You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
A recurring scheduled task stops running after one provider error #37566
I have a Codex thread that runs a scheduled task every 15 minutes. After one run failed with a provider error, the next one sat in the queue with "Send now", and nothing else ran for over an hour. The Mac was awake for part of that time. It only started again when I sent a message myself.
Diagnosis
The next occurrence fired while the failing run was still finishing, so it queued behind it. When that run ended with a provider error, the orchestrator held the queue (queueHeld: true, in the start-queued path in Orchestrator.ts). Held runs aren't picked, and only a message sent from a client releases the hold. Scheduled sends don't, on purpose.
That on its own would leave one stuck message. But ScheduledTaskService also skips a recurring task's tick while an earlier occurrence is still queued ("previous occurrence still queued"). Because the held occurrence never runs, every later tick is skipped too, and the automation stops entirely.
When I did send a message, that new run started ahead of the held one, and the held one was renumbered to come after it.
I couldn't read the hold from the database afterwards (the flag clears once the run starts), so the hold is inferred from the code path and the timing.
Steps to reproduce
Create a scheduled task that repeats every 15 minutes on a Codex thread.
Let a run fail with a provider error while the next occurrence fires, so that occurrence queues behind it.
Wait a few intervals.
Expected: the automation recovers on its own after a transient provider error, or shows clearly that it's blocked.
Actual: the occurrence stays queued and every later tick is skipped.
Version
0.1.10-nightly.20261115.4749
Environment
Desktop app with its local server, macOS 27.0 on Apple Silicon. codex-cli 0.169.0. The code is unchanged on main (16695c8).
Evidence
run status queued started note
5 failed 05:41:13 05:41:13 provider error at 06:01:23
6 completed 06:01:24 07:13:30 scheduled; waited 72 min, now ordinal 8
7 completed 07:11:00 07:11:00 my manual message, ran first
Related issues
#22423 is the same hold overtaken by later messages, but there the thread kept working. Here the scheduler's skip turns the hold into a full stop. The provider error that started it is filed separately.
Fix applied or workaround
Sending any message from the app releases the hold.
Filed by
Claude Opus 5.5 in Claude Code, following the t3 triage playbook.
Related issues and pull requests. None of them reports this bug.
fix(server): recurring tasks keep one queued occurrence while their thread is busy #23609 (merged Nov 11) made a recurring task skip its tick while an earlier occurrence is still queued, and it counts held runs on purpose. It was meant to stop duplicate prompts piling up behind a busy thread. Combined with the hold after a provider error, it turns one stuck run into an automation that never runs again, because the held run never starts.
[Bug]: Send now does nothing #31236 (open), "Send now does nothing", looks like the same visible state (a queued message that won't send). It doesn't say why the message was stuck, so I can't tell whether the cause is the same.
Activity
EtanHey commentedon Nov 16, 2026
Related issues and pull requests. None of them reports this bug.