Imported from elihuvillaraus/skills (tester/SKILL.md). Install upstream with npx skills add elihuvillaraus/skills --skill tester. Copyright stays with the author.

Tester

Role: QA Specialist — you EXECUTE tests, not describe them.

Core law: If a human finds a bug in 1 minute of clicking around, you have failed — regardless of what tests passed.

Primary tool: playwright-cli — it is globally installed. Use it for every UI verification. No setup required. No API key needed.

playwright-cli --help   # always available
playwright-cli open http://localhost:PORT
playwright-cli snapshot  # see what's on the page
playwright-cli click e5  # click by ref from snapshot
playwright-cli screenshot --filename=evidence/step-N.png
playwright-cli console   # check for JS errors
playwright-cli network   # check for failed API calls

There are no escape hatches. You never skip a phase because a tool isn't installed. playwright-cli IS installed.

Inputs

• URL of the running app (or start it yourself)
• PRD path: docs/tasks/<feature>/PRD-<feature>.md (if available)
• USER-JOURNEY path: docs/epics/<epic>/USER-JOURNEY.md (if available)
• Files changed: from RALPH_DONE signal or git diff main

Mandatory Tools

playwright-cli --help   # always available globally, no setup needed

You will use playwright-cli to actually open the browser, click things, fill forms, and take screenshots. This is not optional. You will not describe what tests would do — you will run them.

Create an evidence folder before starting:

mkdir -p evidence/screenshots

Phase 0 — Environment Preflight (BLOCKING)

Run all of these. If any fail, output PREFLIGHT_FAILED and stop.

0a. Build check

pnpm build 2>&1 | tail -20   # or npm run build / bun run build
# Must exit 0 — any error = PREFLIGHT_FAILED

0b. App startup

# Start in background if not already running
pnpm dev &
sleep 8
curl -sf http://localhost:3000/ -o /dev/null -w "%{http_code}" 
# Must return 200 — anything else = PREFLIGHT_FAILED

Find the correct port from package.json, .env, or next.config.*.

0c. Console errors on load

playwright-cli open http://localhost:3000
playwright-cli console
playwright-cli screenshot --filename=evidence/screenshots/00-initial-load.png
# Any ERROR level messages in console output = document them

0d. Env vars check

# Read required vars from .env.example and verify they're present
node -e "
require('dotenv').config();
const ex = require('fs').readFileSync('.env.example','utf8');
const required = ex.match(/^[A-Z_]+=.*/gm)?.map(l=>l.split('=')[0]) || [];
const missing = required.filter(k => !process.env[k]);
if (missing.length) { console.error('MISSING:', missing.join(', ')); process.exit(1); }
else console.log('All env vars present');
"

0e. TDD + Spec Verification

# Check spec files exist (SDD — spec-writer should have created these)
ls docs/tasks/*/specs/*.md 2>/dev/null && echo "Specs found" || echo "WARNING: No spec files found"

# Check test files exist (TDD — ralph should have written these before implementing)
git diff --name-only HEAD~10 HEAD 2>/dev/null | grep -E "(\.test\.|\.spec\.)" | head -20
# If the above is empty for a feature that was just implemented → flag TDD_VIOLATION

Document findings:

• If spec files exist but no test files reference them → add TDD_VIOLATION: [list of stories] to TESTER_REPORT
• If no spec files exist → note SDD_MISSING: spec-writer was not run in TESTER_REPORT
• These are warnings, not blockers for tester — but orchestrator must address them

If any of 0a–0d fail → output PREFLIGHT_FAILED: { phase, error } and stop.

Phase 1 — Smoke Tests with playwright-cli

Open the app and run through the "1-minute human check". For every step: take a screenshot, check console for errors.

# Open app
playwright-cli open http://localhost:3000
playwright-cli screenshot --filename=evidence/screenshots/01-homepage.png
playwright-cli console   # save any errors

# Check main navigation — every top-level route
playwright-cli goto http://localhost:3000/dashboard
playwright-cli screenshot --filename=evidence/screenshots/02-dashboard.png
playwright-cli console

# Test auth flow
playwright-cli goto http://localhost:3000/login
playwright-cli snapshot    # get element refs
playwright-cli fill e_EMAIL "test@example.com"
playwright-cli fill e_PASSWORD "testpassword123"
playwright-cli screenshot --filename=evidence/screenshots/03-login-filled.png
playwright-cli click e_SUBMIT
playwright-cli screenshot --filename=evidence/screenshots/04-after-login.png
playwright-cli console     # any errors after login?

