Expect CLI

Expect is a command-line tool developed by Million Software, Inc. that enables AI coding agents to test application code in a real browser. It works by scanning unstaged changes or branch diffs, using AI to generate a test plan, and then executing those tests against a live browser session. Every bug found is captured on video for easy reproduction and debugging.

The tool operates in two primary modes. As a standalone CLI, developers can run it directly from the terminal with natural language instructions describing what to test. It supports an interactive terminal UI mode for guided test plan creation, as well as a headless mode that skips plan review and runs tests immediately. As an agent skill, it can be installed into AI coding agents like Claude Code or OpenAI Codex using the skills system, allowing agents to autonomously verify their code changes against a running application.

Expect supports multiple targeting strategies for determining what code to test. The default mode analyzes recent changes, while dedicated modes exist for testing all unstaged modifications or the full branch diff against the base branch. Saved flows allow teams to define reusable test scenarios by slug, making it easy to repeatedly run common test sequences like login flows or checkout processes without redefining them each time.

The CLI is installed via npm using npx and requires Node.js to run. It supports both Claude and OpenAI Codex as the underlying AI agent provider for generating test plans. Configuration is handled through an init command that sets up the tool globally and configures the agent skill integration.

Expect is built with TypeScript in a monorepo structure managed by Turborepo and pnpm. The project is licensed under FSL-1.1-MIT, the Functional Source License, which permits use and modification but converts to a full MIT license after two years. The source code is hosted on GitHub under the Million Software, Inc. organization.