Best pattern for agent long-running tasks (30min+) without session timeout?
Running a multi-agent setup (Captain coordinator + permanent agents like Scout/Bolt). When an agent kicks off a long task (e.g., scraping 3K websites for email enrichment, ~30-40 min), the agent's LLM session gets aborted mid-poll but the underlying shell process keeps running orphaned.
Current workaround: Agent writes a self-contained Python script, tests it on small batch, launches it detached (nohup), and exits. The script runs fine but there's no callback — nothing reports the results back to Discord or triggers follow-up work.
Questions:
1. Is there a recommended pattern for "fire-and-forget + callback"? E.g., can a cron job or hook monitor a PID/file and trigger an agent turn when complete?
2. Is there a way to increase the session turn timeout for specific agents so they can poll longer without getting aborted?
3. Would sessions_spawn with a high runTimeoutSeconds work better than sessions_send for these long tasks?
4. Any plans for a built-in "background task" primitive that survives session turns?
Setup: OpenClaw 2026.2.15, macOS arm64, Claude Max (Opus 4.6 for agents). Using sessions_send to permanent agent sessions. The abort seems to come from coordinator sweep cron or internal session timeout, not LLM API timeout.
Thanks!
Current workaround: Agent writes a self-contained Python script, tests it on small batch, launches it detached (nohup), and exits. The script runs fine but there's no callback — nothing reports the results back to Discord or triggers follow-up work.
Questions:
1. Is there a recommended pattern for "fire-and-forget + callback"? E.g., can a cron job or hook monitor a PID/file and trigger an agent turn when complete?
2. Is there a way to increase the session turn timeout for specific agents so they can poll longer without getting aborted?
3. Would sessions_spawn with a high runTimeoutSeconds work better than sessions_send for these long tasks?
4. Any plans for a built-in "background task" primitive that survives session turns?
Setup: OpenClaw 2026.2.15, macOS arm64, Claude Max (Opus 4.6 for agents). Using sessions_send to permanent agent sessions. The abort seems to come from coordinator sweep cron or internal session timeout, not LLM API timeout.
Thanks!
The AI that actually does things. Emails, calendar, home automation - all from your favorite chat app. New shell, ready to help. The lobster way.
134,827Members
Resources
Similar Threads
Was this page helpful?
