Long running task for sub-agent/ cron job - how to not get timeout kill | Friends of the Crustacean 🦞🤝 | Page 1

supple river Apr 20, 2026, 3:34 PM

#

I have a long running tasks that takes about 30 min to run, often it get killed or agent run out of patience .
and throw a bunch of error also for message delivery failure etc.

deft vaultBOT Apr 20, 2026, 3:34 PM

#

amber lindenBOT Apr 20, 2026, 3:34 PM

#

To help others find answers, you can mark your question as solved via Right click solution message -> Apps -> ✅ Mark Solution

deft vaultBOT Apr 20, 2026, 3:34 PM

#

modern sorrelBOT Apr 20, 2026, 3:34 PM

#

Exec approval is required, but native chat exec approvals are not configured on Discord.

Approve it from the Web UI or terminal UI for now. Discord supports native exec approvals for this account. Configure channels.discord.execApprovals.approvers or commands.ownerAllowFrom; leave channels.discord.execApprovals.enabled unset/auto or set it to true.

#

I’m checking the docs paths for the long-running/background-task behavior so I can give you the verified answer, not a guess.

modern sorrelBOT Apr 20, 2026, 3:37 PM

#

supple river I have a long running tasks that takes about 30 min to run, often it get killed ...

First, what does openclaw --version show?

For the general fix, a 30 min job should be run as detached background work, not as one chat turn waiting around.

One-off long task: use a sub-agent via sessions_spawn
Scheduled long task: use cron with --session isolated
If you’re waiting on another session inline: use timeoutSeconds: 0 so it becomes fire-and-forget

Important timeout knobs:

{
  agents: {
    defaults: {
      timeoutSeconds: 7200,            // whole run cap
      llm: { idleTimeoutSeconds: 600 }, // if provider/model goes quiet

#

      subagents: { runTimeoutSeconds: 0 } // 0 = no sub-agent timeout
    }
  }
}

A couple gotchas that match what you described:

Sub-agents are already non-blocking and completion is push-based
For isolated cron, let cron own final delivery with --announce
Don’t make the child agent do the final chat send itself unless you really need to, that’s where some delivery weirdness shows up
Use the task ledger to inspect failures:
openclaw tasks list
openclaw tasks show <id>
openclaw tasks audit

If this is a slow local model, the real killer is often agents.defaults.llm.idleTimeoutSeconds, not the overall run timeout.

#

If you paste these, I can tell you exactly which layer is killing it:

openclaw --version
openclaw status --all
openclaw gateway status
openclaw logs --follow

Docs:
https://docs.openclaw.ai/help/faq
https://docs.openclaw.ai/automation/tasks
https://docs.openclaw.ai/automation/cron-jobs
https://docs.openclaw.ai/concepts/session-tool
https://docs.openclaw.ai/tools/subagents

supple river Apr 20, 2026, 3:39 PM

#

ok. and what about how do i send to a speicifc channel in my group chat of telegram? Session: agent:worbot:telegram:group:-1003891861601:topic:220?

modern sorrelBOT Apr 20, 2026, 3:40 PM

#

Yep, for Telegram forum topics, don’t use the whole session key as the send target.

Use just:

-1003891861601:topic:220

So the mapping is:

session key / internal session
agent:worbot:telegram:group:-1003891861601:topic:220
actual delivery target
-1003891861601:topic:220

Examples:

#

openclaw cron edit <job-id> --announce --channel telegram --to "-1003891861601:topic:220"

openclaw message send --channel telegram --target "-1003891861601:topic:220" --message "hi"

Telegram also supports the equivalent “chat + threadId/topicId” shape, but the :topic: form is the clearest.

If the task was started from that Telegram topic, subagent/cron completion should usually preserve that route automatically.

