preview-testing（预览测试）

Preview Testing

Comprehensive E2E + Security Tests for Vercel Preview Deployments

Trigger

This skill activates on:

• /preview-test - Manual invocation
• After PR creation against main or develop
• Before merge for production approval

Features

Usage

# Standard: Smoke + Visual + Security
/preview-test

# Security tests only
/preview-test --security

# With AI Exploratory Testing (optional)
/preview-test --ai

Security Tests (OWASP LLM Top 10 2025)

Prompt Injection (LLM01) - CRITICAL

• Direct Injection (Jailbreaks, DAN, Role Manipulation)
• System Prompt Extraction Prevention
• Context Hijacking via Fake History
• Indirect Injection via File Upload
• Multi-Language Bypass Attempts

File: tests/security/prompt-injection.spec.ts

Quota Bypass - CRITICAL (Cost Risk)

• API Authentication Bypass
• Email Spoofing Prevention
• Demo Mode Abuse
• Race Condition in Quota Check
• Test User Email Discovery

File: tests/security/quota-bypass.spec.ts

Stripe Webhook Security (PCI-DSS)

• Signature Validation
• Replay Attack Prevention
• Payload Manipulation Detection
• Subscription Fraud Prevention

File: tests/security/stripe-webhook.spec.ts

Workflow

1. PRE-DEPLOY GATES (quality-gate.yml)
   └── TypeScript Check
   └── Unit Tests
   └── Build Validation

2. PREVIEW DEPLOYMENT
   └── Generate Vercel Preview URL
   └── Wait 30s warmup

3. SMOKE TESTS (Playwright, <2min)
   └── Homepage loads
   └── Login flow works
   └── Critical Path: Upload → Analysis

4. VISUAL REGRESSION (Playwright, <3min)
   └── Screenshot comparison with baseline
   └── Mobile + Desktop breakpoints

5. SECURITY TESTS (Playwright, <5min) [CRITICAL]
   └── Quota Bypass Tests
   └── Prompt Injection Tests
   └── Stripe Webhook Security

6. GDPR REGION CHECK
   └── Verifies fra1 (Frankfurt) region

7. APPROVAL GATE
   └── All tests green → PR comment "Ready to merge"
   └── Security failures → BLOCK MERGE

Claude-in-Chrome MCP Integration

In addition to automated Playwright tests, interactive browser tools are available via MCP. These are ideal for:

• Visual debugging during development
• Ad-hoc testing without test scripts
• GIF recordings for PR documentation
• Live console/network inspection

When to Use Which Tool?

Interactive Preview Testing (Claude-in-Chrome)

// 1. Initialize browser tab context
mcp__claude-in-chrome__tabs_context_mcp({ createIfEmpty: true })

// 2. Create new tab for preview
mcp__claude-in-chrome__tabs_create_mcp()

// 3. Navigate to preview URL
mcp__claude-in-chrome__navigate({
  url: "https://your-app-xyz.vercel.app",
  tabId: <id>
})

// 4. Accessibility snapshot (better than screenshot for structure)
mcp__claude-in-chrome__read_page({ tabId: <id> })

// 5. Find interactive elements
mcp__claude-in-chrome__find({
  query: "login button",
  tabId: <id>
})

// 6. Click element
mcp__claude-in-chrome__computer({
  action: "left_click",
  ref: "ref_123",  // from find result
  tabId: <id>
})

// 7. Save screenshot
mcp__claude-in-chrome__computer({
  action: "screenshot",
  tabId: <id>
})

GIF Recording for PR Documentation

// 1. Start recording
mcp__claude-in-chrome__gif_creator({
  action: "start_recording",
  tabId: <id>
})

// 2. Screenshot for first frame
mcp__claude-in-chrome__computer({ action: "screenshot", tabId: <id> })

// 3. Perform user flow (login, upload, etc.)
mcp__claude-in-chrome__computer({
  action: "left_click",
  coordinate: [x, y],
  tabId: <id>
})

