Dark mode
Reading a run's outcome

Reading a run's outcome

When a run finishes, or while it's still going, you can open it to see exactly what the agent did. This page explains what you'll find in an open run and how to tell whether it succeeded.

Open a run

  1. Go to Automation → Activity.

  2. Select a row, or the arrow at the end of it. You can also select a run under Recent runs on a flow.

The run opens in a side panel. Rows marked Skipped, and launches that failed before a run started, have no run to open.

What's in the run panel

Area

What it shows

Header

The agent, the run's current stage (for example Agent working or Completed) and a short description of it

Tokens

How much AI usage the run consumed, when available

Transcript

The conversation between you and the agent, including the steps and tools it used

Agent plan

The agent's plan, when it made one, and whether it was finished

Changes

Files the agent added, modified or deleted, when it worked on code

Approvals and questions

Anything the agent asked you, and how it was answered

Use Refresh at the top of the panel to load the latest updates, and Cancel run to stop a run in progress.

Tell whether a run succeeded

  • Completed means the run ended normally. Read the end of the transcript and check any changes to confirm the agent did what you expected. A completed run doesn't guarantee a good result.

  • Failed means the run stopped with an error. A Run failed message shows the error at the top of the panel. If the agent produced anything useful first, it appears under Recovered output before failure.

  • Cancelled means someone stopped the run. Work done before cancelling isn't undone.

Preview runs

If a run shows Preview run · Repository changes stay local; automatic publishing is disabled, the agent's code changes weren't pushed or published. Use preview runs to check an agent's work before letting it publish.

When a run is waiting for you

If the agent paused for approval, the transcript shows Approval required and what the agent wants to do. If it asked a question, type your answer and send it. The run continues once you respond. See Agent runs.

Where else results appear

An agent often records its work on the item it ran on, for example as a task update, a pull request or a document change. Check that item too. See Agent run details on tasks.

Was this article helpful?