testing

DVstack Testing

Playwright-focused end-to-end testing skills. Every skill carries a working config, fixture, or spec that the agent copies into the project, so the standard arrives as code rather than as prose.

The fifteen skills are written to work together: naming conventions and architecture set the vocabulary, and the rest reference them.

Skills

Foundations

Skill	Covers	Ask for it with
playwright-naming-conventions	kebab-case files with role suffixes, PascalCase types, camelCase values, UPPER_SNAKE_CASE module constants, when to declare a type at all, and which folder anything belongs in	"what should I call this file"
playwright-test-architecture	Folder layout, playwright.config.ts, setup and teardown projects, page objects through fixtures, where each locator belongs, test.step with an assertion closing every step	"set up Playwright"
playwright-test-strategy	What belongs in e2e versus unit or API, risk-based coverage, smoke and regression tiers, when to delete a test, flaky budget	"how many tests does this need"
playwright-fixtures	Test vs worker scope, auto fixtures, option fixtures, setup/teardown around use(), merging fixture files, test locks	"this setup is repeated everywhere"

Writing tests

Skill	Covers	Ask for it with
playwright-step-validation	Zero hard-coded timeouts, the validation ladder — locator state, then waitForResponse with method and status, then response plus rendered UI, then waitForLoadState on quiet pages only	"what should this step wait for"
playwright-locators	Role-first locator order, data-testid policy, strict-mode scoping, web-first assertions	"fix these selectors"
playwright-api-testing	request fixture, seeding state through the API instead of the UI, toBeOK, cleanup, where the API/UI boundary belongs	"setup is slow because it clicks through the UI"
playwright-network-mocking	page.route, stubbing JSON, forcing errors, deferring responses for loading states, what not to mock	"test the error state"
playwright-hard-interactions	iframes, shadow DOM, dialogs, popups, uploads and downloads, drag and drop, clipboard, date pickers, virtualised lists, canvas, frozen time	"this element can't be clicked"

Quality gates

Skill	Covers	Ask for it with
playwright-accessibility-testing	@axe-core/playwright with WCAG tags, aria snapshots, keyboard and focus tests, what cannot be automated	"add a11y checks"
playwright-visual-testing	toHaveScreenshot, masking, animation and font determinism, per-platform snapshots, update policy	"set up visual regression"
playwright-code-review	Review checklist ordered by severity, from vacuous assertions down to unused imports, plus the ESLint config that automates half of it	"review these tests"

Running and fixing

Skill	Covers	Ask for it with
playwright-debugging	UI mode, trace viewer, codegen, --debug, page.pause(), attachments	"why did this fail"
playwright-flaky-tests	Trace triage, replacing sleeps with assertions, state isolation, honest retry policy	"this test is flaky"
playwright-ci	Sharding, browser caching, merged blob reports, traces as artifacts, secrets per environment	"add e2e to CI"

Conventions these skills enforce

• Page objects + fixtures. Intents live on the page object; the page object reaches the test through test.extend. A spec never calls new.
• One definition per element, at the narrowest scope. Used once: inline in the method. Used twice or asserted on from a spec: a getXxx() getter. Used across pages: the value goes in constants/selectors.ts and the page object wraps it.
• A step is a step only if it asserts. Every test is a sequence of test.step blocks and every block closes with at least one web-first expect — navigate then assert the landing element, click then assert the next element is visible and the previous one hidden.
• No hard-coded timeout anywhere. Not in a spec, a page object, a fixture, or a helper. The step waits on the state it needs: a locator, a response with its method and status, both together for data-driven UI, and a load state only on a small, quiet page.
• Web-first assertions everywhere. toBeVisible, toHaveText, toHaveURL. No isVisible() inside an expect, no waitForTimeout, no networkidle.
• Tests are independent — any test, alone, in any order, in parallel.
• Nothing hard-coded in a spec. Routes and endpoints in constants/, accounts in data/, pure functions in helpers/. A spec never reads process.env.
• Setup and teardown are projects, not globalSetup — they show in the report, record traces, and can use fixtures.
• Types are derived or inlined, not declared. type over interface, ReturnType<typeof fn> over a hand-written shape.

Install

claude plugin marketplace add vardanyandavit/dvstack
claude plugin install dvstack-testing@dvstack

In Cursor, add the same repository as a plugin marketplace and install dvstack-testing.

Or copy a single skill folder by hand — see the repository README.

Commands

Claude Code, after install:

• /dvstack-testing:playwright — force the whole pack when the change spans architecture, strategy, writing, or review.
• /dvstack-testing:playwright-locators (and the other skill names) — force one skill.

Cursor and other tools: invoke the skill name, or ask "use dvstack-testing" / "follow playwright-locators". See using commands.