Preventing AI Test Hallucinations by Indexing data-testid and ARIA Attributes
Generating integration and end-to-end tests with AI coding agents is one of the highest-leverage developer workflows. However, agents routinely fail at one specific step: guessing DOM element selectors. An agent will write: await page . click ( " .btn-primary-submit " ); // Fails: Class name does n
Generating integration and end-to-end tests with AI coding agents is one of the highest-leverage developer workflows. However, agents routinely fail at one specific step: guessing DOM element selectors.
await page. click ( ".btn-primary-submit " ); // Fails: Class name does not exist Enter fullscreen mode Exit fullscreen mode In TokenCap, we solve this by parsing JSX AST nodes and extracting explicit testing attributes in src/frontend/testIds.js.
When tokencap make runs, frontend components include their test locator contract:
### Component: UserBillingForm (src/components/UserBillingForm.jsx) Testing Selectors: - Submit Button: `getByTestId("billing-submit-button")` (line 42) - Plan Selector: `getByRole("combobox", { name: "Billing Tier" })` (line 68) - Card Input: `getByTestId("card-element-input")` (line 95) Enter fullscreen mode Exit fullscreen mode With this context provided up front, coding agents generate robust, locator-accurate tests using modern accessibility best practices rather than brittle CSS path queries.
Explore the frontend indexing engine at tokencap.vansharora.app