openclaw - 💡(How to fix) Fix Isolated agent cron jobs failing: stalled before first model call [1 comments, 2 participants]

Bug Description

Multiple isolated agent cron jobs are failing with: cron: isolated agent run stalled before first model call (last phase: attempt-dispatch)

Affected Jobs

All jobs that use sessionTarget: isolated with payload.kind: agentTurn:

• portfolio-briefing (7:30 AM weekdays) — 3 consecutive errors
• cortex-daily-digest (8 AM weekdays) — 4 consecutive errors
• portfolio-daily-perf (8:15 AM) — 1 error
• launchpad-briefing (9 AM weekdays) — 1 error
• max-session-emit (daily) — 1 error

Jobs using sessionTarget: main with payload.kind: systemEvent appear unaffected.

Error Details

lastError: "cron: isolated agent run stalled before first model call (last phase: attempt-dispatch)"
lastPhase: "attempt-dispatch"
lastDurationMs: 60709  (~60s before stall)

The lastPhase: attempt-dispatch suggests the cron scheduler triggers the isolated session but the session dispatch to the LLM provider fails.

Context

• Gateway model: Switched from anthropic/claude-sonnet-4-7-2025-06-05 to minimax/MiniMax-M2.5 around May 8-10, 2026
• The launchpad-scout cron (8:30 AM) runs successfully using openai-codex/gpt-5.5 with timeoutSeconds: 0
• The failing jobs also specify openai-codex/gpt-5.5 but fail with the dispatch stall
• One suspicious difference: scout uses timeoutSeconds: 0 (no timeout) while others use 120-300s limits

Questions

1. Is attempt-dispatch the right phase to fail on? Should this be handled more gracefully?
2. Could the model switch be related — even though the jobs specify openai-codex/gpt-5.5 explicitly?
3. Any recent changes to the isolated session dispatch that might have introduced this?