testdriverai/testdriver:extract
Read information from the screen using AI and return it as a string
npx skills add https://github.com/testdriverai/testdriverai --skill testdriver:extract
<!-- Generated from extract.mdx. DO NOT EDIT. -->
Extract information from the current screen using AI and return it as a string. Describe what you want in natural language, and the AI reads the screen and returns the matching value — text, numbers, labels, status messages, or any other on-screen content.
Unlike assert(), which returns a boolean verdict, extract() returns the actual value so you can store it, compare it, or feed it into later steps and framework assertions.
const value = await testdriver.extract(description)
const value = await testdriver.extract({ description })
<ParamField path="description" type="string" required>
Natural language description of the information to read from the screen.
</ParamField>
<Info>
extract() also accepts an options object — extract({ description }) — which is equivalent to the positional form. The bare string form is the most common.
</Info>
Promise<string> — The information read from the screen. Returns the extracted value as text; parse or cast it yourself if you need a number or other type.
// Read text content
const title = await testdriver.extract('the page title');
const heading = await testdriver.extract('the main heading text');
// Read numbers and prices
const price = await testdriver.extract('the total price shown in the cart');
const count = await testdriver.extract('the number of items in the list');
// Read status and confirmation values
const status = await testdriver.extract('the order status');
const orderNumber = await testdriver.extract('the order confirmation number');
// Store and reuse across steps
const orderNumber = await testdriver.extract('the order confirmation number');
console.log('Order:', orderNumber);
// Combine with framework assertions
import { expect } from 'vitest';
const message = await testdriver.extract('the success message text');
expect(message).toContain('successfully');
// Cast to a number when you need to compare
const totalText = await testdriver.extract('the cart total as a number without currency symbol');
expect(Number(totalText)).toBeGreaterThan(0);
<Check>
Be specific about what to read
Precise descriptions produce cleaner values:
// ❌ Too vague — may return extra surrounding text
const price = await testdriver.extract('price');
// ✅ Specific — targets a single value
const price = await testdriver.extract('the total price in the order summary, digits only');
</Check>
<Check>
Ask for the format you want
Steer the output by describing the desired shape in the prompt:
// Strip currency symbols
const total = await testdriver.extract('the order total as a number without the dollar sign');
// Isolate a single field
const email = await testdriver.extract('the email address shown in the profile header');
</Check>
<Check>
Extract for detailed assertions
Use extract() when a boolean assert() isn't enough and you need the actual value to inspect:
const confirmation = await testdriver.extract('the confirmation number');
expect(confirmation).toMatch(/^ORD-\d{6}$/);
</Check>
<AccordionGroup>
<Accordion title="Capturing Confirmation Details">
const submitBtn = await testdriver.find('place order button');
await submitBtn.click();
// Read the confirmation the app generated
const orderNumber = await testdriver.extract('the order confirmation number');
const eta = await testdriver.extract('the estimated delivery date');
console.log(`Order ${orderNumber} arrives ${eta}`);
</Accordion>
<Accordion title="Reading Tooltip and Hover Content">
const icon = await testdriver.find('info icon next to the price');
await icon.hover();
const tooltipText = await testdriver.extract('the tooltip text');
expect(tooltipText).toContain('tax included');
</Accordion>
<Accordion title="Verifying Dynamic Values">
// Read a value before an action
const before = await testdriver.extract('the account balance');
const addBtn = await testdriver.find('add funds button');
await addBtn.click();
// Read it again after and compare
const after = await testdriver.extract('the account balance');
expect(Number(after.replace(/[^0-9.]/g, ''))).toBeGreaterThan(
Number(before.replace(/[^0-9.]/g, ''))
);
</Accordion>
<Accordion title="Passing Data Between Steps">
// Read a generated code on one screen...
const resetCode = await testdriver.extract('the password reset code');
// ...and type it into the next
const codeField = await testdriver.find('reset code input');
await codeField.click();
await testdriver.type(resetCode);
</Accordion>
</AccordionGroup>
import { beforeAll, afterAll, describe, it, expect } from 'vitest';
import TestDriver from 'testdriverai';
describe('Extraction', () => {
let testdriver;
beforeAll(async () => {
testdriver = new TestDriver(process.env.TD_API_KEY);
await testdriver.auth();
await testdriver.connect();
});
afterAll(async () => {
await testdriver.disconnect();
});
it('should capture the order confirmation', async () => {
await testdriver.focusApplication('Google Chrome');
// Complete a checkout
const checkoutBtn = await testdriver.find('checkout button');
await checkoutBtn.click();
const placeOrderBtn = await testdriver.find('place order button');
await placeOrderBtn.click();
// Verify we reached confirmation
await testdriver.assert('the order confirmation page is displayed');
// Extract the details the app generated
const orderNumber = await testdriver.extract('the order confirmation number');
const total = await testdriver.extract('the order total as a number without currency symbol');
// Assert on the extracted values
expect(orderNumber).toBeTruthy();
expect(Number(total)).toBeGreaterThan(0);
});
});
<Note>
Like assertions, extract() reads the screen fresh on every call — it is not cached — so it always reflects the current state of the app.
</Note>
assert() - Verify screen state with a boolean AI judgmentfind() - Locate elements to interact withparse() - Detect all UI elements on screenTake testdriverai/testdriver:extract from the repository into ~/.claude/skills for personal
use, or into .claude/skills inside a project.
The agent identifies a skill by the name field in its header. Two skills with the
same name cannot sit side by side — one of them will be ignored.