AI E2E Test Generator (Playwright, Cypress & Selenium)

About the AI E2E Test Generator (Playwright, Cypress & Selenium)

Describe a user journey in plain steps, as a Gherkin scenario or as a user story with acceptance criteria, and get an end-to-end test for Playwright Test (TypeScript, JavaScript or Python), Cypress, Selenium WebDriver (Java, Python or C#) or WebdriverIO: the test file, page objects if you want them, a sign-in that runs once and is reused, the framework’s config and a README with the run commands.

Paste the HTML of your pages and every locator is checked before the code is written. Each target is found in the HTML and gets locators in the order role and accessible name, label, test id (data-testid, data-test, data-cy), then text; each one is counted against the page with the matching rules of the framework that runs it, and a step is marked as checked only when its locator matches exactly one element. Every wait is a web-first assertion that waits by itself: no fixed sleeps.

The AI is optional. It turns looser wording into steps, and the code is then written from those steps, so it never invents a locator or an API. Migration mode converts existing Cypress and Selenium tests to Playwright, with a checklist of the commands that have no direct equivalent. Without a pass the result is a free preview: the step plan with its checked locators, the checks and the first lines of the test file; the whole project, copying and the downloads unlock with a Pro pass.

Common uses

• Turn the acceptance criteria of a ticket into a first end-to-end test before the feature ships.
• Give a manual tester’s steps to a developer as a runnable Playwright, Cypress, Selenium or WebdriverIO test.
• Check which locators of a page are unique before writing tests by hand, and find elements that need a test id.
• Move a Cypress or Selenium suite to Playwright one file at a time, with a list of what to finish.

How locators are chosen and checked

Each step’s target, such as the "Sign in" button or the email field, is matched to an element of the pasted HTML by its accessible name, label, placeholder, test id, id, name and text, with the kind of element the words name (button, link, field, checkbox, drop-down …). The element then gets its candidate locators in the order role and accessible name, label, test id, text, then placeholder, alt text, title, id, name, CSS and XPath, and each is counted against its page:

• Playwright’s getByRole, getByLabel and getByText match names as a case-insensitive part of the text unless exact is set; the count follows that rule, and exact: true is added when only the whole name is unique.
• CSS selectors are counted with the browser’s own querySelectorAll, and XPath with its document.evaluate.
• A locator is marked ✓ 1 match only when it matches exactly one element. When two elements match the words equally, the step is marked ambiguous and the code uses the first, with a comment; give the element a test id to make it unique.

Each framework uses the best checked locator it supports: role and label locators in Playwright (locators); data-cy, data-test and data-testid, then ids, names and cy.contains() in Cypress (selecting elements); ids, names, CSS, link text and XPath in Selenium (locator strategies); CSS, button=Sign in and XPath in WebdriverIO, whose $ must match one element (selectors).

No fixed waits

Every check is written so it waits until it holds: await expect(locator).toBeVisible() in Playwright, .should(…) in Cypress, await expect(…).toBeDisplayed() in WebdriverIO, and explicit WebDriverWait conditions in Selenium, which never mixes in an implicit wait (Selenium on waits). A line such as “wait 3 seconds” is left out with a note: write what to wait for instead, for example Wait for the "Saved" message. The checks list flags any sleep left in the code, and migration removes cy.wait(1000) and Thread.sleep().

Sign in once and reuse the session

With Sign in once and reuse the session, the steps that sign in (a password field and its submit) run once and the tests start signed in:

• Playwright Test: a setup project saves storageState to playwright/.auth/user.json, and the browser projects depend on it (authentication); in Python, a session fixture saves storage_state the same way.
• Cypress: a cy.login() command around cy.session() (cy.session).
• Selenium and WebdriverIO: the session cookies are saved after signing in and set in each later browser.

Passwords are never written into the code: they are read from the E2E_PASSWORD environment variable (in Cypress through cy.env(), environment variables), and the saved session folder is listed in .gitignore.

What the project contains

• Playwright Test: tests/*.spec.ts (or .js), page objects in pages/, playwright.config.ts with the base URL, package.json and a README. Run npm install, npx playwright install chromium and npx playwright test.
• Playwright for Python: a pytest test, pytest.ini with base_url, requirements.txt (pytest-playwright). Run pytest.
• Cypress: cypress/e2e/*.cy.js, cypress.config.js and the support files. Run npx cypress run.
• Selenium: JUnit with Maven (pom.xml, mvn test), pytest (pytest), or NUnit test files with the dotnet new nunit commands to create the project around them (dotnet test).
• WebdriverIO: test/specs/*.e2e.js, wdio.conf.js and package.json. Run npx wdio run ./wdio.conf.js.

Migrating Cypress and Selenium tests

Migration reads the test statement by statement. cy.visit, cy.get(…).click(), .type() with {enter}, .should(…) checks, cy.url() and cy.title() become Playwright locators and assertions; Selenium’s findElement, sendKeys, Select, Actions, explicit waits and JUnit, TestNG, unittest or pytest assertions become their Playwright equivalents, and driver set-up and quit() are dropped. A Python unittest.TestCase class becomes plain pytest functions that get the page, and its setUp becomes an autouse fixture (or goes, when it only started the driver).

Network stubbing (cy.intercept), aliases, .then() callbacks, custom commands, frames, dialogs and scripts stay in the code as TODO comments with their line, the whole statement when it runs over several lines, and the checklist says what to use in Playwright instead. In Java, try/catch and counted for loops keep their shape; an if or while block stays whole as one TODO, so none of its steps runs unconditionally.

Limitations

• Locators are checked against the HTML you paste, as it is: style sheets are not applied (an element hidden by CSS counts as shown), and what the page loads later is not in it. Paste the page in the state the step sees it.
• Accessible names follow the common rules of the accessible name computation (labels, aria-label, aria-labelledby, alt, title and the text inside); unusual markup can get another name in a real browser.
• The rule-based reader reads English, one action or check per clause; a line it does not understand is listed with its number, never guessed. Rephrase it, or use Plan with AI.
• The project files name minimum versions of each framework; newer versions usually work, but check the release notes when you upgrade.
• Migration converts common commands; callbacks, custom commands, network stubbing and frames are left as TODO comments to finish by hand.
• With Selenium and WebdriverIO, Sign in once and reuse the session carries the session’s cookies over. An app that keeps its sign-in in local storage instead needs that storage copied too (Playwright’s storage state and Cypress’s cy.session() keep both).
• Write tests only for apps you own or are allowed to test. Steps that solve CAPTCHAs, sign up many accounts or collect data from a site are not written.

Privacy

Cloud AI only if you choose it: you see what is sent first

The journey, the pasted HTML and the generated code stay in your browser. Plan with AI runs on your device (your browser’s built-in AI or a model on your device), or, with a pass, on MySmartCoPilot’s cloud AI, which receives the journey with passwords masked and the role and name of the page’s elements, never the HTML.