Bug Description

One-shot cron jobs (schedule.kind: "at" + deleteAfterRun: true + sessionTarget: "main" + payload.kind: "systemEvent") fire repeatedly when the gateway restarts after the scheduled time has passed.

The jobs get lastStatus: "skipped" (likely because the main session is busy), which does not count as a successful run, so deleteAfterRun never triggers. On each subsequent gateway restart, the overdue job is re-evaluated and re-injected, causing dozens of duplicate systemEvents.

Environment

• OpenClaw Version: 2026.2.6-3
• OS: macOS 15.7.3 (ARM64)
• Node: v22.22.0
• Channel: Telegram

Steps to Reproduce

1. Create a one-shot reminder:

{
  "schedule": { "kind": "at", "at": "2026-02-10T22:30:00.000Z" },
  "payload": { "kind": "systemEvent", "text": "⏰ Reminder: Time to workout!" },
  "sessionTarget": "main",
  "deleteAfterRun": true
}

2. Let the scheduled time pass while the main session is active (chatting with user)

3. The systemEvent fires but gets lastStatus: "skipped"

4. Restart the gateway (e.g., via config.patch or openclaw gateway restart)

5. Observe: The same reminder fires again. Repeat step 4 → fires again each time.

In our case, multiple config changes triggered ~10 gateway restarts in an evening, resulting in 20+ duplicate reminder messages flooding the session.

Expected Behavior

• Overdue at jobs with deleteAfterRun: true should be deleted (or at minimum disabled) after the scheduled time passes, regardless of whether the delivery was "skipped"
• A systemEvent injection should not be skippable — it is a fire-and-forget text injection, not an agent turn that can fail

Actual Behavior

• The job stays alive with lastStatus: "skipped" and deleteAfterRun never fires
• Each gateway restart re-injects the systemEvent, causing duplicates
• The only fix is manually removing the job via cron remove

Cron Job State After Bug

{
  "lastStatus": "skipped",
  "lastDurationMs": 0,
  "runningAtMs": 1770774567831
}

Workaround

Use sessionTarget: "isolated" + payload.kind: "agentTurn" instead of main + systemEvent for one-shot reminders. Isolated jobs complete independently and trigger deleteAfterRun correctly.

Related

• [Bug] Telegram bot stuck in infinite loop sending duplicate messages - session state persists across restarts #5806 (session state persists across restarts causing infinite loops)
• Herbert Yang's report on orphaned .tmp files in cron recovery logic