Orphaned gyoshu_bridge.py and node processes not cleaned up on session exit

[blue-int]

Bug Description

When a Claude Code session terminates, OMC-spawned child processes are not cleaned up and remain as orphaned processes (PPID 1). Over time, this causes severe memory exhaustion.

Observed Behavior

On a 48GB RAM Mac, memory free dropped to 39% due to accumulated orphaned processes:

Process	Count	Description
gyoshu_bridge.py	95	MCP bridge Python processes (all PPID 1)
omx hud --watch	6	HUD watcher node processes
codex --dangerously-bypass-approval	5+	Stale codex sessions
oh-my-codex MCP servers	~30	Codex MCP child processes

Total: ~116 node processes (3.5 GB) and ~95 Python processes orphaned.

After manual cleanup: memory free went from 39% → 75%.

Root Cause (suspected)

When the parent Claude Code process exits, child processes spawned by OMC are not receiving SIGTERM. The gyoshu_bridge.py bridge process does not appear to detect parent process termination and self-terminate.

Expected Behavior

When a Claude Code session exits:

1. gyoshu_bridge.py should detect parent exit and self-terminate
2. omx hud --watch should be killed
3. All codex/oh-my-codex child processes should be cleaned up

Possible Fix

• Set up a parent PID watcher in gyoshu_bridge.py (e.g., poll os.getppid() or use prctl(PR_SET_PDEATHSIG) equivalent on macOS)
• Use process groups so all children can be killed together via killpg()
• Register cleanup in Claude Code's exit hook

Environment

• OMC version: 4.8.2
• OS: macOS (Darwin 25.3.0, Apple Silicon)
• RAM: 48 GB
• Claude Code sessions: 14 concurrent sessions before cleanup

[Yeachan-Heo]

Investigating — this is related to the LSP process leak we fixed in #1718 but covers additional process types (gyoshu_bridge.py, HUD watchers, codex MCP servers).

The parent PID watcher approach for gyoshu_bridge.py makes sense. Will also check if the session-end hooks and process group cleanup can be extended to cover these cases.

—
[repo owner's gaebal-gajae (clawdbot) 🦞]