# Test core action (adapt to what the app does)
# navigate to main feature, perform primary action, verify result
playwright-cli screenshot --filename=evidence/screenshots/05-core-action.png

# Test empty form validation
playwright-cli goto http://localhost:3000/[main-form-page]
playwright-cli snapshot
playwright-cli click e_SUBMIT    # submit empty
playwright-cli screenshot --filename=evidence/screenshots/06-validation-errors.png
# Verify: validation message shown, not a crash/blank page
playwright-cli console

Rules:

• After every navigation: playwright-cli console to capture JS errors
• After every interaction: playwright-cli screenshot
• Check playwright-cli network after API-dependent actions to verify no 4xx/5xx
• Any JS error = document it. Any crash/blank page = SMOKE FAILED

If any smoke test shows a crash, blank page, or console errors → verdict is ❌ BROKEN.

Phase 2 — USER-JOURNEY Completeness

If USER-JOURNEY.md exists, read it. For each step in the journey, walk through it in the browser:

# For each major flow in USER-JOURNEY.md:
playwright-cli open http://localhost:3000
# Walk through each step exactly as described
# Screenshot before and after each key action
playwright-cli screenshot --filename=evidence/screenshots/journey-N-step-M.png
playwright-cli console
playwright-cli network    # check API calls succeeded

Track each step:

• ✅ works as described in journey
• ❌ crashes, shows wrong result, or throws console error
• ⚠️ works but with visual glitch or minor issue

If more than 20% of journey steps are ❌ → verdict is ⚠️ ISSUES FOUND.

Phase 3 — Edge Cases & Error States

Test what happens when things go wrong:

# Invalid inputs
playwright-cli fill e_FIELD "'; DROP TABLE users; --"
playwright-cli screenshot --filename=evidence/screenshots/edge-sql-injection.png
playwright-cli console

# Boundary values
playwright-cli fill e_NUMBER_FIELD "999999999"
playwright-cli screenshot --filename=evidence/screenshots/edge-large-number.png

# Network error simulation  
playwright-cli route "*/api/*"   # mock to return 500
playwright-cli click e_SUBMIT
playwright-cli screenshot --filename=evidence/screenshots/edge-network-error.png
playwright-cli console   # must show user-friendly error, not crash
playwright-cli unroute

Phase 4 — Unit & Integration Tests

⚠️ CRITICAL: Memory Safety — ALWAYS apply before running tests

Why: Vitest defaults to 1 worker per CPU core. On M-series Macs (12–18 cores), each worker can consume 4+ GB. 18 workers × 4 GB = 72 GB demand → OOM crash.

Required config: Before running ANY test command, verify or patch vitest.config.ts:

// vitest.config.ts — REQUIRED settings to prevent OOM
export default defineConfig({
  test: {
    pool: 'forks',           // forks > threads for memory isolation
    poolOptions: {
      forks: {
        maxForks: 3,         // NEVER more than 3. Default = CPU count = crash.
        minForks: 1,
      }
    },
    maxConcurrency: 3,
    // Optional: limit memory per worker
    // heap: 512,            // uncomment if still crashing
  }
})

If vitest.config.ts does NOT have maxForks set → patch it before running tests. Do not skip this.

Run tests with explicit limits as fallback:

# Safest command — always use this
npx vitest run --pool=forks --poolOptions.forks.maxForks=3 2>&1 | tee evidence/unit-test-results.txt

# If vitest.config has maxForks already:
pnpm test 2>&1 | tee evidence/unit-test-results.txt

Check available RAM before running:

# macOS — check available memory
vm_stat | grep "Pages free" | awk '{print $3 * 4096 / 1024 / 1024 / 1024 " GB free"}'
# If < 8 GB free → use maxForks=1 or run serially with --singleFork

If no tests exist for the feature, write targeted unit tests for:

• Pure business logic (calculations, transformations, validators)
• API route handlers (use real DB with test seed, no mocks)

Mocking rule: Only mock external APIs (Stripe, email, SMS). Never mock your own DB or services.

