Write browser tests in English, let Claude drive the clicks

What it does

Shortest is an end-to-end testing framework that sits on top of Playwright and lets you describe browser actions in plain English. You pass a sentence like shortest("Login to the app using email and password") along with credentials, and the framework uses the Anthropic Claude API to translate that instruction into actual browser interactions. It also handles API testing and supports lifecycle hooks and callback assertions for verifying database state or specific page values after the AI execution finishes.

The interesting bit

The framework replaces the tedious selector-and-action choreography of traditional E2E tests with an LLM intermediary, while still letting you drop down into raw Playwright page objects when you need precise control. It is essentially opinionated glue between Anthropic and Playwright, which means you get the reliability of a mature browser automation engine without writing brittle click paths.

Key highlights

• Tests are written as natural language strings; the AI determines the DOM interactions.
• Built on Playwright, so underlying browser context and page objects remain accessible.
• Supports test chaining and reusable flow composition via arrays and spread operators.
• Includes helpers for GitHub 2FA authentication and Mailosaur email validation.
• Offers lifecycle hooks and post-execution callbacks for custom assertions.

Caveats

• Requires an active Anthropic API key, so test runs are neither free nor fully offline.
• The project is largely an abstraction layer over Playwright and Claude; if either service has issues, Shortest inherits them.

Verdict

Teams drowning in brittle E2E selectors should try this, provided they are comfortable with Anthropic API costs and the slight opacity of LLM-driven interactions. If you need deterministic, millisecond-precise browser choreography without network dependencies, stick to raw Playwright.