AndreiBanu1

clarify-scope

Use before writing or planning any test when the ask is not yet precise — "add tests for checkout", "cover the new endpoint", "scope this first". Resolves the open decisions one question at a time, each with a recommended default, looking up in the repo anything that is a fact rather than a choice, and ends in a written brief the user confirms before any code is generated. Skip it when the request is already unambiguous.

handoff-notes

Use when test-automation work must continue in a fresh session or be picked up by someone else — the context window is filling mid-task, the day is ending, or the work is being passed on. Writes a handoff document to a temporary location (never the repo) containing what a successor needs and nothing they could read for themselves, with secrets excluded and open decisions recorded so they are not re-litigated. Not for finished work.

harden-locators

Use when a locator broke, is about to break, or is XPath — replacing positional and structural selectors with ones tied to what the user perceives. Covers XPath and brittle-CSS replacement patterns, disambiguating between multiple matches, handling text that varies across environments, and when a missing hook should be requested from the app team instead.

probe-conventions

Run this first in any repository, before authoring or fixing tests. Inspects the actual Playwright suite — directory layout, how specs import test, locator habits, data sources, run commands — and writes a short profile to .claude/test-profile.md that every other skill in this pack reads instead of assuming conventions. Use when the profile is missing or stale, when a skill reports it cannot find the profile, or after the suite is restructured.

register-fixture

Use when a newly written page object, API client, or helper needs to become available to specs — the Playwright fixture wiring step. Covers picking the right fixture file, scope choice, dependencies between fixtures, teardown for anything that holds a resource, and why a fixture is preferable to a beforeEach. Reads conventions from .claude/test-profile.md.

run-suite

Use when asked to run Playwright tests — one spec, a group, or everything — and the right command, target and flags are not obvious. Resolves the invocation from .claude/test-profile.md, prefers the repo's own runner script over hand-built CLI lines, chooses flags that make failures readable, and reports results without inflating them.

triage-red-suite

Use when a suite has multiple failures and needs to be sorted out — "run it and fix what's broken", "get the suite green". Classifies every failure by cause before touching code, fixes only the test-side ones, escalates product bugs instead of absorbing them, and reports a before/after that separates the two. For a single failing test, use the healer agents instead.

write-api-client

Use when a request spec needs a call that no existing client method covers — adding a domain client, adding a method to one, or extending the shared transport helper. Covers where the endpoint path belongs, what the method should return, and when a new client is warranted at all. Reads conventions from .claude/test-profile.md.

write-api-test

Use when adding a Playwright test that exercises an HTTP endpoint directly with no browser — REST or GraphQL — including response-shape validation, error paths, and cleanup after writes. Reads conventions from .claude/test-profile.md and mirrors the existing request specs. Not for tests that drive a page.

write-db-test

Use when adding a Playwright test that queries a database directly — verifying persistence, migrations, or that a write through the app landed correctly. Covers parameterised statements only, ordering for shared-row tests, teardown that survives failure, and keeping connection details out of the spec. Reads conventions from .claude/test-profile.md.

write-e2e-test

Use when adding a new browser-level Playwright spec — an end-to-end or journey test that drives a real page. Reads the repo's conventions from .claude/test-profile.md, mirrors the nearest existing spec, keeps interaction behind whatever model layer the repo uses, and verifies the spec actually runs before finishing. Not for API-only or component tests.

write-page-object

Use when a spec needs a screen or component wrapper that does not exist yet — the class that owns a page's locators and exposes intent-named methods. Follows whatever model conventions .claude/test-profile.md recorded, including the repo's base class and locator priority. Use harden-locators instead when editing an existing model's selectors.