Deciding approvals
When an agent reaches a Human approval step, the run pauses and a pending item appears for whoever has the required role.
What you see
| Element | Content |
|---|---|
| Title | What is being asked |
| Summary | The context whoever built the agent provided |
| Origin | Which agent and which run |
| Deadline | When it expires, if there is one |
If you need more context, open the run: the timeline shows everything that happened before that step.
Deciding
| Decision | What happens |
|---|---|
| Approve | The run continues to the following steps |
| Reject | The run ends as not approved |
You can record a comment. It stays in the step's output and can be used by the following steps — for example, to include the rationale in a report.
There is no alternative path for "rejected". If the flow needs to handle both outcomes differently, the correct design is to decide before the approval, with a condition.
Who can decide
Anyone with the role required by the step, within the agent's space. The step can further restrict it to specific people.
The decision is single
The first decision stands. A second attempt on the same pending item is refused with a notice that it has already been resolved — two people do not decide the same case in opposite directions.
Deadline
If the step declares a deadline and it runs out, the run ends as expired. With no deadline, the pending item waits indefinitely — without consuming resources.
What is recorded
Who decided, when, and the comment. The information lives in the step's output and in the run's timeline.
Next steps
Was this page helpful?
Report a problem on this pageDo not send passwords, keys, tokens, or customer data.