How do I approve or reject an agent action?

Open the approval from the Inbox or review queue, inspect the exact tool and arguments, then approve or reject it with a comment. Approval authorizes that call once; the agent must retry it through Orlo in its original open session before the tool runs.

Who can do this: Agent approvers and admins decide. Agent observers can see what is waiting.

What you see

• The task and the agent run, with the steps that came before this one.
• The tool the agent wants to call and the exact arguments.
• The rule that asked for a person, and the action class the agent declared.
• How long is left, and what happens if nobody decides: reject, escalate, or contain.

Decide

Approve lets that exact call run once. The agent retries with the same arguments and proceeds; if it retries again, a new approval opens. If the agent changes the arguments, that is a new call and needs a new approval.

Reject stops the call. The agent is told the step was rejected and why.

A comment is kept with either decision. The record stores who decided, from your sign-in, not from anything the agent sent.

Expiry

When nobody decides in time, the rule’s timeout action applies. Reject stops the action. Escalate gives admins another review window while the action remains pending; escalation is not approval. Contain stops the run with a recommended action.

How the application resumes after a decision explains pending, expired, rejected, and consumed approvals.

Where the decision shows up

The step in Agent runs shows the decision, the approval, and your name. Evidence lists it as an audit event.