Build apps
Tests
Watch recordings of the browser tests the agent writes for each requirement.
Watch recordings of the browser tests the agent writes for each requirement.
While it builds what you ask for, the agent writes a browser test for each of your requirements: a script that opens the app, clicks through it the way a person would, and checks that the result is right. It runs them before it finishes, and fixes what broke. Each run is recorded, so you can watch exactly what was checked.
If some requirements don’t have tests yet (for example, ones written before OneDrop added tests), the Tests tab says so. Click Write them with agent to ask the agent in the chat to write and run them; it’s queued if the agent is busy.
Open the Tests tab from the + button next to Preview, or click the badge next to a requirement in the Requirements tab to go straight to its tests.
Tests are grouped by the requirement they check, with each requirement’s name. Each shows whether it passed or failed on its last run, or hasn’t run yet, and how long it took. Tests not linked to a requirement are listed at the end.
Pick a test to:
In the Requirements tab, each requirement with tests shows a badge: Test passes, Test failing (or how many of its tests fail), or how many haven’t run yet.
| Button | Runs |
|---|---|
| Run all | Every test |
| ▶ next to a requirement | That requirement’s tests |
| Run, on a test | Only that test |
Tests run in the sandbox, one at a time, against the app the preview shows. Only one run goes at a time. The tab updates when it finishes. Only the latest recording of each test is kept.
Click Open test runner to open Playwright’s test runner in its own, larger window. There you can:
Only people who can change the project can open the runner. It stops by itself after 30 minutes without use; click Open test runner again to start it. Runs in the runner don’t change the results shown in the Tests tab.
On a laptop install, the test runner opens only on the machine running OneDrop, like the preview.
Sometimes you want to get the app to a certain point (signed in, three things in the cart, on the checkout page), then look around, try things, and ask the agent to change what you see. A test that already does those steps can take you there.
In the Tests tab, pick a test. Its Steps list every action and check from its last run. Hover the step you want to stop after and click Take over.
The test runs up to and including that step, then stops, and its page opens in a Browser tab exactly as the test left it, with any form input or open dialog. Click, type, scroll and paste as usual; Bigger window opens the same page in its own window.
Chat as usual (“make this button bigger”, “why is this total wrong?”). The agent is told which test and step your page came from, and can look at it. Its changes appear on the page as it works, without losing what’s on it.
The page uses the app’s own dev server and database, like the preview: whatever you do there really happens. Close the Browser tab to close it; it also closes after 30 minutes unwatched. Only people who can change the project can take over a test.
Tests come with requirements tracking: the switch at the top of the Requirements tab, or in Tools → Agent Skills, turns both on or off. With it off, the agent’s turns are shorter, but it doesn’t write or run tests. Tests it already wrote stay, and you can still run them from the tab.
The tests are Playwright Test files in tests/e2e, committed with the app. Each is tagged with the requirement it checks:
import { expect, test } from '@playwright/test';
test('User should see an error when their email isn\'t a company one', { tag: '@REQ-001' }, async ({ page }) => {
await page.goto('/login');
await page.getByLabel('Email').fill('someone@gmail.com'
@playwright/test as a dev dependency (the agent adds version 1.63.0). You don’t need to install a browser: the sandbox has Chromium.$PORT, one worker, with a video and trace of each test. Use relative URLs in page.goto./opt/onedrop/run-tests runs them all; pass a file, file:line, or a tag such as @REQ-003 to run some. Runs from the Shell show in the tab too..onedrop/tests, which is never committed.