Summary

An unhandled TypeError: fetch failed rejection kills the gateway process. With KeepAlive: true in the LaunchAgent, launchd restarts it — but if the underlying network issue persists, it crash-loops until launchd throttles and gives up. The gateway then stays dead silently until manually restarted.

Steps to Reproduce

1. Gateway running normally via LaunchAgent with KeepAlive: true
2. Network blip or transient API failure (Anthropic, Telegram, Brave search, etc.)
3. Gateway crashes with unhandled rejection:

2026-02-11T14:39:26.618Z [clawdbot] Unhandled promise rejection: TypeError: fetch failed
    at node:internal/deps/undici/undici:13484:13

4. launchd restarts → crashes again → restarts → crash loop
5. After ~50 runs, launchd throttles and stops restarting
6. Gateway stays dead — Telegram bot (and all cron jobs) go offline silently

Expected Behavior

• Network fetch failures should be caught and retried with backoff, not crash the process
• The gateway should be resilient to transient network issues

Actual Behavior

• Unhandled promise rejection kills the process
• last exit code = 1, runs = 50 in launchctl print
• Gateway offline for ~24 hours until manual clawdbot gateway start

Environment

• clawdbot v2026.1.24-3
• macOS (Darwin 25.2.0)
• Node v22.12.0
• Channel: Telegram (@Ghostinasshellbot)

Suggested Fix

Add a global process.on('unhandledRejection') handler that logs + recovers instead of crashing. Specifically, fetch calls to external APIs (Anthropic, Telegram Bot API, Brave Search) should have try/catch with retry logic.