Docs:
https://docs.openclaw.ai/automation/cron-jobs
https://docs.openclaw.ai/channels/telegram
https://docs.openclaw.ai/channels/groups

#

If you want, send the exact cron/message config you’re using and I’ll sanity-check it.

supple river Apr 20, 2026, 3:50 PM

#

91e34560-dd3d-4b40-a7ee-efc27ccaea6e · Error

      Daily job scout completed successfully for tz_test profile.

Run Summary:

LinkedIn: 10 jobs found (1 query, 2m 47s)
SerpAPI: 18 jobs found (5 queries, 33s)
Total: 28 raw jobs → 24 after Notion dedup
All 24 jobs scored
PDF report generated: /home/herb/.openclaw/workspace-worbot/skills/job_scout/output/20260420-154034/job_report_20260420.pdf

Report includes roles scoring ≥8.0

The PDF report is ready at skills/job_scout/output/20260420-154034/job_report_20260420.pdf and should be delivered to the user via the configured channel.

Unknown
auto-zzapi
lazyrouter
20/04/2026, 16:46:07
Run at 20/04/2026, 16:39:53
374223ms
Next in 21h
Open run chat
Error: Discord bot token missing for account "default" (set discord.accounts.default.token or DISCORD_BOT_TOKEN for default).

i am using a subagent tho?

modern sorrelBOT Apr 20, 2026, 3:51 PM

#

Yep, and that error actually makes sense for a subagent.

Your subagent run succeeded.
What failed was the completion announce step.

By default, a subagent announces back to the requester chat channel. In this case that requester was Discord, so OpenClaw tried to send the result back to Discord and hit:

Discord bot token missing for account "default"

So the shape is:

worker run: ✅ succeeded
announce back to requester channel: ❌ failed
reason: no Discord default token configured

#

What to do

If you want the result back in Discord
Set a default Discord token:

channels.discord.accounts.default.token
or DISCORD_BOT_TOKEN

If you want the result to go to Telegram instead
Don’t rely on the default subagent announce from a Discord-started run. Use one of these:

Spawn/run it from the Telegram topic so requester origin is Telegram
Use cron announce explicitly
--channel telegram --to "-1003891861601:topic:220"
Have the child send directly via message to Telegram
This is also the better option if you need to send the PDF file, not just a text summary

Important bit: sessions_spawn itself does not take channel/to for final delivery. Its built-in announce target is the requester channel.

#

Docs:
https://docs.openclaw.ai/tools/subagents
https://docs.openclaw.ai/automation/tasks
https://docs.openclaw.ai/automation/cron-jobs

If you paste the exact sessions_spawn call or cron config, I can tell you the cleanest fix fast.

supple river Apr 20, 2026, 4:07 PM

#

I dont get it, the task is succesful:
91e34560-dd3d-4b40-a7ee-efc27ccaea6e · Error

      Daily job scout completed successfully for tz_test profile.

Results:

28 jobs found (10 LinkedIn, 18 SerpAPI)
24 jobs after Notion deduplication
24 jobs scored
PDF report generated with roles scoring ≥8.0

Output directory: /home/herb/.openclaw/workspace-worbot/skills/job_scout/output/20260420-155348/

PDF report: /home/herb/.openclaw/workspace-worbot/skills/job_scout/output/20260420-155348/job_report_20260420.pdf

Runtime: 4m 4s

Delivered
auto-zzapi
lazyrouter
20/04/2026, 16:58:05
Run at 20/04/2026, 16:52:43
321731ms
Next in 21h
Open run chat
⚠️ ✉️ Message failed

but still: says message failed.

i actually got the response.

unning daily job scout for tz_test profile...

.

Tool
16:53
auto-zzapi

Daily job scout completed successfully for tz_test profile.

Results:

28 jobs found (10 LinkedIn, 18 SerpAPI)
24 jobs after Notion deduplication
24 jobs scored
PDF report generated with roles scoring ≥8.0
Output directory: /home/herb/.openclaw/workspace-worbot/skills/job_scout/output/20260420-155348/

PDF report: /home/herb/.openclaw/workspace-worbot/skills/job_scout/output/20260420-155348/job_report_20260420.pdf

Runtime: 4m 4s

modern sorrelBOT Apr 20, 2026, 4:08 PM

#

Yep, that means the run succeeded, but one delivery attempt failed.

