When a Swarmbot AI script meets a sign-in or permission prompt, stop and hand it back to the user
Tell the Swarmbot script assistant which sign-in, permission, and unexpected screens should interrupt a workflow. Have it report the current screen before a person decides what to do next.
Swarmbot AI can help turn repeated phone actions into a script. A sign-in screen, permission request, or unexpected dialog changes the screen that the next action would touch. This guide is for people writing app workflows with Swarmbot's script assistant. Open your project in Swarmbot's script development view, choose one authorized test phone, and describe both the expected starting screen and the screens that should interrupt the task.
Name the screens where the script must stop
Suppose the task is to open an asset list and find a particular card. The normal path starts searching when the list appears. If the app instead opens a sign-in screen or Android asks for a permission, tapping where the card used to be is no longer the right action. Tell AI what the actual screens on your app look like. When a screen cannot be identified confidently, ask it to stop and show you the uncertainty.
You can adapt this request:
The test phone is open to the sample app's asset page. Using the current Swarmbot script documentation, first write a small step that only checks the screen. If it confirms the asset list, report “ready to continue.” If it finds a sign-in screen, Android permission request, or another dialog covering the list, report “needs a person” and stop this part. State the screen clue and the last confirmed page. Do not enter an account, choose a permission, or tap the list's old position. If the screen is unclear, stop and tell me what information is missing.
The key is to treat an interruption as a distinct result. It should not become another long wait or an unverified tap. Android's permission request guide describes a system decision shown to the user. Your Swarmbot script should hand that decision back to the appropriate person.
Give AI just enough visual context
If the assistant cannot tell the list from a sign-in page, provide a cropped screenshot that keeps the page title, nearby button labels, and dialog boundary. Hide account names, avatars, messages, one-time codes, and device identifiers. A password or full private screen is not needed to explain the interruption. Point out which area is the normal list and which is the new prompt.
Ask AI to check the screen before the next action and return one of three outcomes: expected page, needs a person, or cannot determine the page. The third outcome should also stop. Have the assistant check Swarmbot's current page-reading or text-finding function documentation instead of borrowing function names from a different automation product.
Resume from a page you have checked
On one test phone, run only the screen-checking step first and compare its report with the picture shown in Swarmbot. When sign-in or permission appears, let an authorized person handle it on the device. Then check the app's actual page again before running the next small part. Saying “I handled it” does not establish that the app returned to the old screen.
If the workflow may already have sent, submitted, or deleted something before it stopped, check the real result before trying again. Give AI the last confirmed successful step, a redacted view of the current screen, and the original error. The draft review guide helps identify actions that change data; the state-based wait guide covers ordinary loading delays. This guide is about handing an interrupted Swarmbot script back to the person who can decide what happens next.