---

name: radulator-qa-tester description: Automated QA testing for Radulator's 18 medical calculators across radiology, hepatology/liver, and urology specialties. Tests accuracy, collects browser diagnostics, generates Playwright tests, and manages three-branch Git workflow (dev1→test1→main). Use when testing Radulator calculators, reviewing PRs with qa label, verifying medical formulas, generating test reports, or creating comprehensive test suites.

Radulator QA Tester

Comprehensive automated testing system for the Radulator medical calculator web application. This skill acts as a highly-competent manual tester that reads test specifications, runs the web app, collects diagnostics, verifies calculations, and produces detailed test reports.

Supported Calculators (18 Total)

Radiology (6)

1. Adrenal CT Washout
2. Adrenal MRI Chemical Shift Index (CSI)
3. Prostate Volume & PSA Density
4. Renal Cyst (Bosniak Classification)
5. Spleen Size (Upper Limit of Normal)
6. Hip Dysplasia Indices (DDH)

Hepatology/Liver (9)

7. ALBI Score (Albumin-Bilirubin Grade)
8. Adrenal Vein Sampling - Cortisol (Cushing)
9. Adrenal Vein Sampling - Aldosterone (Hyperaldosteronism)
10. BCLC Staging (Barcelona Clinic Liver Cancer)
11. Child-Pugh Score
12. Milan Criteria (HCC Transplant Eligibility)
13. MELD-Na Score
14. MR Elastography (Liver Fibrosis Staging)
15. Y-90 Radiation Segmentectomy

Urology (3)

16. IPSS (International Prostate Symptom Score)
17. RENAL Nephrometry Score
18. SHIM Score (Sexual Health Inventory for Men)

Quick Start

When a Radulator PR is opened with the qa label:

1. Identify which calculator(s) changed
2. Run automated tests using Playwright MCP
3. Verify calculations against expected formulas
4. Collect diagnostics (console logs, network requests, screenshots)
5. Generate Playwright regression tests
6. Post test report to PR via GitHub MCP
7. Merge if tests pass, otherwise request fixes

Prerequisites

This skill requires two MCP servers configured via Docker Desktop MCP Toolkit (recommended):

• Playwright MCP: Browser automation for testing (21 tools)
• GitHub MCP: Repository management and PR operations (40+ tools)

Quick Setup (Claude Code CLI)

1. Install Docker Desktop with MCP Toolkit enabled
2. Add MCP servers via Docker Desktop UI:
   • Playwright (from catalog)
   • GitHub Official (from catalog, requires OAuth)
3. Create .mcp.json in Radulator project root:

   {
     "mcpServers": {
       "MCP_DOCKER": {
         "command": "docker",
         "args": ["mcp", "gateway", "run"],
         "type": "stdio"
       }
     }
   }
   

4. Connect Claude Code:

   docker mcp client connect claude-code
   

5. Verify: docker mcp tools ls should show ~67 tools

See references/mcp_setup.md for detailed installation and configuration instructions.

Core Testing Process

Step 1: Detect Calculator Changes

When a PR is opened:

Use GitHub MCP to:
1. Get PR details and diff
2. Search diff for: src/components/calculators/*.jsx
3. Extract calculator name from changed files

Step 2: Setup Test Environment

# Clone PR branch
git clone <repo-url> --branch <pr-branch>
cd radulator

# Install dependencies
npm install

# Start dev server
npm run dev
# Wait for "ready" message, note the port (typically 5173)

Step 3: Run Playwright Tests

For each changed calculator:

1. Navigate: Use Playwright MCP to open http://localhost:PORT
2. Select Calculator: Click sidebar item matching calculator name
3. Load Test Data: Read test cases from references/test_cases.md
4. Fill Inputs: Use browser_type to enter test values
5. Execute: Click "Calculate" button
6. Capture Results: Extract output values from results section

Step 4: Verify Calculations

# Run verification script
python scripts/verify_calculators.py <calculator_name> '<test_case_json>'

# Compare actual vs expected values
# Tolerance: ±0.1 for percentages, ±0.01 for densities

Step 5: Collect Diagnostics

Use Playwright MCP to gather:
- browser_console_messages: Capture JavaScript errors/warnings
- browser_network_requests: Verify no unexpected API calls
- browser_take_screenshot: Document test state