// 4. Wait for page transition
mcp__claude-in-chrome__browser_wait_for({
  text: "Welcome",
  tabId: <id>
})

// 5. Screenshot for last frame
mcp__claude-in-chrome__computer({ action: "screenshot", tabId: <id> })

// 6. Stop recording
mcp__claude-in-chrome__gif_creator({
  action: "stop_recording",
  tabId: <id>
})

// 7. Export GIF
mcp__claude-in-chrome__gif_creator({
  action: "export",
  tabId: <id>,
  filename: "login-flow-preview.gif",
  download: true,
  options: { quality: 15 }  // 1-30, lower = smaller file
})

Debugging: Console & Network

// Check JavaScript errors in console
mcp__claude-in-chrome__read_console_messages({
  tabId: <id>,
  onlyErrors: true,
  pattern: "error|exception"
})

// Inspect API requests
mcp__claude-in-chrome__read_network_requests({
  tabId: <id>,
  urlPattern: "/api/"  // Backend calls only
})

Example: Complete Interactive Preview Test

User: "Test the preview https://your-app-abc123.vercel.app"

Claude executes:
1. tabs_context_mcp → Get tab IDs
2. tabs_create_mcp → Create new tab
3. navigate → Open preview URL
4. read_page → Check accessibility snapshot
5. find → Search for "login button"
6. computer(screenshot) → Baseline screenshot
7. computer(left_click) → Click login button
8. read_console_messages → Check for JS errors
9. Report: "Landing page correct, no console errors"

Local Execution

# All preview tests (against local dev server)
npm run test:e2e

# Security tests only
npx playwright test tests/security/ --project=chromium

# Against Vercel Preview URL
BASE_URL=https://preview-xxx.vercel.app npx playwright test tests/security/

CI/CD Integration

The workflow is defined in .github/workflows/preview-test.yml and runs automatically on PRs against main or develop.

PR Comment (automatic)

After each run, a comment with test results is created:

## Preview Deployment Test Results

| Test Suite             | Status     |
| ---------------------- | ---------- |
| Smoke Tests            | ✅ success |
| Visual Regression      | ✅ success |
| Security Tests (OWASP) | ✅ success |
| GDPR Region (fra1)     | ✅ success |
| npm audit              | ⚠️ failure |

Blocking Logic

• Smoke Tests: Must pass
• Security Tests: Must pass (CRITICAL)
• Visual Regression: Warning on failure, doesn't block
• npm audit: Warning on failure, doesn't block

GDPR Compliance

• EU Data Residency: All tests verify Frankfurt (fra1) region
• No Real User Data: Synthetic test data (faker.js)
• Screenshots in EU: Playwright reports stored in GitHub Actions (EU region)
• Audit Trail: 30-day retention for security test reports

Expected Output

After running /preview-test:

Preview Testing Complete

RESULTS:
✅ Smoke Tests: 8/8 passed
✅ Visual Regression: 0 diffs
✅ Security Tests: 45/45 passed
  - Prompt Injection: 15 tests
  - Quota Bypass: 18 tests
  - Stripe Webhook: 12 tests
✅ GDPR Region: fra1 verified
⚠️ npm audit: 2 moderate vulnerabilities

RECOMMENDATION: Ready to merge to main

Troubleshooting

Security Tests Are Skipped

Tests use test.skip(!isVercelDeployment,...) and only run against production-like Vercel Preview. Locally they're skipped because some tests (e.g., rate limiting) require a real serverless environment.

Solution:

# Test against Vercel Preview
BASE_URL=https://your-preview-url.vercel.app npx playwright test tests/security/

Visual Regression Baseline Missing

On first run, baseline screenshots are created. Changes are marked as diffs.

Solution:

# Update baseline
npx playwright test tests/e2e/visual-regression.spec.ts --update-snapshots

Sources

• OWASP LLM Top 10 2025
• Playwright Best Practices
• Stripe Webhook Security
• Vercel Preview URL Polling