Severity: High

Makes the bot completely unresponsive until manual process kill.

Steps to reproduce

1. Send a message while the configured model returns 529 (overloaded)
2. Reply fails, failover also fails
3. sendChatAction ("typing...") enters infinite retry loop every ~3 seconds
4. Loop persists across SIGUSR1 in-process restarts
5. Only a full process kill (launchctl kickstart -k or bootout+bootstrap) clears it

Log evidence

[telegram] sendChatAction failed: Network request for 'sendChatAction' failed!
(repeated every 3s for 10+ minutes, 200+ times)
[telegram] Polling stall detected (no getUpdates for 94.89s); forcing restart.
[telegram] Polling runner stop timed out after 15s; forcing restart cycle.

Expected behavior

• sendChatAction should have a max retry count (e.g., 5 attempts)
• Exponential backoff between retries
• After max retries, abandon the typing indicator and move on
• SIGUSR1 restart should clear all pending sendChatAction loops

Actual behavior

• Infinite loop, no backoff, no max retries
• Blocks the entire Telegram channel
• Survives soft restarts

Impact

Bot was completely unresponsive for 11+ hours. User sent 10+ messages over that period with no reply. Only resolved by manual process kill.

Environment

• OpenClaw 2026.3.24 (cff6dc9)
• macOS 12.7.6 (x64)
• Node 24.14.0
• Telegram channel