web-qa-regression

Turn a user flow of a web application into an automated end-to-end regression test, written in the project's own test framework and proven to pass.

Arguments

• A flow (required; ask for it when missing): what the user does and what must be true at the end, or a bug report whose fix the test should protect.
• A URL (optional): where the application runs; otherwise find how the project starts it.

Flow

1. Find the project's end-to-end setup: the framework (Playwright, Cypress, WebdriverIO or another), its configuration, where tests live, how they are named, and the helpers, fixtures and page objects that exist. If the project has none, ask whether to add Playwright Test; never add a second framework next to an existing one.
2. Walk the flow by hand first by delegating it to the web-qa-tester agent (or with the playwright MCP server yourself): record each step, the element used and the state that proves the step worked. If the flow does not work by hand, stop and report that instead of writing a test.
3. Write the test following the project's conventions:
   • locate elements the way a user finds them: role and accessible name, label, visible text; a test id where the project uses them; never a generated class name or a position in the page;
   • assert what the user sees after each meaningful step, and wait on those assertions, never on fixed delays;
   • make the test create the data it needs and leave nothing behind, so it can run alone and in any order;
   • read credentials and base URLs from the project's configuration or environment variables.
4. Run it with the project's command, three times in a row. A test that fails once is flaky: fix the cause (usually a missing wait on a visible state) rather than adding a retry.
5. Prove it can fail when it protects a bug fix: say how you checked that it fails without the fix, or that you could not check.

Report

• The test file and what it covers, step by step.
• The command to run it and the result of the three runs.
• Anything in the application that made the flow hard to test (missing labels, no stable way to locate an element), as suggestions.
• What the test does not cover.