hermes update (Windows) stops the dashboard server but only restarts the gateway — web UI chat left dead until manual relaunch

[cfkleins]

Environment

• hermes-agent 0.18.0, checkout at 7203898
• Windows 11 Home (10.0.26200), Python 3.11.15
• Standalone web dashboard started with hermes dashboard (default port 9119), used from Chrome

Symptom

After running hermes update, the web dashboard's chat never reconnects. The update output shows the asymmetry:

→ Stopping Windows gateway process(es) before updating Hermes...
  ✓ Paused gateway profile(s): default
...
✓ Update complete!
  ✓ Restarting Windows gateway profile(s): default

The pre-update stop takes down the dashboard server along with the gateway, but the post-update resume only restarts the gateway profile. Nothing listens on 9119 afterward, so any open dashboard tab shows a dead chat connection with no hint that the server is gone. The user-visible failure is "chat not connecting" — it looks like a gateway or network problem, not a missing process.

Related: hermes dashboard --status reports false positives

While diagnosing, hermes dashboard --status claimed dashboards were running when none was listening:

3 hermes dashboard process(es) running:
    PID 16488
    PID 25324
    PID 31924

• 16488 / 25324 were the desktop app's internal serve --port 0 backend (IPC-gated, not a usable dashboard)
• 31924 no longer existed at all

So the natural post-update health check also lies, which makes the missing-dashboard state harder to spot.

Suggested fix

Track the dashboard server the same way gateway profiles are tracked, and either restart it in the post-update resume step or print an explicit "dashboard was stopped — run hermes dashboard to restore the web UI" notice. Separately, --status should verify the recorded PIDs are alive and actually bound to a listening port before reporting them as running.