openclaw - 💡(How to fix) Fix Long-running exec processes killed on session heartbeat timeout

Problem

When an agent starts a long-running exec process (e.g., data collection taking 15-30 minutes), the process is terminated with SIGTERM when the parent session's heartbeat times out. This happens even when the process output is still being produced.

Steps to Reproduce

1. Start a long-running command via exec (e.g., a Python data pipeline that takes 20+ minutes)
2. Wait for heartbeat timeout on the session
3. Process receives SIGTERM and dies

Expected Behavior

Long-running exec processes should survive session heartbeat timeouts, or at minimum there should be an opt-in mechanism (e.g., detach: true option on exec) to decouple the process lifecycle from the session.

Current Workaround

Using nohup + pid file + log file pattern, but this is fragile and loses the nice integration with process tool (poll/log/write).

Environment

• OpenClaw: 2026.5.3
• OS: WSL2 Ubuntu
• Channel: webchat

Impact

Any workflow involving long API-bound data collection or model training is unreliable. The process must be restarted multiple times before it can complete within a heartbeat window.