[BUG] orphaned processes from subagents crashing computer #18405

Activity

1. JasenNelson

   added

   bugSomething isn't workingSomething isn't working

   on Jan 15, 2026
2. github-actions

   added

   has reproHas detailed reproduction stepsHas detailed reproduction steps

   platform:windowsIssue specifically occurs on WindowsIssue specifically occurs on Windows

   area:tools

   area:core

   perf:memory

   on Jan 15, 2026
3. github-actions commented on Jan 15, 2026

   

   github-actionsBot

   on Jan 15, 2026 – with GitHub Actions

   More actions

   Found 3 possible duplicate issues:

   1. Orphaned processes after terminal close cause runaway memory consumption (34GB+ per instance) #17391
   2. [BUG] v2.0.76 on Apple Silicon: AVX warning + spawns 40+ zombie node processes causing OOM #15975
   3. [BUG] Found 30+ "2.1.1" processes in my Activity Monitor, despite having 3 Terminals open. #17219

   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
4. JasenNelson commented on Jan 15, 2026

   

   JasenNelson

   on Jan 15, 2026

   Author

   More actions

   This bug is related to 3 others identified by the github-actions[bot], but has one significant difference, it's crashing my Windows 11 Pro and forcing my computer to reboot. The underlying reason of subagents using tools that get stuck in loops and don't close with subagent completion and lack of primary agent awareness or tracking (via/tasks). Literally typing /exit in Claude Code results in immediately crash of Windows and loss of anything I had open.

   The 3 other related posts provided by the bot include:
   #17391
   #15975
   #17219
5. Muhtasham commented on Jan 17, 2026

   

   Muhtasham

   on Jan 17, 2026

   More actions

   Also having same issue:

   Version & Environment

   • Claude Code version: 2.1.11
   • OS: macOS 15.3 (Darwin 25.2.0) ARM64
   • CPU: Apple M4
   • Shell: fish

   Observed Behavior

   Multiple claude processes running with high CPU, some orphaned (PPID=1):
   Total: ~420% CPU from zombie claude processes.

   Expected Behavior

   When a Claude Code session terminates, all spawned background agents/subagents should be terminated via:

   • Process group signals (SIGTERM/SIGKILL to pgrp)
   • Explicit child process tracking and cleanup

   Reproduction

   1. Use Claude Code with Task tool to spawn background agents
   2. Exit the session (Ctrl+C, close terminal, etc.)
   3. Check for orphaned claude processes: ps -Ao pid,ppid,pcpu,etime,comm | grep claude

   👍React with 👍1Reacted by Nozir
6. github-actions

   added

   staleIssue is inactiveIssue is inactive

   on Feb 12, 2026
7. 0xarkstar

   added a commit that references this issue on Feb 13, 2026

   feat(plugin): add agent-teams-pipeline plugin

   ...

   87ec51b
8. alexeyv commented on Feb 25, 2026

   

   alexeyv

   on Feb 25, 2026

   More actions

   Additional data point: subagent persists at high CPU during a live session (macOS, v2.1.53)

   Same root cause, slightly different scenario from the original report. This happened during an active session, not after /exit.

   What happened:

   1. Parent session spawned two Task tool subagents (review agents, model: haiku and model: sonnet) for parallel file review
   2. Both agents completed their work and returned results to the parent
   3. The parent session continued working normally with the results
   4. At least one subagent process persisted as an invisible zombie at 34% CPU sustained

   Discovery: Only found it hours later via ps aux. The process:

   • Was not visible in /tasks or any UI element
   • Showed as an independent Claude process (claude --dangerously-skip-permissions --plugin-dir pan-out) under the same terminal process tree
   • Had caffeinate child processes (sleep prevention timers)
   • Was working in the parent session directory (/Users/alex/src/cooking)
   • pstree showed it under the same Ghostty terminal shell, but in no visible pane/tab

   Killing the orphan killed the parent: When the user killed this unknown process, it took down the parent Claude session too -- suggesting the subagent process was still structurally linked to the parent despite appearing independent in ps.

   Key distinction from original report: This is not about cleanup after /exit. The subagent finished its work, delivered results, and the parent kept going. The process just never terminated after completing its task.

   Environment: macOS 26.3, ARM64, Claude Code 2.1.53, Bun runtime, --dangerously-skip-permissions
9. github-actions

   removed

   staleIssue is inactiveIssue is inactive

   on Feb 25, 2026
10. theQuert commented on Mar 2, 2026

    

    theQuert

    on Mar 2, 2026

    More actions

    Community workaround: cc-reaper

    I ran into the same orphan process accumulation issue on macOS — after several sessions, claude subagents and MCP servers pile up to 7+ GB of RAM (all with PPID=1, no controlling TTY).

    I built cc-reaper as a three-layer cleanup tool:

    1. Stop hook — automatically kills orphan processes when a session ends normally
    2. proc-janitor daemon — scans every 30s and cleans up orphans from crashed sessions (60s grace period)
    3. Shell commands — claude-ram (RAM usage report) and claude-cleanup (manual kill)

    It targets claude.*stream-json (subagents), MCP servers (mcp-server, context7, supabase, etc.), and chroma-mcp processes, using TTY filtering (awk '$7 == "??"') to only kill true orphans.

    One-line install: git clone https://github.com/anthropics/cc-reaper && cd cc-reaper && bash install.sh

    Not a fix for the root cause, but keeps RAM under control until the team addresses process lifecycle management properly. Hope this helps others hitting this issue.
11. abix- commented on Mar 4, 2026

    

    abix-

    on Mar 4, 2026

    Last edited by abix-

    More actions

    statusline.sh orphans on Windows

    Statusline spawns bash.exe (Git Bash) per update. On Windows, they're never killed when the parent exits. 88 orphans in ~1h43m (~1/min). Same root cause -- no signal propagation through MSYS2.

    Inline jq command still spawns bash.exe, so no user-side fix. We've disabled statusline on Windows.

    Get-CimInstance Win32_Process -Filter "Name = 'bash.exe'" |
        Where-Object { $_.CommandLine -match 'statusline\.sh' } |
        ForEach-Object {
            try { Get-Process -Id $_.ParentProcessId -EA Stop | Out-Null }
            catch { [PSCustomObject]@{ PID = $_.ProcessId; ParentPID = $_.ParentProcessId } }
        }
    
12. github-actions

    added

    staleIssue is inactiveIssue is inactive

    on Mar 18, 2026
13. github-actions commented on Apr 1, 2026

    

    github-actionsBot

    on Apr 1, 2026 – with GitHub Actions

    More actions

    Closing for now — inactive for too long. Please open a new issue if this is still relevant.
14. github-actions

    closed this as not plannedon Apr 1, 2026
15. github-actions commented on Apr 9, 2026

    

    github-actionsBot

    on Apr 9, 2026 – with GitHub Actions

    More actions

    This issue has been automatically locked since it was closed and has not had any activity for 7 days. If you're experiencing a similar issue, please file a new issue and reference this one if it's relevant.
16. github-actions

    locked as resolved and limited conversation to collaborators on Apr 9, 2026