When an automation does not run
Why a departure passes with nothing happening, from a paused agent to an unwired tool, and what each message on the run tells you to do.
Fix a problem
An automation is a recurring order handed to an agent: instructions, a schedule, and a version frozen at the moment it is queued.
What you see is one of three things. The next departure passed and nothing moved. The card shows Needs a look. Or the run history shows a failure with one sentence on it.
Read the schedule first
MaShop queues due automations every 15 minutes, wakes the executor every minute, and passes a recovery round every 5 minutes. A departure is never exact to the second.
A cadence carries an hour and no minute, so an agent leaves at minute zero, daily or on one weekday. The time zone is a project setting, and only the owner changes it. Saving a new one recomputes every schedule of the project.
After an outage MaShop replays at most the last missed occurrence. Three silent days do not produce three runs.
The screen says automations are not available
That sentence means their table is missing from the database, so nothing can be created or run. Nothing was deleted. No setting on your side changes it, and the wait is on MaShop.
Nothing left because something was paused
- The agent is paused. The run stops and says so.
- The agent changed version after the queueing. The run stops without doing anything at all.
- The automation is paused or archived. An archived automation never leaves again, and its history is kept.
The tools are not wired
MaShop executes two tools today: reading Supabase data and saving an email draft. The tool names shown on screen describe intent, not capability.
An agent that asks for tools the executor cannot hand over is suspended in Needs a look on its first attempt. Nothing was attempted and nothing was charged, and the run says so rather than failing every morning.
Its tools and its function are chosen at creation and never after. To change them, create another agent. An agent with no tool ticked hands back a written report, which is often what you actually want today.
The work waits for your signature
A tool set to Ask files an approval and holds the call. No counter announces it anywhere in the rail, so open Approvals yourself.
Saving an email draft is the exception, and it is the tool that fills the queue today. It counts as harmless, so it runs, then files its own request. Signing that request releases nothing: the draft is already written, and MaShop has no way to send. Copy it and send it yourself.
Only the owner signs an action that touches money, unless the owner opened that right to editors. A request that cannot be read is treated as the most expensive case, so it needs the same signature.
The run stopped on a ceiling
| What the run says | What it means |
|---|---|
| It hit its 45 s ceiling and was stopped | Nothing was returned and nothing was charged. Shorten the instruction, or split it. |
| Not enough credits | The work already done is kept and applied. The automation is paused so it does not leave tomorrow. Top up, then resume it. |
| The run cost more than its cap | The cost cap is 0.30 dollar by default and 5 dollars at most. That automation alone is suspended. Others held by the same agent keep running. |
| Skills declared but not found | The agent quoted a method name you never wrote. Write it on Skills, or correct the name. |
A run that fails is billed, because the model turns were spent. A run stopped by its own ceiling is not. Every run is charged after the fact, at real cost, never reserved in advance.
A billing fault of its own is told apart from an empty balance. Only a real shortage pauses an automation.
Several of these messages in English still carry the automation's older name. Glossary and old names lists where.
What the history can and cannot tell you
A failure carries its reason in clear text on its card. When nothing was recorded, the screen says so. No screen shows the step by step of a past run.
The count reads runs loaded, not runs total: 20 per page, infinite scroll.
Editing instructions, schedule or model writes a new version. A run keeps the version that left the queue, so an edit made while it runs changes nothing for that run. Editing only the title writes no new version.
Delete an agent only if you accept losing its history
Deleting an agent deletes its run history with it, and that history is the trace of what was done in your name. Archiving an automation keeps its history.
