qa-test-author

ISOLATION CONTRACT

This skill is permitted to use only:

1. The structured claim list from qa-spec-reader.
2. The running TEST application — navigated and observed via mcp__plugin_playwright_playwright__* tools.
3. The TxAdmin page objects as a locator library (structural import only — not a source of test intent).

Forbidden inputs — refuse immediately if offered:

• Code diffs, PR contents, or commit messages.
• Author-side test files or dev-skill outputs.
• Direct reads from main/ source tree.

ISOLATION GATE: If a forbidden input is detected, state: "I have detected a forbidden input: [name it]. Authoring proceeds from the claims list and TEST-app observation only."

---

AUTHORING PROCEDURE

Step 1: Project scaffold

Produce a Playwright TypeScript test file. Use this file header for every suite:

import { test, expect } from '@playwright/test';
// Admin surface: import TxAdmin page objects for locator reuse
// import { HomePageAdm } from '../pages/TxAdmin/homePageAdm';
// import { AircraftsPageAdm } from '../pages/TxAdmin/aircraftsPageAdm';
// 360 surface: scaffold fresh page objects in src/pages/Tx360/

// Auth: storageState loaded from playwright.config.ts project dependency
// 'Setup User Login' project must run first and write .auth/storageState.TEST.json

Global use block to include in playwright.config.ts (or test-level test.use):

use: {
  baseURL: process.env.TEST_BASE_URL,
  storageState: '.auth/storageState.TEST.json',
  trace: 'retain-on-failure',
  screenshot: 'only-on-failure',
  video: 'retain-on-failure',
}

Step 2: Observe the TEST application

Before writing any selector, navigate the TEST application using Playwright MCP tools to observe the actual UI:

1. Use mcp__plugin_playwright_playwright__browser_navigate to open the relevant screen.
2. Use mcp__plugin_playwright_playwright__browser_snapshot to capture the accessibility tree.
3. Derive locators from the accessibility tree using the priority order below.

Never guess a locator. If a TEST-app observation is not possible (environment offline), note the assumption in a comment and use a getByRole/getByLabel placeholder with a // TODO: verify in TEST marker.

Step 3: Locator selection (strict priority order)

Priority	Method	Use when
1	getByRole('button', { name: '...' })	Buttons, links, checkboxes, radio buttons
2	getByLabel('...')	Form inputs with associated labels
3	getByText('...')	Static text elements, headings, status messages
4	locator('[data-testid="..."]')	When a data-testid attribute is present
5	TxAdmin page object method	Admin surface only — import from src/pages/TxAdmin/

Never use: CSS class selectors, nth-child, dynamic IDs (containing GUIDs or numeric DB keys), layout-position selectors (first() without a semantic anchor), or any selector derived from reading source code.

Step 4: Wait strategy (deterministic, no hard sleeps)

Use these wait patterns. Never use page.waitForTimeout.

Scenario	Correct wait
Navigation / page load	await page.waitForURL('**/expected-path**') or await expect(page).toHaveURL(...)
Element becomes visible	await expect(locator).toBeVisible()
Form save / server postback	await page.waitForResponse(resp => resp.url().includes('/save') && resp.status() === 200)
Async SignalR notification	await page.waitForEvent('console', msg => msg.text().includes('notification-key')) or await page.waitForResponse(resp => resp.url().includes('/hub'))
Element disappears (spinner)	await expect(spinner).not.toBeVisible()
Hard timeout guard (ERROR case)	Wrap in Promise.race([action, new Promise((_, reject) => setTimeout(() => reject(new Error('TIMEOUT')), 30000))]) and catch to record ERROR

For Web Forms partial-page postbacks (Admin surface): after triggering a postback, wait for the UpdatePanel completion via page.waitForResponse matching the __EVENTTARGET request, or assert that the expected DOM change is visible — never sleep.

Step 5: Write test cases from claims

For each CLAIM-n from the spec-reader output, write at minimum:

• One happy test covering the primary success path.
• Document edge, negative, abuse, and cross-module cases as test stubs with test.skip and a comment like // adversarial-critic will expand — the critic phase fills these in.

Test naming convention: [CLAIM-n] [type] — [brief description]

test('CLAIM-1 happy — technician marks task complete, compliance status updates', async ({ page }) => {
  // STEP: Navigate to maintenance task screen
  await page.goto('/admin/tasks');
  // STEP: Select the task
  await page.getByRole('link', { name: /task-name/i }).click();
  // STEP: Mark as complete
  await page.getByRole('button', { name: /mark complete/i }).click();
  // WAIT: Server postback completes
  await page.waitForResponse(resp => resp.url().includes('/tasks/complete') && resp.status() === 200);
  // ASSERT: Compliance status refreshes
  await expect(page.getByRole('status', { name: /compliance/i })).toContainText('Up to Date');
});

Step 6: Page objects for the 360 surface

When the surface is 360, scaffold a new page object class in src/pages/Tx360/. Minimum structure:

// src/pages/Tx360/PartOrderPage360.ts
import { Page, Locator } from '@playwright/test';

export class PartOrderPage360 {
  readonly page: Page;
  readonly vendorSelect: Locator;
  readonly quantityInput: Locator;
  readonly partNumberInput: Locator;
  readonly submitButton: Locator;
  readonly validationMessage: Locator;

  constructor(page: Page) {
    this.page = page;
    this.vendorSelect = page.getByLabel('Vendor');
    this.quantityInput = page.getByLabel('Quantity');
    this.partNumberInput = page.getByLabel('Part Number');
    this.submitButton = page.getByRole('button', { name: /submit order/i });
    this.validationMessage = page.getByRole('alert');
  }

  async navigate() {
    await this.page.goto('/360/inventory/part-orders/new');
    await expect(this.submitButton).toBeVisible();
  }
}

All locators in the page object must come from TEST-app observation, not source reading.

Step 7: Auth fixture

Include in the test file:

// Auth is pre-loaded from .auth/storageState.TEST.json via playwright.config.ts
// The 'Setup User Login' project dependency writes this file.
// Do NOT call login() inside test bodies.

If a permission-boundary test requires a different role:

test.use({ storageState: '.auth/storageState.TEST.readonly.json' });

---

OUTPUT

Return the Playwright TypeScript test file contents and any new page object files as code blocks, followed by the updated JSON envelope with all test cases populated at result: "NOT_RUN".

---

FORBIDDEN ACTIONS

• Do NOT call Read, Grep, or Glob against main/.
• Do NOT use page.waitForTimeout in any test.
• Do NOT embed credentials or hard-code environment URLs.
• Do NOT import from author-side test files (UnitTests.Core, TransactionScope).
• Do NOT emit URLs pointing to production or staging.
• Do NOT read eval/rubric.private.md or any path under eval/.