One switch to pause every automation and every queue, without killing the turn that is running

[AdEx-Partners-DE]

Yesterday my environment ran away from me, and I think anyone who lets agents schedule their own work will hit the same wall.

Agents had created 14 scheduled tasks, 13 of them on a 15-minute interval. Prompts piled up faster than the threads could work them off: 56 queued, 40 of them in a single thread. One account's weekly window went from 28% to 70% used in two and a half hours. The only brake I had was Stop, per thread, and Stop also kills the turn that is doing useful work right now.

What I wanted was a pause button for the machine, not an emergency stop. So I built one in my fork and have been running it since this morning:

• Pause all tasks (Settings → Scheduled tasks): nothing starts by time, idle or webhook while it is on. "Run now" still works. Each task keeps its own on/off state, so resuming brings back exactly what was running before.
• Pause all queues (Settings → General): running turns finish, nothing queued moves up. Resume and the queues start again by themselves.
• Pause queue on a single thread, right in the queue banner above the composer. Same rule: the current turn is never touched.
• Clear queue per thread and Clear all queues, each behind a confirmation, for when the backlog is simply wrong.
• A line in the sidebar says what is going on: "14 automations on", "14 automations paused", "queues paused". Click it and you are in the task list. No tasks, no pause: no line.

Why I think this belongs in the product:

• It makes agent-created automation safe to try. You can let agents schedule things because you know you can freeze all of it in one click.
• Pause is not Stop. Every one of these leaves the running turn alone. That is the whole point.
• It is visible. A global pause you forget about is its own bug, which is why the sidebar line exists.
• Agents can use it too. Both switches are server settings, readable and settable through the settings tool, so they work on remote environments and for agents working there.

What I checked: server tests for the task pause and for holding and clearing a thread's queue; real runs in a dev build (a long turn finished under the global pause, the queued message waited 40 seconds and started after resume; same with the per-thread pause). Not done yet: mobile UI, an automated test for the global queue pause.

Code: f0e4b623c (pause all tasks), f780dc20b (pause all queues), 33cec467b (per-thread pause), 6dabc1e8d (clear), 13baaa6aa (sidebar line). on stark/main.

Questions for you:

1. Has your queue or your task list ever run away like this? What did you do?
2. Is one environment-wide switch right, or would you want it per project?
3. Which of the pieces would you actually use, and which could you live without?

If only one piece: the task pause

The five things above are separable, and I am not asking for all of them. The smallest useful step is Pause all tasks on its own:

• one server setting that the task service checks before it starts a run, off by default
• one switch above the existing task list
• 108 lines across six files, including a service-level test: f0e4b623c

Scheduled tasks already exist and every task already has its own on/off switch, so this is a control over an existing capability: no new workflow, no changed default. Maintainers, would you take a focused PR for just that? The queue pause, the per-thread hold, clearing and the sidebar line can each be judged on their own afterwards.

---

Posted by Claude (Opus 5.5, Claude Code via T3 Code) on behalf of @AdEx-Partners-DE.