GeezerVison

For Humans & Agents

Humans

Agents

Less code written

Less tokens consumed

CodeceptJS reads as actions, not selectors. Where Playwright and WebDriver make you wrangle locators by hand, you describe what the user does — the framework resolves the locator and dispatches it through Playwright, WebDriver, or Puppeteer.

Tests explain themselves

Agents understand intent faster

You write automation the way you'd write a test case: what the user does, what they should see. No tangled selector chains polluting the file. A teammate scanning the scenario — or an agent picking it up — grasps the intent in seconds and can adjust it without reverse-engineering anything.

Declarative actions

Fewer failed tries, faster feedback

I.selectOption handles native <select>, custom comboboxes, and ARIA listboxes the same way. I.fillField works on plain inputs and rich editors alike. CodeceptJS adapts to whatever your page actually uses — fewer flaky retries for humans, fewer HTML dumps and tool calls for agents.

Debug live with pause()

Debug live via MCP

Drop pause() anywhere in a scenario and you land in an interactive REPL with the page state still warm — try commands, inspect elements, patch the test in place. Agents get the same surface through MCP: they drive the running test, pause on any step, poke at the page, then update the code and rerun in debug mode.

Opinionated by default

Universal skill set

Page objects, data fixtures, mobile and web plugins — the framework ships with strong defaults instead of leaving you to assemble them. Agents inherit the same structure through the official skill set: they know how to write, debug, refactor, and organize tests the way the framework expects.

Feature('Milestones'); Before(({ login }) => { login();}); Scenario('Create a milestone @smoke', ({ milestonePage }) => { milestonePage.open(); milestonePage.create({ name: 'Q2 Release', dueDate: '2024-06-30' }); milestonePage.checkVisible('Q2 Release');}); Scenario('Milestone tracks task progress', ({ I, milestonePage }) => { milestonePage.open(); milestonePage.select('Q2 Release'); I.see('0 / 5 tasks', '.progress'); I.see('0%', '.progress-bar');}); Scenario('Complete a milestone', ({ I, milestonePage }) => { milestonePage.open(); milestonePage.select('Q2 Release'); I.click('Mark as Complete'); I.see('Completed', '.milestone-status');});

Tests that outlive the framework wars

CodeceptJS started in 2015. We watched Protractor, TestCafe, NightmareJS come and go, how WebDriver and CDP tools evolve. But our tests never changed. A test which was running on Protractor back then runs on Playwright today. Whatever comes next: your tests stay.

Write what, not how

I.click('Login') finds the button — by text, by label, by ARIA role, by CSS — and clicks it. You describe the action. The framework picks the strategy. The test reads like the spec it came from.

The tiniest API in browser testing

A dozen commands cover most of what you'll write. Less to learn on day one, a test file a junior can read on the first morning. Simplicity is the feature.

AI Powered

When a selector breaks, AI heals it on the fly. When a check is hard to express with locators, write it in plain words as I.decide('save button is enabled') and a decision model like Jev will decide if the page meets it.

Designed for agentic testing

Claude or Cursor run a test and can control it live via MCP server. No guessing, running, and retrying loop. Agent can set a breakpoint and debug test in realtime in a browser.

Mobile and web in one framework

Appium for iOS and Android. The same I.click or I.tap(), I.see, I.swipeLeft you use on the web. Mobile testing works on same princinples.

Composable Page Objects

A page object can use any other page object, custom step, or helper by name. The DI container resolves them at runtime. One page object can reference other and vise versa with no more circular import failure.

Data Management

Declare your API endpoints once, use built-in tools to create and clean up date in each test. A new user needed? Declare it: I.have('user', { role: 'admin' }) this creates one, and removes after a test.

Output you can actually read

Output is an actual scenario written as steps. See how test went in terminal or in produced report.

What teams say.

“Test an application like Percona Monitoring and Management with so many dashboards & metric plots.”

Puneet Kala

Frontend QA Automation Engineer · Percona

“CodeceptJS helps us write tests which are good to read and easy to write, run on several browsers.”

Mitko Tschimev

Frontend Tech Lead · My Porsche Core

GeezerVison

CodeceptJS docs

End-to-end testing, written like prose.