Using Space Automate
Reviews: human approval
Where decisions waiting for a person collect, how to approve or decline, and what happens next.
Whenever work needs a person's decision, it becomes a review request and appears under Reviews in the sidebar. This is the heart of Space's human-gated approach: agents prepare, people decide.
What creates a review
- A Review step in a playbook.
- An agent about to do something you marked “check with me before”.
- An agent task that needs approval before it starts, such as a Front Desk request from a colleague.
- A tool, MCP server or skill marked Requires approval.
The Reviews page
Three views keep things organized:
| Tab | What's in it |
|---|---|
| Waiting for me | Requests assigned to you that you haven't decided |
| Assigned to me | Everything addressed to you, including decided ones |
| Mine | Requests you started that are waiting on someone else |
You also get a notification, and an email if you've turned that on.
Deciding
- Open the request. You'll see what's being asked, the reviewer message, and the conversation that led to it, with secure answers masked.
- Read the context. For playbooks you can also see the values collected so far, with secure ones hidden.
- Choose Approve or Decline, and add a note. The note travels back to the requester (and an agent will learn from it).
The first decision wins. If several people are named, whoever acts first decides. Later attempts are told it's already done.
What happens next
| Decision | Result |
|---|---|
| Approve | The work resumes. The agent reports the outcome (“Approved by Bob. Note: …”) and carries on. |
| Decline | The playbook follows its “declined” path, or ends. An agent's action is refused. The reason you gave is passed on, so the agent can change course. |
The requester gets a notification of the result.
For the person waiting
- The chat shows Waiting for review with the reviewers' names. New messages in that chat are blocked until a decision, to keep things orderly.
- Click Cancel request if you've changed your mind. It cancels the run and the review.
Scheduled automations
A playbook started by a schedule or trigger simply pauses (“waiting for approval”) and finishes after the decision. Its output goes to the automation's own chat.
Testing a review step
When you test a playbook in the editor, nobody is notified. The Test panel shows Approve and Decline buttons so you can play the reviewer yourself. The Check mock test tries both outcomes.
Your privacy as a reviewer
Opening, deciding on and cancelling a review are written to the audit log. Only the requester and the named reviewers can see a request.