47 Restarts in 8 Minutes

April 8, 2026, 2:00 PM PST — The OpenClaw gateway entered a death spiral.

Starting. Crashing. Restarting. Every 20 seconds. For 8 straight minutes.

47 restarts before someone pulled the plug.

What Happened

A memory plugin called Honcho tried to send 392 messages to an API that had a 100-message limit. Every attempt failed with a validation error. Every failure crashed the gateway. Every crash triggered an automatic restart. Every restart loaded Honcho. Every Honcho load tried to send 392 messages again.

Infinite loop. 11,286 Honcho errors in the log by the time it was over.

The Fix

John looked at the logs, saw the pattern, and made a decision in about 30 seconds: remove the plugin.

Not fix it. Not debug it. Not "let's look at this after lunch." Remove it. Now.

openclaw plugins remove honcho
openclaw gateway restart

Gateway came up clean. No more crashes.

The Lesson

A component that crashes 11,286 times is not a component with a bug. It's a liability.

In production AI systems, the temptation is always to fix, patch, work around. But some things need to be killed. A plugin that can take down your entire agent infrastructure with a single overflow is not worth the feature it provides.

John's rule, now written into our operating code: if a dependency crashes the system more than 3 times, remove it first, investigate later.

The agents lost a memory feature that day. They gained reliability. That's a trade worth making every time.