AI Integration

TWD produces structured, deterministic output that AI agents can parse and act on autonomously. Every test run returns the same pass/fail signals for the same inputs — no flakiness, no ambiguity. Whether you want your AI assistant to write better tests, generate tests from browser interactions, or run tests autonomously, TWD has you covered.

Quick Start: Claude Code Plugin

The fastest way to get AI-powered TWD testing is with the Claude Code plugin:

claude plugin install BRIKEV/twd-ai

This gives Claude Code two capabilities:

Command	Description
/twd:setup	Interactive setup — detects your project config and generates .claude/twd-patterns.md
twd skill	Autonomous agent — writes tests, runs them via twd-relay, fixes failures, and re-runs until green

Other AI Tools

For Cursor, Copilot, Windsurf, and other AI tools, use the Agent Skills CLI:

npx skills add BRIKEV/twd-ai

This copies TWD context into your AI tool's configuration file (.cursorrules, .github/copilot-instructions.md, etc.).

---

Features at a Glance

1. AI Context

Teach your AI assistant (Claude, Cursor, Copilot, Windsurf) how to write correct TWD tests by providing a comprehensive prompt with API reference, patterns, and common pitfalls.

Best for: Getting AI to write correct TWD tests on the first try.

Read the AI Context guide

---

2. AI Remote Testing (twd-relay)

A WebSocket bridge that lets AI agents trigger test runs and stream results back, without launching a browser automation tool. Your Vite dev server is already running with TWD loaded -- the relay just connects to it.

Best for: AI agents that need to run tests, read failures, and iterate.

Read the AI Remote Testing guide

---

3. Auto-Invocation (Claude Code)

When you install the TWD plugin (claude plugin install BRIKEV/twd-ai), Claude Code can automatically invoke the testing agent when it detects the task is relevant. For example:

• You ask: "Add a search filter to the orders page"
• Claude implements the feature
• Claude sees the twd skill and spawns it as a sub-agent
• The agent writes tests, runs them via npx twd-relay run, reads failures, fixes, and re-runs until green
• Claude continues with your task

The agent works autonomously in a forked context — your main conversation isn't cluttered with test-writing steps. You can also set up your project interactively:

/twd:setup

---

How They Work Together

You can use these features independently or combine them:

AI Context           →  AI writes better tests
AI Remote Testing    →  AI runs tests and reads results
Claude Code Plugin   →  AI writes, runs, and fixes tests autonomously

A typical workflow:

1. Plugin / Skills install TWD context into your AI agent
2. The TWD agent writes tests, runs them via twd-relay, and fixes failures
3. The autonomous validation loop continues until all tests pass

MCP Integration

TWD also provides an experimental MCP server that works with Playwright MCP to generate test code from browser automation data. This is an early feature — if you're interested, check the TWD MCP package for details.