On this page
Test as Agent
Choose an enabled action, expand Effective action definition, and enter matching JSON. Run action executes with your signed-in identity through the same action service used in production. Results show status, duration, identity mode, source, and risk.
Tests are real: a create action inserts a record and a delete action deletes it. Destructive and consequential tests require explicit confirmation. Changing the selected action or input clears that confirmation.
Plan action uses the existing AI Studio allowance to translate a short request into one proposed action and input. It never executes automatically. Review and run the proposed action explicitly. If the model cannot express your request using an enabled action, MiniUp reports that instead of inventing a capability. Provider failure does not prevent direct structured testing.
Recent activity
Recent activity lists the newest 25 invocations. Use Newer and Older to browse history or Refresh after external agent use. Errors use safe codes rather than raw database, provider, or credential details. Request and response bodies are not retained.
Use a controlled app and harmless Function for testing. Clean up records you create. Validate with a viewer and editor when checking role-specific workflows; testing as an owner alone does not prove another user’s permissions.