[BUG] Subagent processes not terminated after parent session ends, causing memory exhaustion

Preflight Checklist

• [x] I have searched existing issues and this hasn't been reported yet
• [x] This is a single bug report
• [x] I am using the latest version of Claude Code

What's Wrong?

Subagent processes spawned via the Task tool persist indefinitely after the parent Claude Code session ends. Over approximately 2 days, 43 orphaned subagent processes accumulated, consuming ~10.8 GB of RAM. This exhausted all system memory (15 GB RAM + 7.7 GB swap at 100%), causing the system to become severely unresponsive with load averages exceeding 112-173 on a 16-core machine.

The orphaned processes all had the signature:

/root/.local/bin/claude --output-format stream-json --verbose --input-format stream-json --model claude-sonnet-4-5 --resume <session-id> --disallowedTools Bash,Read,Write,Edit,Grep,Glob,WebFetch,WebSearch,Task,NotebookEdit,AskUserQuestion,TodoWrite --setting-sources --permission-mode default

Manual intervention (kill -9) was required to terminate these processes and recover the system.

What Should Happen?

Subagent processes should automatically terminate when:

• Their parent Claude Code session exits (gracefully or via interrupt)
• Their assigned task completes
• The session they belong to is closed

Error Messages/Logs

No explicit error messages. The symptom was system-wide resource exhaustion:

$ uptime
14:58:47 up 2 days, 20:32, 2 users, load average: 112.54, 173.17, 103.66

$ free -h
              total        used        free      shared  buff/cache   available
Mem:           15Gi        14Gi       446Mi        81Mi       731Mi       762Mi
Swap:         7.7Gi       7.7Gi        34Mi

$ pgrep -c claude
45

$ ps aux | grep '[c]laude' | awk '{sum+=$6} END {print "Total Claude RSS: " sum/1024 " MB"}'
Total Claude RSS: 10780.5 MB

Steps to Reproduce

1. Use Claude Code with features that spawn subagents (Task tool with Explore, Plan, or other subagent types)
2. End the session (Ctrl+C, close terminal, or /exit)
3. Start new sessions and repeat over 1-2 days
4. Check for orphaned processes: ps aux | grep 'claude.*--output-format stream-json'
5. Observe accumulated subagent processes that should have terminated

Note: Reproduction may require multiple sessions over time. The exact trigger for orphaning is unclear - it may be related to session interruption (Ctrl+C) rather than graceful exit.

Claude Model

claude-sonnet-4-5 (subagents)
claude-opus-4-5 (main session)

Is this a regression?

Unknown - first time encountering this issue after extended use with subagents.

Claude Code Version

2.1.19 (Claude Code)

Platform

Anthropic API (Direct)

Operating System

CentOS Stream 9 (Linux 5.14.0-604.el9.x86_64)

Terminal/Shell

Bash (/bin/bash), xterm-256color

Additional Information

System specs: 16-core Intel i7-10875H, 16 GB RAM, 7.7 GB swap

Recovery: Killing orphaned processes immediately recovered ~8 GB RAM and ~6.6 GB swap:

ps aux | grep '[c]laude.*--output-format stream-json' | awk '{print $2}' | xargs -r kill -9

Potential root cause: Subagent processes may not be receiving SIGTERM/SIGHUP when parent process exits, or there's no process group management ensuring children terminate with parent.