generate-tests（生成测试）

Generate Tests

Generate reusable Playwright .spec.ts test files from your app-navigator map and playbooks.

Core principle: Mapped knowledge should produce executable tests. This skill reads what app-navigator discovered and writes a test suite that lives in your repo and runs with npx playwright test.

When to Use

• After running /app-navigator setup and you want persistent, rerunnable tests
• When you need a baseline test suite for CI
• When trust-but-verify reveals gaps that should be codified as assertions
• When recommended by app-navigator or trust-but-verify

Not for: Backend/API tests, replacing hand-written integration tests, or one-shot verification (use trust-but-verify for that).

Invocation

• /generate-tests — generates smoke-depth tests (default)
• /generate-tests --depth functional — smoke + interaction tests
• /generate-tests --depth full — functional + edge cases + responsive

Process

digraph generate_tests {
    "Phase 1:\nPrerequisites" [shape=box];
    "Phase 2:\nGenerate Auth" [shape=box];
    "Phase 3:\nGenerate Page Tests" [shape=box];
    "Phase 4:\nValidate (optional)" [shape=box];
    "Done" [shape=doublecircle];

    "Phase 1:\nPrerequisites" -> "Phase 2:\nGenerate Auth";
    "Phase 2:\nGenerate Auth" -> "Phase 3:\nGenerate Page Tests";
    "Phase 3:\nGenerate Page Tests" -> "Phase 4:\nValidate (optional)";
    "Phase 4:\nValidate (optional)" -> "Done";
}

Phase 1: Prerequisites

1. App map: Check ~/.claude/skills/app-navigator/app-map.md exists

- If no: "Run /app-navigator setup first — I need the app map to generate tests." Offer to invoke it.

1. Playwright installed: Check if @playwright/test is in root package.json devDependencies

- If no: run pnpm add -D -w @playwright/test and npx playwright install chromium

1. Config: Check if playwright.config.ts exists in the project root

- If no: scaffold one with this template:

import { defineConfig, devices } from '@playwright/test';
import dotenv from 'dotenv';

dotenv.config({ path: '.env.test.local' });

export default defineConfig({
  testDir: './tests/e2e',
  timeout: 30_000,
  expect: { timeout: 10_000 },
  fullyParallel: true,
  retries: 1,
  use: {
    baseURL: process.env.BB_TEST_BASE_URL || 'http://localhost:5173',
    trace: 'on-first-retry',
  },
  projects: [
    { name: 'setup', testMatch: /auth\.setup\.ts/, teardown: undefined },
    {
      name: 'chromium',
      use: {
        ...devices['Desktop Chrome'],
        storageState: 'tests/e2e/.auth/user.json',
      },
      dependencies: ['setup'],
    },
  ],
});

1. Directories: Create tests/e2e/ and tests/e2e/.auth/ if they don't exist
2. Gitignore: Add tests/e2e/.auth/ and .env.test.local to .gitignore if not present

Phase 2: Generate Auth Setup

Read:

• ~/.claude/skills/app-navigator/playbooks/auth.md for the login flow
• ~/.claude/projects/<project>/memory/reference_local_auth.md for credentials

Create .env.test.local (gitignored) from memory credentials:

BB_TEST_EMAIL=<email from memory>
BB_TEST_PASSWORD=<password from memory>
BB_TEST_BASE_URL=<app URL from memory, default http://localhost:5173>

Generate tests/e2e/auth.setup.ts:

• Read BB_TEST_EMAIL and BB_TEST_PASSWORD from process.env
• Navigate to the login URL (e.g., http://localhost:5173/login)
• Follow the SSO/OAuth redirect to the auth domain
• Fill the email and password fields on the login form
• Click the submit/login button
• Wait for redirect through the OAuth callback path (e.g., /auth/callback) back to the app domain
• If an org selector appears: select the first org or one matching BB_TEST_ORG env var
• Save storageState to tests/e2e/.auth/user.json

The auth setup captures cookies from all domains visited (app + auth), which preserves the auth session across tests.

Important: The exact login flow (field names, redirect chain, org selector) comes from playbooks/auth.md. Read it carefully — don't assume a generic login form.

Phase 3: Generate Page Tests

Read:

• ~/.claude/skills/app-navigator/app-map.md — routes, key elements, interactions
• ~/.claude/skills/app-navigator/playbooks/navigation.md — how to navigate
• ~/.claude/skills/app-navigator/playbooks/interactions.md — common UI patterns

For each route in the app map, generate tests/e2e/<page-slug>.spec.ts.

Every generated file starts with:

// AUTO-GENERATED by generate-tests skill from app-navigator map
// Regenerate with: /generate-tests --depth <level>
// Last generated: YYYY-MM-DD

Depth: smoke (default)

test.describe('Page Name', () => {
  test('page loads', async ({ page }) => {
    await page.goto('/route');
    await expect(page.locator('text=Expected Heading')).toBeVisible();
  });

  test('key elements present', async ({ page }) => {
    await page.goto('/route');
    // Assert 2-3 key elements from app-map "Key Elements" field
    await expect(page.getByRole('button', { name: 'Create' })).toBeVisible();
  });
});

Depth: functional — adds interaction tests:

  test('create button opens dropdown', async ({ page }) => {
    await page.goto('/route');
    await page.getByRole('button', { name: 'Create' }).click();
    await expect(page.getByText('Create Document')).toBeVisible();
  });

Depth: full — adds edge cases + responsive:

  test('responsive: mobile layout', async ({ page }) => {
    await page.setViewportSize({ width: 375, height: 812 });
    await page.goto('/route');
    await expect(page.locator('text=Expected Heading')).toBeVisible();
  });

Test conventions:

• Tests grouped by depth with comments: // --- SMOKE ---, // --- FUNCTIONAL ---, // --- FULL ---
• All tests reuse storageState from the setup project — no re-login
• Selectors use getByRole, getByText, getByLabel — resilient to DOM changes
• Each test is independent — no ordering dependencies
• Routes sharing a component (e.g., tracked items) are grouped in one file
• Settings sub-pages are included in settings.spec.ts

Phase 4: Validate (optional)

After generating, offer:

"Tests generated. Want me to run them to make sure they pass?"

If yes:

1. Check dev server is reachable
2. Run npx playwright test
3. If tests fail: read the error, fix the test (adjust selector/wait/assertion), re-run. Max 3 fix iterations per test, then skip with a note.
4. Report: X/Y tests passing, Z skipped with reasons

If no: files are ready. Do NOT auto-commit — let the user decide.

Regeneration

Running /generate-tests again overwrites all files with the AUTO-GENERATED header. Hand-written test files without the header are NOT touched.

Red Flags

• Never hardcode credentials in test files. Always use env vars.
• Never commit .env.test.local — it contains real credentials.
• Never auto-commit generated tests. Let the user review and commit.
• Don't generate tests without an app-map. The map is the source of truth.
• Don't invent selectors. Every element name and assertion comes from the app-map.

After Generation

"Tests generated at tests/e2e/. Run them with npx playwright test. Want to verify your current feature branch against its plan? Run /trust-but-verify."

If the user says yes, invoke the trust-but-verify skill.

Integration

• Depends on: app-navigator (reads app-map.md and playbooks)
• Credential source: reference_local_auth.md in project memory → .env.test.local
• Recommended by: app-navigator (after setup), trust-but-verify (if no tests exist)
• Recommends: trust-but-verify (after generation)