Expect

Visit Website

Introduction

Expect is a skill that lets a coding agent test its own work in a real browser. You add it to Claude Code, Codex, Cursor or another supported agent, and when you run /expect it reads your git changes, generates a test plan for them and executes that plan with Playwright.

It spawns subagents that act like real logged-in users, using your actual cookies, to look for issues and regressions. The README lists what it checks: performance (long animation frames, INP and LCP), security (npm dependencies, CSRF and known vulnerabilities), design problems (broken hover states, links and buttons) and app completeness (missing metadata and dead links). The agent then fixes what Expect found and reruns it to confirm.

The maintainers position it against general computer-use agents. Rather than screenshots and mouse coordinates, Expect uses Playwright for fast DOM automation, which is quicker and more reliable for testing. It runs entirely on your machine, and a --ci flag (or add github-action) sets up a workflow that tests every pull request headlessly, skipping cookie extraction, auto-approving the plan and enforcing a 30-minute timeout.

Expect is maintained by Million (million.dev), and it auto-detects which coding agents are on your PATH.

The source code is at millionco/expect on GitHub under the FSL-1.1-MIT (Functional Source License with an MIT future licence) licence, with about 3,600 stars; the most recent commit was on 10 April 2026.

Browser agents that verify a coding agent’s work are a new category, and Expect is one of the more concrete open implementations. The key question when trying it is whether its checks are reproducible: run the same verification twice on an unchanged branch and confirm the results agree before trusting it to gate merges.

It also helps to be clear about what Expect is not. It does not replace a maintained end-to-end suite with stable assertions; it complements one by checking new or changed behaviour that nobody has written a test for yet. A practical pattern is to let it explore the change, then turn any important behaviour it confirms into a permanent Playwright or Cypress test.

Because it is pre-1.0 and under a Functional Source License, review the licence terms and pin versions in CI so an upgrade does not change behaviour unexpectedly.

Treat it as an assistant to human review, not a replacement.

Launch and background: There’s An AI For That lists the Expect repository in its test automation category.

Features

1. Change-aware test plans: Reads the git diff and plans tests for what changed.
2. Real-browser runs: Executes plans with Playwright.
3. Logged-in subagents: Simulates users with your real cookies locally.
4. Performance checks: Long animation frames, INP and LCP.
5. Security checks: npm dependencies, CSRF and vulnerabilities.
6. Design and completeness checks: Hover states, broken links and buttons, metadata and dead links.
7. CI mode: --ci runs headless with a 30-minute timeout.
8. Agent auto-detection: Works with Claude Code, Codex, Cursor and others; -a selects one.

Pros

1. Closes the agent loop: The agent that wrote the code also verifies it in a browser.
2. Local-first: Runs on your machine without sending your app to a hosted service.
3. Free to use: Open source under the FSL-1.1-MIT (Functional Source License with an MIT future licence) licence, so the core tool costs nothing to use.
4. Actively maintained: The repository had commits as recent as 10 April 2026.
5. Community: About 3,600 GitHub stars.

Cons

1. Source-available licence: FSL-1.1-MIT is not an OSI open-source licence until the MIT future-licence date applies.
2. Agent-dependent: Needs a supported coding agent installed.
3. Young project: Version numbers are still 0.0.x.
4. Not a regression suite: It plans fresh tests from diffs rather than maintaining a fixed suite.

Usage and Links (Example)

Inside your coding agent:

Fetch https://www.expect.dev/ and run init script
/expect

For pull requests:

expect add github-action   # or run with --ci in your own workflow

Resources:

• Expect website

Pricing

Expect is free to use; the source is published under the Functional Source License 1.1 with an MIT future licence (FSL-1.1-MIT).

Recommended for?

1. Developers using coding agents who want the agent to check its own UI changes.
2. Small teams without QA that need a quick browser pass on every PR.

Conclusion

Expect is a lightweight way to add browser verification to agent-written code, and its Playwright-first design keeps runs quick.

Use it as a fast pre-merge check alongside, not instead of, a maintained regression suite.

On CheckOps, Expect is filed under the End-to-end & functional and Release & CI gates stages. Related tools listed here include Playwright MCP, Chrome DevTools MCP, Browser Use, Stagehand, Shortest and Vibium. A short trial on a real project, with your own code, data and pipeline, is usually the fastest way to choose between them, because fit matters more than feature lists. Facts here come from the official site, documentation and repository, checked on 11 November 2026.