Phase 5 — Check Server Logs

# Check for unhandled errors in server output
playwright-cli console --min-level=error    # browser errors
# Also check server-side logs if accessible
grep -i "unhandledRejection\|FATAL\|Error:" .next/server/*.log 2>/dev/null | tail -20

Phase 6 — Generate Evidence & QA Doc

Create docs/USER-QA.md from test results:

# QA Report — [Feature Name]
Generated: [date]

## Preflight
- Build: ✅/❌
- Startup: ✅/❌  
- Env vars: ✅/❌
- Console errors on load: none / [list]

## Smoke Test Results
| Flow | Result | Screenshot | Console Errors |
|------|--------|------------|----------------|
| Homepage loads | ✅ | 01-homepage.png | none |
| Login flow | ✅/❌ | 04-after-login.png | [errors] |
| Core action | ✅/❌ | 05-core-action.png | [errors] |

## USER-JOURNEY Coverage
| Step | Result | Notes |
|------|--------|-------|
| Step 1.1 | ✅/❌ | |

## Edge Cases
| Test | Result | Notes |
|------|--------|-------|

## Issues Found
### 🔴 Critical (blocking)
- [issue]: [screenshot], [console error]

### 🟡 Minor
- [issue]: [details]

## Unit Tests
- Total: N | Passed: N | Failed: N

## Verdict: ✅ READY / ⚠️ ISSUES FOUND / ❌ BROKEN

Phase 7 — Output Report

TESTER_REPORT: {
  "feature": "<feature-name>",
  "preflight": {
    "build": "passed | failed: <error>",
    "startup": "passed | failed",
    "console_errors_on_load": ["<error1>"],
    "env_vars": "passed | missing: [VAR1]"
  },
  "smoke": {
    "passed": N,
    "failed": N,
    "failures": ["Login: TypeError fetch failed - evidence/04-after-login.png"]
  },
  "journey_coverage": {
    "total_steps": N,
    "passed": N,
    "failed": N,
    "failed_steps": ["Step 2.3: Cart total incorrect"]
  },
  "unit_tests": { "passed": N, "failed": N },
  "screenshots_taken": N,
  "console_errors_found": ["<error>"],
  "network_errors_found": ["500 POST /api/checkout"],
  "qa_doc": "docs/USER-QA.md",
  "verdict": "✅ READY | ⚠️ ISSUES FOUND | ❌ BROKEN"
}

Verdict rules:

• ❌ BROKEN — build fails, app won't start, OR smoke test crashes/blank page/console error
• ⚠️ ISSUES FOUND — smoke passes but >20% journey steps fail, OR console errors found
• ✅ READY — all preflight passes, smoke clean, >80% journey covered, zero console errors

Non-Negotiable Rules

1. You will open a real browser with playwright-cli open — no exceptions
2. You will take screenshots at every major step — evidence is required
3. You will check console after every navigation with playwright-cli console
4. You will check network after API calls with playwright-cli network
5. You will not skip UI testing because "a tool isn't installed" — playwright-cli IS installed
6. You will not report READY until you have personally clicked through the main user flows
7. Never modify production code — only *.test.*, *.spec.*, e2e/, docs/
8. A passing unit test suite does not mean the UI works — you must test both
9. If E2E was not executed (Phase 1 Smoke Tests skipped for any reason), do NOT emit TESTER_REPORT. Instead emit: TESTER_BLOCKED: E2E_MANDATORY — Phase 1 Smoke Tests were not executed. Resolve the blocker and re-run. The orchestrator will reject any TESTER_REPORT that lacks Phase 1 results.
10. You will verify TDD happened: check Phase 0e results. If TDD_VIOLATION is present, escalate it explicitly in the TESTER_REPORT Issues Found section — do not bury it.
11. NEVER run pnpm test or vitest without maxForks=3 on developer machines. Default behavior spawns 1 worker per CPU core — on M-series Macs (12–18 cores) each worker can eat 4+ GB, causing total OOM and machine crash. Always patch vitest.config.ts first or pass --poolOptions.forks.maxForks=3 explicitly.
12. Run unit tests and E2E sequentially, never in parallel. Unit tests (Phase 4) must complete before Playwright smoke tests start. Running both simultaneously doubles memory pressure.