10x-e2e

Provide a change-id or plan path and the Skill picks each browser-level risk, maps the flow against the running app, and generates one test per risk from a seed exemplar. Each test is reviewed against five anti-patterns and verified green before completion.

Reliable E2E tests use role-based locators like getByRole, run independently with their own setup and cleanup, wait for state instead of time, and assert a business outcome tied to a named risk. A deliberate-break check confirms the test fails when the risk materializes.

The shipped seed, rules, and prompt template are tuned for Playwright, but the principles are tool-agnostic. On Cypress, WebdriverIO, or Selenium you encode your tool's locator and wait idioms into your own variant of the rules file.

Skip E2E when an isolated function, endpoint contract, or integration test can prove the risk, such as pure logic, validators, config, or scaffolding. E2E fits only risks crossing several system boundaries or existing solely in the rendered UI.

The assertion is likely decorative, checking something like a page title instead of the business outcome the risk threatens. The deliberate-break technique inverts the protected behavior and confirms the test goes red, exposing assertions that protect nothing.