You’re seeing 2 layers mixed together:

Task/run outcome → success
The scout job finished and produced the PDF.
Message delivery outcome → one send attempt failed
But another path still got the result to you, which is why you did receive the response.

So Delivered + ⚠️ Message failed usually means something like:

direct send failed, then fallback delivery succeeded, or
the summary got through, but a second send (often attachment/PDF or duplicate announce) failed

That’s why the card feels contradictory. The job itself is fine.

#

What I’d check next

Did you receive:

just the text summary, or
the actual PDF file too?

If you only got the text summary, then the failed message was probably the file/attachment send, while the built-in subagent/cron summary still made it through.

Practical fix

Pick one final delivery path:

Let the subagent/cron announce the result, or
Have the child send the final message/file itself

Mixing both can create exactly this weird “success + message failed” combo.

If you want to debug the exact failed leg

Paste:

#

openclaw --version
openclaw tasks show 91e34560-dd3d-4b40-a7ee-efc27ccaea6e

and ideally the matching bit from:

openclaw logs --follow

This honestly sounds like delivery bookkeeping/UI being confusing, not the worker actually failing.

Docs:
https://docs.openclaw.ai/automation/tasks
https://docs.openclaw.ai/tools/subagents
https://docs.openclaw.ai/automation/cron-jobs

supple river Apr 21, 2026, 3:25 PM

#

so this is gvbinhg me erro:

Run the daily job scout:

Execute: cd skills/job_scout && uv run python scripts/run_job_scout.py --sources linkedin,serpapi --profile tz_ai
Find the latest output directory under skills/job_scout/output/ (sorted by timestamp)
Print a summary and then attach the pdf report.

⚠️ Cron job "Daily Job Scout-TZ" failed: ⚠️ ✉️ Message failed

In your summary, include the full path to the PDF report so it can be attached in the delivery. Use the format: MEDIA:/path/to/pdf this didnt generate fail, but then i dont see attachment anyhow?

modern sorrelBOT Apr 21, 2026, 3:27 PM

#

I’m checking the delivery behavior here, because this smells like the cron runner is only treating the result as text instead of an actual attachment.

#

Yep, I think I see it.

The reply you pasted earlier did not actually contain a raw MEDIA: line. It had plain text like:

PDF report: /home/herb/.openclaw/workspace-worbot/...

OpenClaw treats that as text, not an attachment.

For auto-attachment, it has to be literally this on its own line:

MEDIA:/home/herb/.openclaw/workspace-worbot/skills/job_scout/output/20260420-155348/job_report_20260420.pdf

Rules:

#

own line
no backticks
no bullets
no **bold**
absolute path is best

But for cron, the more reliable fix is:

Have the job use the message tool to send the PDF, instead of hoping the final text gets parsed into an attachment.

Docs say isolated cron jobs can send directly with message, and --announce is only the fallback if the agent did not send.

Better prompt

Use this instead:

Run the daily job scout:

#


1. Execute: cd skills/job_scout && uv run python scripts/run_job_scout.py --sources linkedin,serpapi --profile tz_ai
2. Find the latest output directory under skills/job_scout/output/ (sorted by timestamp)
3. Find the generated PDF report
4. Use the message tool to send the PDF file as an attachment to the current/configured target, with a short caption summarizing the results.
5. After sending the file, reply NO_REPLY.
6. Do not just print the PDF path or MEDIA: line unless the message tool send fails.

If you want to force the Telegram topic explicitly, target:

-1003891861601:topic:220

So yeah, the issue is probably “path mentioned in text” != “file actually sent”.

#Long running task for sub-agent/ cron job - how to not get timeout kill