code CLI hangs indefinitely after successfully completing a task and sending the final {"type":"result","subtype":"success"} event. The process remains running with stdout open, never exiting cleanly, requiring manual SIGINT (CTRL+C) or SIGKILL to terminate.

Environment

• Platform: Linux (Ubuntu 22.04, kernel 6.8.0-100-generic)
• Claude CLI version: 2.1.38
• Node.js: v20.20.0
• Output Format: --output-format stream-json
• Flags: --verbose --dangerously-skip-permissions
• Runtime: ~18 minutes of active session

Steps to Reproduce

1. Run Claude Code CLI in headless mode with stream-json output:

``bash
claude --output-format stream-json --verbose --dangerously-skip-permissions --model claude-opus-4-5-20251101 -p "..." --append-system-prompt "..."
``

1. Let Claude work on a task that involves multiple tool uses (in this case: 82 turns, ~18 minutes)

1. Claude completes successfully:

• Outputs {"type":"result","subtype":"success","duration_ms":1082218,...,"total_cost_usd":5.573962}
• Task is functionally complete (PR marked ready, comments posted, etc.)

1. Process hangs - stdout remains open, process doesn't exit

1. Wait 5+ minutes - no activity, process still running

1. Only CTRL+C (SIGINT) or SIGKILL terminates the process

Expected Behavior

After sending {"type":"result"} event, the CLI should:

1. Close stdout stream
2. Exit with code 0

Actual Behavior

• {"type":"result","subtype":"success"} is sent
• stdout remains open
• Process hangs indefinitely
• No further output
• Exit code upon SIGINT: 130 (128 + SIGINT)

Evidence from Logs

Timeline (from session 2ab1d239-9581-4d03-a895-af10c9fcb863):

| Timestamp | Event |
|-----------|-------|
| 21:10:17.168Z | Result event sent: {"type":"result","subtype":"success",...} |
| 21:10:17.169Z | Cost captured, result summary processed |
| 5 min 26 sec | NO ACTIVITY - PROCESS HUNG |
| 21:15:43.123Z | User interrupted with CTRL+C |
| 21:15:43.127Z | Exit code: 130 |

Result Event (last output before hang):

{
  "type": "result",
  "subtype": "success",
  "is_error": false,
  "duration_ms": 1082218,
  "duration_api_ms": 940758,
  "num_turns": 82,
  "result": "I've completed the implementation...",
  "session_id": "2ab1d239-9581-4d03-a895-af10c9fcb863",
  "total_cost_usd": 5.573962
}

Related Issues

This appears to be the same or similar root cause as:

• #1920 - Missing Final Result Event in Streaming JSON Output (intermittent)
• #24478 - Claude Code CLI freezes and becomes unresponsive
• #24481 - CLI hangs indefinitely on simple queries

However, in this case:

• The result event IS sent (unlike #1920 where it's missing)
• The hang occurs AFTER the result event, not during execution
• The process appears to be in a state where it's waiting for something that never happens

Hypothesis

Based on code analysis, the issue appears to be that after sending the result event:

1. Claude Code doesn't close its stdout stream
2. The process remains in some waiting state
3. No cleanup/exit routine is triggered

This causes any consumer using for await iteration or stream.on('end') to never receive the end signal.

Impact

• Automation: Completely breaks CI/CD pipelines using Claude Code CLI
• Resources: Stuck processes accumulate, consuming memory
• Reliability: Users must manually monitor and kill processes

Workaround

The consuming application can implement a timeout after receiving the result event:

if (data.type === 'result') {
  // Process result...
  setTimeout(() => {
    process.kill(childPid, 'SIGKILL');
  }, 30000);
}

Session Information

• Session ID: 2ab1d239-9581-4d03-a895-af10c9fcb863
• Full log file: Available at https://gist.githubusercontent.com/konard/309a5a535c0afc50e6a444f7f8b30d2a/raw/dc882237021646a4aa09320ecf26b8d03c434c65/2ab1d239-9581-4d03-a895-af10c9fcb863.log
• Reporting issue: https://github.com/link-assistant/hive-mind/issues/1280

Possible Fix Areas

1. Ensure stdout is flushed and closed after writing result event
2. Call process.exit() explicitly after result event
3. Check for any pending timers/promises that prevent clean exit
4. Investigate if MCP servers are keeping the process alive