Auto Playwright

Run Playwright tests using AI.

Setup

1. Install auto-playwright dependency:

npm install auto-playwright -D

2. This package relies on talking with OpenAI (https://openai.com/). You must export the API token as an enviroment variable or add it to your .env file:

export OPENAI_API_KEY='sk-..."

3. Import and use the auto function:

import { test, expect } from "@playwright/test";
import { auto } from "auto-playwright";

test("auto Playwright example", async ({ page }) => {
  await page.goto("/");

  // `auto` can query data
  // In this case, the result is plain-text contents of the header
  const headerText = await auto("get the header text", { page, test });

  // `auto` can perform actions
  // In this case, auto will find and fill in the search text input
  await auto(`Type "${headerText}" in the search box`, { page, test });

  // `auto` can assert the state of the website
  // In this case, the result is a boolean outcome
  const searchInputHasHeaderText = await auto(`Is the contents of the search box equal to "${headerText}"?`, { page, test });

  expect(searchInputHasHeaderText).toBe(true);
});

Setup with Azure OpenAI

Include the StepOptions type with the values needed for connecting to Azure OpenAI.

import { test, expect } from "@playwright/test";
import { auto } from "auto-playwright";
import { StepOptions } from "../src/types";

const apiKey = "apikey";
const resource = "azure-resource-name";
const model = "model-deployment-name";

const options: StepOptions = {
   model: model,
   openaiApiKey: apiKey,
   openaiBaseUrl: `https://${resource}.openai.azure.com/openai/deployments/${model}`,
   openaiDefaultQuery: { 'api-version': "2023-07-01-preview" },
   openaiDefaultHeaders: { 'api-key': apiKey }
};

test("auto Playwright example", async ({ page }) => {
  await page.goto("/");

  // `auto` can query data
  // In this case, the result is plain-text contents of the header
  const headerText = await auto("get the header text", { page, test }, options);

  // `auto` can perform actions
  // In this case, auto will find and fill in the search text input
  await auto(`Type "${headerText}" in the search box`, { page, test }, options);

  // `auto` can assert the state of the website
  // In this case, the result is a boolean outcome
  const searchInputHasHeaderText = await auto(`Is the contents of the search box equal to "${headerText}"?`, { page, test }, options);

  expect(searchInputHasHeaderText).toBe(true);
});

Usage

At minimum, the auto function requires a plain text prompt and an argument that contains your page and test (optional) objects.

auto("<your prompt>", { page, test });

Browser automation

Running without the test parameter:

import { chromium } from "playwright";
import { auto } from "auto-playwright";

(async () => {
  const browser = await chromium.launch({ headless: true });
  const context = await browser.newContext();
  const page = await context.newPage();
  // Navigate to a website
  await page.goto("https://www.example.com");

  // `auto` can query data
  // In this case, the result is plain-text contents of the header
  const res = await auto("get the header text", { page });

  // use res.query to get a query result.
  console.log(res);
  await page.close();
})();

Debug

You may pass a debug attribute as the third parameter to the auto function. This will print the prompt and the commands executed by OpenAI.

await auto("get the header text", { page, test }, { debug: true });

You may also set environment variable AUTO_PLAYWRIGHT_DEBUG=true, which will enable debugging for all auto calls.

export AUTO_PLAYWRIGHT_DEBUG=true

Supported Browsers

Every browser that Playwright supports.

Additional Options

There are additional options you can pass as a third argument:

const options = {
  // If true, debugging information is printed in the console.
  debug: boolean,
  // The OpenAI model (https://platform.openai.com/docs/models/overview)
  model: "gpt-4-1106-preview",
  // The OpenAI API key
  openaiApiKey: 'sk-...',
};

auto("<your prompt>", { page, test }, options);

Supported Actions & Return Values

Depending on the type of action (inferred by the auto function), there are different behaviors and return types.

Action

An action (e.g. "click") is some simulated user interaction with the page, e.g. a click on a link. Actions will return `undefined`` if they were successful and will throw an error if they failed, e.g.

try {
  await auto("click the link", { page, test });
} catch (e) {
  console.error("failed to click the link");
}

Query

A query will return requested data from the page as a string, e.g.

const linkText = await auto("Get the text of the first link", { page, test });

console.log("The link text is", linkText);

Assert

An assertion is a question that will return true or false, e.g.

const thereAreThreeLinks = await auto("Are there 3 links on the page?", {
  page,
  test,
});

console.log(`"There are 3 links" is a ${thereAreThreeLinks} statement`);

Why use Auto Playwright?

Supported Playwright Actions

• locator.blur
• locator.boundingBox
• locator.check
• locator.clear
• locator.click
• locator.count
• locator.fill
• locator.getAttribute
• locator.innerHTML
• locator.innerText
• locator.inputValue
• locator.isChecked
• locator.isEditable
• locator.isEnabled
• locator.isVisible
• locator.textContent
• locator.uncheck
• page.goto

Adding new actions is easy: just update the functions in src/completeTask.ts.

Pricing

This library is free. However, there are costs associated with using OpenAI. You can find more information about pricing here: https://openai.com/pricing/.

Implementation

HTML Sanitization

The auto function uses sanitize-html to sanitize the HTML of the page before sending it to OpenAI. This is done to reduce cost and improve the quality of the generated text.

ZeroStep

This project draws its inspiration from ZeroStep. ZeroStep offers a similar API but with a more robust implementation through its proprietary backend. Auto Playwright was created with the aim of exploring the underlying technology of ZeroStep and establishing a basis for an open-source version of their software. For production environments, I suggest opting for ZeroStep.

Here's a side-by-side comparison of Auto Playwright and ZeroStep:

[^3]: Uses ZeroStep proprietary API. [^4]: Uses some Playwright API, but predominantly relies on Chrome DevTools Protocol (CDP).