Summary

Manual cron job execution via openclaw cron run <jobId> enqueues successfully but the embedded runner never starts the LLM call. Job is marked as "running" (runningAtMs set) but produces zero activity until it times out after 1200s.

Scheduled automatic execution works fine. Only manual triggers are broken.

Environment

• OS: Linux 6.8.0-100-generic (x64)
• Node: v22.22.0
• OpenClaw: 2026.3.8 (3caab92)
• Gateway: local loopback 127.0.0.1:18789

Steps to reproduce

1. Have a working cron job (ours ran successfully on schedule at 12:15 UTC today)
2. Run openclaw cron run <jobId>
3. CLI returns { enqueued: true, runId: "manual:..." }
4. runningAtMs is set in jobs.json
5. Zero LLM/agent/cron activity in gateway logs
6. Job sits "running" until timeout (1200s) with lane wait exceeded: lane=cron

What works ✅

• Scheduled automatic cron execution (ran fine at 12:15 UTC)
• cron add, list, status, remove, runs, edit
• Simple one-shot --at crons (first run after gateway restart)
• Sub-agent spawns from main session

What fails ❌

• openclaw cron run <jobId> — tested 8+ times across multiple gateway restarts
• Tested with both opus and sonnet models
• Tested with isolated and main session targets
• Tested with delivery.mode: "none" and delivery.mode: "announce"
• Tested with fresh sessions (purged all 78 cron session mappings from sessions.json)

Log behavior

# Enqueue succeeds
{ "enqueued": true, "runId": "manual:36f76eaf-...:1773069058800:1" }

# Gateway logs show NOTHING cron-related after enqueue
# Only inbound/outbound WhatsApp from other sessions
# No LLM call, no tool invocation, no error

# Eventually times out:
lane wait exceeded: lane=cron waitedMs=1199897 queueAhead=0
FailoverError: LLM request timed out

Key observations

1. A simple one-shot --at cron that just sends a WhatsApp message worked immediately after a gateway restart
2. The next manual cron run on a different job (same restart) did not work
3. This suggests the manual trigger path has a different code path than scheduled execution
4. openclaw status shows the cron session exists with 46k tokens cached, but no new activity

Possibly related

• Bug: cron run (manual trigger) times out without execution on macOS #28866 reports the same symptoms on macOS — this confirms it also affects Linux
• bug: Stale cron locks persist across gateway restarts #30096 about stale cron locks

Workaround

None found. Scheduled execution works, so time-critical crons can be scheduled with --at as a workaround, but cron run for ad-hoc testing is broken.