Getting started

Quick start

Two slash commands in Claude Code, one browser sign-in, and one plain-English prompt. That's everything between a fresh install and a passing AI-generated test.

Step 1Install

Add the Diffie plugin to Claude Code

In any Claude Code conversation, run these two slash commands. The first registers Diffie's plugin marketplace; the second installs the Diffie plugin.

/plugin marketplace add diffie-ai/diffie-skill
/plugin install diffie@diffie

Once installed, the plugin is available in every Claude Code conversation for the current project.

Step 2Sign in

Authenticate on first use

The first time you ask the plugin to do anything, it opens a browser window and signs you into Diffie. No prior account setup needed. Every session afterwards authenticates automatically; if you ever get signed out, the plugin re-opens the login for you.

Step 3First test

Describe a flow in English

Ask Claude Code, in plain English, for the test you want. The plugin plans it with you before touching Diffie.

> Create an E2E test for my app at https://app.example.com
  that covers logging in with email and password.

The plugin will ask about credentials (it can store them as LOGIN_EMAIL and LOGIN_PASSWORD secrets), scan your source for selector hints, present a spec for approval, then create and run the test. You'll get pass or fail with a link to the recording.

State is tracked in a .diffie-qa.md file in your project root, so the next conversation already knows what tests you have and which are passing.

What you get

  • A real Playwright test, generated from your prompt and the selector hints in your source code.

  • A video recording of every run, viewable by anyone with the link (no Diffie account required).

  • Encrypted secrets storage so tests can log in without credentials leaving your workspace.

  • Auto-fix on failure. Reprocess a test with a short fix prompt and the plugin regenerates and reruns.

Next: understand what happens under the hood

Every Diffie test flows through the same five stages, whether you triggered it from Claude Code, a PR comment, or the dashboard. Read how that lifecycle works so failures are easier to diagnose.