#My OpenClaw always hung in the midway
1 messages · Page 1 of 1 (latest)
I’m checking the docs/code path for this, because “it just pauses until I poke it” usually has a few specific causes.
What you added makes sense, but my first guess is this:
the main session lane is getting stuck, so heartbeat never gets a chance to “rescue” it.
A few important gotchas here:
AGENTS.md/HEARTBEAT.mdonly affect behavior once a turn is actually runningdiagnostics.stuckSessionWarnMs/ session logging help detect hangs, they do not unstick them- heartbeat is main-session, so if that lane is already busy/stuck, it can get skipped as
requests-in-flight
That would match “it only woke up when I asked again”.
Please check these first and paste the output:
openclaw --version
openclaw status --all
openclaw system heartbeat last
openclaw tasks list
openclaw tasks audit
openclaw logs --follow
What I’d look for in logs:
heartbeat skippedwithreason=requests-in-flightstuck session: ... processingNo output for ... It may be waiting for input.
A few concrete tips:
- If this is detached work (ACP/subagent), inspect tasks, not heartbeat
openclaw tasks list
openclaw tasks show <taskId-or-runId>
openclaw tasks notify <taskId-or-runId> state_changes
By default, ACP/subagent tasks are often done_only, so mid-run stalls can feel “silent”.
-
Do not rely on heartbeat to recover a wedged main turn
Heartbeat can be skipped when the session is already busy. For watchdog/recovery, isolated cron or detached tasks are usually better than “please resume hung work” inHEARTBEAT.md. -
Make sure heartbeat is not skipping for a boring reason
Common ones are:quiet-hoursempty-heartbeat-fileno-tasks-duealerts-disabled