Subagent processes not cleaned up after session ends

[jdongwoo]

Problem

After using claude-mem for a while, I noticed my machine slowing down significantly. Checked ps aux and found 14+ orphaned claude subagent processes still running in the background, some dating back 3 days.

All of them were children of the worker-service.cjs --daemon process (bun), each consuming 300-600MB of memory.

PID 1172  - started Monday (3 days ago)
PID 51407 - started Tuesday
PID 51593 - started Tuesday
...total 14 orphaned processes

Root cause (what I found)

The Stop hook flow queues /api/sessions/summarize but doesn't call session complete or delete. So from the daemon's perspective, the session stays "active" and the spawned claude subprocess just sits there waiting for more queue input forever.

The internal reaper also skips these because they're tied to "active" sessions.

Steps to reproduce

1. Use claude-mem normally across multiple sessions
2. Close terminal / end sessions
3. After a day or two, run ps aux | grep claude
4. Observe accumulated background claude processes under the worker-service daemon

Workaround

I set up a SessionEnd hook in ~/.claude/settings.json that kills background claude subagent processes older than 5 minutes on session end. Works fine as a band-aid but the daemon itself should handle cleanup.

Suggestion

• Call session complete/delete after summarize finishes in the Stop flow
• Add an idle TTL so stale sessions get cleaned up automatically
• Reconcile orphaned child processes on daemon startup

Environment

• claude-mem 9.0.12
• macOS (Apple Silicon)
• Claude Code CLI

[github-actions]

A GitHub user reported a significant performance issue stemming from orphaned Claude subagent processes not being cleaned up after sessions end in the claude-mem tool. After prolonged use, the user observed that their machine slowed down and discovered over 14 orphaned subagent processes, each consuming 300-600MB of memory, still running days later as child processes of the worker-service.cjs --daemon. The root cause was identified as an incomplete Stop hook flow that fails to mark sessions as complete, leaving subprocesses in an idle state indefinitely. Steps to reproduce the issue and a temporary workaround were provided, where a custom SessionEnd hook kills stale processes. The user suggests calling session complete/delete after session summarization, adding a time-to-live (TTL) for idle sessions, and reconciling orphaned processes during daemon startup to address the issue properly.