Test runs your agent on your real workspace data and shows what it would do. Nothing is sent, posted, or changed. A test doesn't count as a run of the agent. Test every new agent before you click Create agent.
Test an agent
In the left sidebar, click Agents.
Click a starter, or describe an agent and click Build agent, to open the agent settings.
Review the trigger, prompt, notifications, and visibility.
Scroll to Test this agent and click Test.
While the test runs, the button shows Testing.... To stop it, click Cancel next to the button.
Review the result.
If something is off, change the prompt or settings and click Test again.
When the result looks right, click Create agent.
What the test result shows
Which records the agent read, by type — for example, "Read 47 records from your workspace," broken down by the specific data it pulled.
Which tools the agent used to gather that information.
The draft message, labeled Draft message — not sent. For a digest-style agent, this includes:
A short Summary highlighting the most important issues, with counts.
Recommended actions — a short numbered list of what to do first.
A breakdown per record, each with the record's name and value, its owner, why it was flagged, a short plan, and a draft message you could send.
Where a live run would deliver — the real Slack channel or email recipient, not a placeholder.
A warning if only part of your pipeline was reviewed. The same limit applies to live runs.
If the test says the agent can't do something
The prompt asks for something outside what agents can do. Examples include sending a text message, updating a field that syncs from your CRM, or posting somewhere the agent has no access to. Do one of the following:
Rewrite the prompt around what the agent can do, such as reading records, notifying you, and drafting messages.
Start from a starter that already does something similar.