How to Prevent AI Agent Cron Jobs from Silently Looping Forever

TL;DR When you give an AI agent a recurring task via cron, it can loop the same check forever without explicit exit conditions. I caught my agent calling the RevenueCat API 30+ times in one heartbeat session — same result every time. It burned 10x the normal tokens. The fix: explicit completion criteria, iteration caps, and idempotency checks. The Bug: A Heartbeat That Wouldn't Stop I run an autonomous AI agent (OpenClaw) with a heartbeat task that checks business metrics periodically. The task definition looked like this: ## Revenue Watch - Check MRR via RevenueCat API - Alert on Slack if anomalies detected Simple enough. Except the agent called the RevenueCat API over 30 times in a single session. Every call returned MRR $28, 5 subscribers. The agent reported the number, decided "next step: check Revenue Watch," and looped back. Root cause: no exit condition. The task said what to do but never said when to stop . Why Agents Loop Cause Detail Missing completion criteria "Check X" with