E2e

E2E setup signals

!ls -d e2e tests/e2e cypress playwright.config.* cypress.config.* tests/integration 2>/dev/null

!grep -rilE "@playwright/test|cypress|puppeteer|selenium|supertest|testcontainers|webdriver" package.json pyproject.toml 2>/dev/null | head

Instructions

Write an end-to-end / integration test for the flow in $ARGUMENTS. If the flow isn't clear, ask which user journey to cover before writing. If no e2e framework is set up, identify the idiomatic one and confirm before adding it.

Method

1. Detect the framework and convention. From the signals above, find the e2e tool (Playwright, Cypress, Puppeteer, Selenium for browser; supertest/requests + testcontainers for API) and read an existing e2e test to match its structure, selectors, fixtures, base URL/config, and naming. Mirror it. If none exists, propose the standard for the stack (Playwright for web, supertest-style for an API) and confirm.
2. Test the flow as a user/client experiences it, across real boundaries — UI → server → db, or HTTP request → response → persisted state. Assert on observable outcomes (visible text, URL, status code, returned/stored data), not internal implementation.
3. Use resilient selectors. Prefer role/label/test-id (getByRole, data-testid) over brittle CSS/XPath tied to styling. For APIs, assert status + body shape, not incidental fields.
4. Handle async and state properly. Use the framework's auto-waiting/explicit waits instead of fixed sleeps. Set up and tear down test data so runs are isolated and repeatable; don't depend on a prior test's leftovers or on a specific pre-seeded record that may not exist.
5. Keep it focused and stable. One coherent journey per test, named for the behavior. Cover the happy path plus a key failure (e.g. invalid login), not every permutation — e2e is expensive, so test the critical paths, not what unit tests already cover.

Rules

1. Match the project's existing e2e framework, selectors, and fixtures exactly; don't introduce a second framework or a new selector style unprompted.
2. No fixed sleep/arbitrary timeouts — use proper waiting; flaky timing makes e2e worthless. Call out any inherent nondeterminism.
3. Make tests self-contained: create the data they need and clean up, so they pass in any order against a fresh environment.
4. Assert real user-visible/client-visible outcomes; don't reach into internals to make a test pass.
5. After writing, give the exact command to run it and note any prerequisites (app/server running, env vars, a seeded test db, browser install like npx playwright install).

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: Pradnyeshp
• Source: Pradnyeshp/Claude-Skills
• License: MIT

Install and usage instructions live in the source repository linked above.