Agent Survival Check

Having run an openclaw agent for 4 months, I run the same check every morning to prevent silent failures.

The key is to catch status anomalies first, within 2 minutes.

• Check auth status, gateway health, and pending events with openclaw status --all.
• Check with openclaw cron list that scheduled jobs haven't disappeared. A past update once deleted a cron job, so I check daily.
• Check /context list in each active session to inspect loaded files, whether truncation occurred, and per-file context usage. This is also where I catch MEMORY.md growing excessively large.
• Check ~/.openclaw/workspace/memory/ to confirm both yesterday's and today's daily logs exist, to see if memory flush is working normally.
• Send test messages to Telegram, Discord, and WhatsApp to confirm a response within 30 seconds. This is especially effective at catching WhatsApp OAuth token expiration or Telegram polling stalls.

With this routine, I caught problems early 8 to 10 times over 4 months, and in each case prevented downtime before the client saw it. The conclusion is that even when an agent performs well, it can't self-monitor, so without proper health checks and alerting, this kind of manual check is the most practical approach.