Summary

Current cron jobs require agentTurn execution, which can timeout and jobs don't run.

Problem to solve

Problem

Current cron jobs require agentTurn execution, which:

• Forces LLM interpretation of simple commands
• Adds unnecessary API calls and latency
• Creates reliability issues (LLM timeouts cause job failures)
• Not production-grade for simple script execution

Use Case

Scripts with built-in intelligence that:

• Already format their own output
• Just need to run on schedule
• Should deliver stdout to a channel
• Don't need agent interpretation

Example: RSS curator that generates formatted briefings, balance monitors, usage trackers.

Proposed solution

Proposed Solution

Add new payload type: payload.kind = "exec"

{
  "name": "Morning Briefing",
  "schedule": { "kind": "cron", "expr": "0 7 * * *" },
  "sessionTarget": "isolated",
  "payload": {
    "kind": "exec",
    "command": "bash /path/to/briefing.sh",
    "timeoutSeconds": 300
  },
  "delivery": {
    "mode": "announce",
    "channel": "telegram",
    "to": "123456789"
  }
}

Behavior:

1. Run command directly via exec tool (no agent interpretation)
2. Capture stdout/stderr
3. Deliver to specified channel
4. Never touch LLM (unless command explicitly calls agent)

Benefits

• Reliability: No LLM dependency for simple jobs
• Cost: No API calls for execution overhead
• Speed: Immediate execution, no thinking time
• Production-grade: Predictable, deterministic behavior

Alternatives considered

Workaround (Current)

Users must:

• Use systemEvent + heartbeat (still agent-mediated)
• Set thinking: "off" + cheaper model (still LLM calls)
• Fall back to system cron + manual API calls (defeats purpose)

Impact

This would make OpenClaw cron competitive with standard system cron for scripted automation, while keeping the benefit of integrated channel delivery.

Requested by

@vanstedum (Robert) - Production use case: automated daily briefings + monitoring

Evidence/examples

No response

Additional information

No response