Run history
What each past run records, what it deliberately does not, and how to turn every failure message into a fix you can apply now.
Automate
Run history is the vertical timeline of past runs, loaded twenty at a time with infinite scroll.
What an entry shows
- The agent's name, or a mention that the agent was deleted.
- A result pill: succeeded, failed or cancelled.
- One closed line: when it ran, how long it took, how many tokens it used.
- On a failure, the reason in plain words, above that line.
What it does not show
- What a successful run produced. Nothing on the card says what it read or prepared. Look in Approvals, or at the thing itself.
- The step by step of a past run. Steps are shown only while a run is in progress, in the work queue on the Briefing screen, which refreshes every ten seconds.
- The real total. The count reads runs loaded, not runs. A project with 400 runs shows 20 on opening.
The agent filters are built from what is loaded, not from your list of agents. An agent whose runs sit further down the timeline gets no button until you scroll to them.
Turning a failure message into a fix
| What the card says | What happened | What to do |
|---|---|---|
| The run hit its ceiling and was stopped, returning nothing and charging nothing | The job did not fit in 45 seconds. | Split the instruction in two, or cut the volume it reads. |
| Not enough credits, the work is kept | The balance ran out when the bill landed. The work done was applied and the automation was paused. | Top up, then resume the automation. |
| This run cost more than its cap | The run was charged, then that one automation was suspended. Others handed to the same agent keep going. | Raise the cap in the editor, or shorten the instruction. |
| This automation asks for tools the runner cannot supply | The agent declares tools, none of them has an executor, and no skill is served to it. Nothing was attempted and nothing was charged. | Create an agent whose tools MaShop runs, or serve this one a skill. |
| Skills declared but not found | The agent quoted a method that does not exist under that identifier. | Create it on the Skills screen, or fix the name in the instruction. |
| The tool was refused, nothing was done | Either the agent's sheet forbids it, or its connector permission is off. | Open the connector's sheet and set that tool to Ask or Allow. |
| The tool is waiting for your signature | The permission sits on Ask, or the agent prepares and waits. | Go to Approvals and decide. |
| The agent changed since the run was queued | Someone saved the agent between the queue and the start. Nothing ran. | Queue it again, and make edits before queueing next time. |
| The agent was paused since the run was queued | Nothing ran. | Resume the agent, then queue it again. |
| No reason recorded for this failure | Nothing was written down. | Queue it manually and watch the work queue on the Briefing screen, which shows the steps live. |
The fourth row deserves a second look. That run is recorded as failed like any other, but it also suspends the automation. It will not be queued again tomorrow, so the timeline stops filling with the same line every morning.
Interruptions and resumes
A run holds a lease of 90 seconds, renewed every 30. If the runner dies, the lease expires and the run becomes claimable again. It is requeued once at most, for two attempts in all.
The tokens of the second attempt add to the first. The total describes what the run cost in all, not what its last try cost.
Reading the token column
Do not read it as an invoice. Some finished runs record zero tokens and zero credits. It was measured on a real run, the code now writes a warning when it happens, and the cause is not yet found. Check your credit balance rather than adding up this column.
Getting to a single automation's runs
From the Automations screen, the card offers a shortcut that opens this timeline filtered on that automation. The empty state tells you when the filter found nothing in what is loaded, and invites you back to the full list.