TestDriver End-to-End Testing

TestDriver End-to-End Testing teaches an assistant to author, run, and debug UI tests using TestDriver and Vitest. TestDriver runs applications in a desktop sandbox, finds bugs, and builds regression tests. Its tests interact with browsers and native applications through natural language descriptions and visual element recognition.

The workflow starts by scaffolding a project:

npx testdriverai init

Initialization writes package.json, example tests, vitest.config.mjs, and a GitHub Actions workflow. Because it writes into the current directory, the command belongs in a project the user intends to modify. An API key is configured before sandbox provisioning.

Test authoring uses Vitest describe, it, and expect together with the TestDriver hooks. A test creates its TestDriver instance from the test context and provisions a browser or application. The login example opens Chrome at a login URL, finds an email field by its visible description, clicks and types into it, then finds and clicks the sign-in button. A final assertion checks that the dashboard is visible and passes that result to a Vitest expectation. This provides an explicit outcome rather than treating successful clicks as proof of a working flow.

Element descriptions should be specific enough to distinguish the intended control, such as a colored sign-in button in a particular location. Every TestDriver call is awaited. Slow elements can use a find timeout of 30000 milliseconds, as in the loading-indicator example. Electron provisioning accepts an application path, after which visible controls such as a File menu can be located and clicked.

Use the skill to write end-to-end tests for web or desktop applications, automate canvas, Electron, or native interfaces without stable selectors, check whether a page or flow works, or investigate a failing TestDriver test. The runner is Vitest, and provisioning requires network access and a TestDriver API key. Environment-specific validation and expert review still apply. The related webapp-testing skill is a Playwright-based alternative for local web applications with stable DOM selectors.

Inputs include the application URL or Electron path, the visible controls to operate, and the expected result. The output is a Vitest suite plus its execution evidence. Screenshots are captured automatically before and after commands and saved under .testdriver/screenshots with source-line identifiers for debugging.

Run the suite through vitest run directly, avoiding npx vitest and CSS-selector dependencies. Set both testTimeout and hookTimeout to 900000 in the Vitest configuration, because teardown and recording uploads can exceed the default hook timeout. If scrolling fails because an input has focus, press Escape or click the background before scrolling.

The TestDriver API key uses TD_API_KEY in the scaffold's gitignored .env file and must not be committed. Test execution takes place in an isolated remote sandbox rather than on the local machine.