Debugging an OpenClaw agent that won't respond: 5 things to check first

If your OpenClaw agent has gone quiet and you're staring at a chat that won't reply, the cause is almost always one of five things. Run through these in order before you tear the install down.

Step 1. Is the daemon actually running?

The agent stops the moment the daemon dies. Check the process is alive. If you ran `onboard --install-daemon` twice in the past, you might have two competing instances. Kill them both and start one cleanly.

Step 2. Is the model endpoint reachable?

If you're routing through OpenRouter, log into the OpenRouter dashboard and check the Activity tab. No requests showing means the daemon isn't getting through. Usually that's an expired token or a typo'd token-provider value. Re-run the onboard command with a fresh key if you're not sure.

Step 3. Has the channel webhook been claimed twice?

Telegram, Slack and Discord will silently swallow messages if a webhook is set to a different URL. In Telegram, `getWebhookInfo` against the bot token tells you exactly what URL is registered. If it's wrong, `setWebhook` to the right one (or `deleteWebhook` and let OpenClaw register itself).

Step 4. Did the skill actually load?

A malformed SKILL.md (bad YAML frontmatter, wrong indentation) fails silently, and the agent falls back to "no relevant skill" behaviour. Check the daemon logs. A parse error will be sitting there if that's the cause.

Step 5. Are you out of credits or rate-limited?

Free models hit usage caps mid-conversation. Paid endpoints fail when the balance is empty. Check the provider dashboard before assuming it's an OpenClaw bug.

Nine times out of ten it's #1 or #3. Drop the symptom in the comments if you're stuck and the community will help you narrow it down.

What did you have to check the last time your agent went quiet?

2 comments