Describe the bug

After a transient network timeout connecting to api.telegram.org, the Telegram polling worker exits silently (exit code not logged, no error output). The health monitor detects it as "stopped" and keeps restarting every ~10-15 minutes, but each new polling process starts and exits immediately without producing any error log or consuming updates. The only recovery is a full gateway restart (systemctl restart openclaw-gateway.service).

The network itself was fine during the crash loop — subsequent curl and ping to api.telegram.org returned 200 and 0% packet loss.

Logs

Initial failure (19:51-19:52)

[Jun 15 19:51:37] [telegram] transport attempt marked temporarily unhealthy for 40000ms (codes=UND_ERR_CONNECT_TIMEOUT)
[Jun 15 19:52:07] [telegram] [default] channel exited: telegram transport attempt temporarily unhealthy; retry after 9972ms | Telegram ingress worker exited with code 1
[Jun 15 19:52:07] [telegram] [default] auto-restart attempt 1/10 in 5s
[Jun 15 19:52:28] [fetch-timeout] fetch timeout after 3509ms (elapsed 3509ms) operation=fetchWithTimeout url=https://api.telegram.org/botxxx/getMe
[Jun 15 19:52:28] [telegram] [default] starting provider
[Jun 15 19:52:29] [telegram] deleteWebhook failed: Network request for 'deleteWebhook' failed!
[Jun 15 19:52:29] [telegram] deleteWebhook failed with a recoverable network error; continuing to polling so getUpdates can confirm webhook state

Crash loop begins (19:57 onwards)

[Jun 15 19:57:03] [health-monitor] [telegram:default] health-monitor: restarting (reason: disconnected)
[Jun 15 19:57:08] [telegram] [default] released stopped Telegram polling lease
[Jun 15 19:57:56] [telegram] [diag] isolated polling ingress started spool=...
[Jun 15 20:12:03] [health-monitor] [telegram:default] health-monitor: restarting (reason: stopped)
[Jun 15 20:22:03] [health-monitor] [telegram:default] health-monitor: restarting (reason: stopped)
[Jun 15 20:37:03] [health-monitor] [telegram:default] health-monitor: restarting (reason: stopped)
... (continues every 10-15 minutes until midnight)

After gateway restart (fixed)

[Jun 16 00:51:49] [telegram] [default] starting provider (@yachiyo1680_bot)
[Jun 16 00:51:50] [telegram] [diag] isolated polling ingress started spool=...
[Jun 16 00:51:50] [telegram] Inbound message telegram:xxx -> @yachiyo1680_bot (direct, 26 chars)

Expected behavior

1. The Telegram polling worker should recover from transient network failures (e.g., exponential backoff with reconnection)
2. If recovery fails, the crash/exit should be logged with a clear error
3. The health monitor restart should actually resolve the issue, OR the system should escalate (e.g., auto-graceful-restart after N consecutive failures)
4. At minimum, a polling failure should produce an error message visible to the user so they know their bot is down

Actual behavior

• Polling process exits silently with no error output
• Health monitor restarts every ~10-15 minutes but the new process immediately exits again
• No error is surfaced to the user via any channel
• Only a full gateway restart resolves the state

Environment

• OpenClaw version: 2026.6.1
• Platform: Raspberry Pi 4B (arm64, Linux 6.12.75+rpt-rpi-2712)
• Node: v22.22.2
• Gateway: local, systemd user service
• Telegram provider: polling mode (no webhook URL configured; webhook URL empty in getWebhookInfo response)
• Network: Behind NAT, standard home broadband (China)