hermes - 💡(How to fix) Fix Background terminal processes lose tracking after gateway restart

Problem

When the gateway restarts (config change, crash recovery, systemd restart), all terminal(background=true) processes lose their tracking. The OS process may survive (re-parented to init) but Hermes can no longer poll/log/wait/kill it via the process tool. Additionally, systemd-managed gateway restarts kill the entire cgroup, destroying all child processes.

Steps to Reproduce

1. Start a long-running background process: terminal(command="sleep 600 && echo done", background=true, notify_on_complete=true)
2. Restart the gateway: hermes gateway restart
3. Try to poll the process: process(action="poll", session_id="...") → process not found

Expected Behavior

Background processes should either:

• Survive gateway restarts and be re-attachable, OR
• Be explicitly checkpointed so the agent knows what was running and can resume

Actual Behavior

• Process tracking is in-memory in the gateway process; lost on restart
• systemd Restart=on-failure cleans the cgroup, killing all child processes
• No mechanism to recover or reconnect to orphaned background processes

Impact

Long-running tasks (ML training, data processing, batch operations) are lost on any gateway restart. Users must manually restart everything.

Related

• #11614: Gateway exits when Telegram disconnects, killing embedded cron ticker
• #25738: Dashboard chat runs should survive browser disconnects
• #29129: WhatsApp bridge drops messages on restart

Environment

• OS: Ubuntu 24.04 (WSL2)
• Hermes version: latest
• systemd user service