Your agents act. You set the boundary.

Agent approval design checklist

Use this checklist to specify one workflow before adding a control. The worksheet is a planning aid. Completing it does not establish that a control is deployed or effective.

Free text template. No email required.

1. Define the exact action

   Name the operation, target, environment and parameters the reviewer must see. For a refund, include the order, amount and currency.

2. Choose the decision owner

   Identify the designated approver, how you authenticate them and how you check their authority. Decide who can delegate or escalate.

3. Set the policy

   Separate routine actions, actions requiring human approval and prohibited actions. Check what your existing framework already enforces.

4. Locate enforcement

   Identify the check before the protected effect. Review every credential and execution path that could bypass it.

5. Handle changed or expired requests

   Compare the final parameters with the approved proposal. Define which changes require a new decision, the expiry time and rejection behavior.

6. Plan retries and recovery

   Store pending state durably. Define how duplicate or concurrent decisions are consumed and how downstream actions avoid unintended repetition.

7. Retain distinct evidence

   Link the request, approver decision and execution result. Record signature verification details where used, and what the records cannot prove.

8. Test the failure paths

   Use synthetic actions to test rejection, changed parameters, expiry, duplicate retries, restart, concurrent decisions and the wrong approver. Record actual outcomes.

Start with the authorization foundation and the human-approval implementation guide. When you have a concrete control gap, share your workflow. Early access registers interest; availability must be confirmed for your integration.

Worksheet v1. Updated October 20, 2026.