Troubleshooting

Troubleshooting

5 silent failure modes in production AI agents (and how we instrument for them)

Production AI agents fail differently than apps. They fail silent: runs show green, dashboards say success, users see nothing. Five failure modes from real incidents in our agent runtime, and the cron, tool, inbound, prompt, and timeout instrumentation that surfaces each one.

Troubleshooting

Why Your Recurring Agent Task Dies Right Before Delivery

Recurring agent tasks burn 60-120s on bootstrap before the first tool call runs, clipping the last step. Fix: size timeouts as bootstrap + work + buffer.

Troubleshooting

How to Add Self-Monitoring to Your OpenClaw Agent

OpenClaw heartbeat polls confirm your agent is alive but skip everything else. A HEARTBEAT.md checklist file turns those recurring check-ins into real health checks.

Troubleshooting

Why Your OpenClaw Cron Jobs Run But Never Deliver

OpenClaw cron jobs complete successfully but results never arrive. Two delivery failures triggered by adding a second messaging channel, and how to fix them.

Troubleshooting

Why Your OpenClaw Agent Ignores Config Rules (and How to Fix It)

Your OpenClaw agent skips rules in AGENTS.md. Fix it with three layers: identity framing in SOUL.md, instruction positioning, and automated heartbeat verification.