Step 6: Generate Regression Tests

# Use Playwright MCP
browser_generate_playwright_test

# Or use helper script
python scripts/generate_playwright_test.py "<calculator_name>" '<test_data_json>'

# Save to: tests/<calculator-name>.spec.js
# Commit to branch: qa/<calculator-name>

Step 7: Report Results

Create a test report comment on the PR with:

## QA Test Report: <Calculator Name>

### Test Case: <Test Name>

**Inputs:**
- Input 1: <value>
- Input 2: <value>
- ...

**Expected Results:**
- Output 1: <expected> (formula: <formula>)
- Output 2: <expected> (formula: <formula>)

**Actual Results:**
- Output 1: <actual> ✅ / ❌
- Output 2: <actual> ✅ / ❌

**Status:** PASS / FAIL

**Diagnostics:**
- Console Errors: <count> errors (details below)
- Network Requests: <count> requests
- Render Time: <time> ms

<If tests failed:>
**Recommended Fixes:**
1. <specific fix suggestion>
2. <code reference if applicable>

**Screenshots:**
![Test Results](<screenshot_url>)

**Console Logs:**

Step 8: Merge or Request Changes

If ALL tests PASS:
  - Post comment: "✅ QA: All tests passed"
  - Merge PR into test1 via GitHub MCP
  
If ANY test FAILS:
  - Post comment with detailed failure report
  - Do NOT merge
  - Wait for developer to fix and push updates

Calculator-Specific Testing

Adrenal CT Washout

Test Inputs (from references/test_cases.md):

• Typical adenoma: unenh=10, portal=100, delayed=40
• Non-adenoma: unenh=20, portal=80, delayed=70

Verification Script:

from scripts.verify_calculators import adrenal_ct_washout
result = adrenal_ct_washout(10, 100, 40)
# Expected: absolute_washout=66.7%, relative_washout=60.0%

UI Selectors:

• Input: input[name="unenh"], input[name="portal"], input[name="delayed"]
• Results: .results container

Prostate Volume & PSA Density

Test Inputs:

• Normal: length=4, height=3, width=3.5, psa=2
• Elevated: length=3.5, height=2.5, width=3, psa=5.5

Verification Script:

from scripts.verify_calculators import prostate_volume_psa_density
result = prostate_volume_psa_density(4, 3, 3.5, 2)
# Expected: volume=21.84 mL, psa_density=0.092 ng/mL²

Other Calculators

See references/test_cases.md for complete test data for all six calculators:

1. Adrenal CT Washout
2. Adrenal MRI Chemical Shift
3. Prostate Volume & PSA Density
4. Renal Cyst (Bosniak)
5. Spleen Size (Upper Limit)
6. Hip Dysplasia Indices

Three-Branch Workflow

Radulator uses: dev1 → test1 → main

PR from dev1 → test1

1. Developer creates PR with qa label
2. This skill runs automated tests
3. If tests pass → merge to test1
4. If tests fail → request fixes

PR from test1 → main

1. After successful test1 merge, create PR to main
2. Re-run tests to catch regressions
3. If tests pass → merge to production
4. Deploy to radulator.com

See references/workflow.md for detailed workflow documentation.

Quality Gates

Before Merging to test1:

• ✅ Calculations accurate within tolerance
• ✅ No console errors
• ✅ Results render in < 500ms
• ✅ All inputs properly validated
• ✅ Interpretation text correct

Before Merging to main:

• ✅ All test1 gates passed
• ✅ No regressions detected
• ✅ Playwright tests committed
• ✅ Documentation updated
• ✅ References section complete

Example Test Report

## QA Test Report: Adrenal CT Washout Calculator

### Test Case: Typical Adenoma

**Inputs:**
- Unenhanced HU: 10
- Portal Venous HU: 100
- Delayed (15 min) HU: 40

**Expected Results:**
- Absolute Washout: 66.7% (formula: ((portal - delayed)/(portal - unenh)) × 100)
- Relative Washout: 60.0% (formula: ((portal - delayed)/portal) × 100)
- Interpretation: "Suggests benign adenoma"

**Actual Results:**
- Absolute Washout: 66.7% ✅
- Relative Washout: 60.0% ✅
- Interpretation: "Suggests benign adenoma" ✅

**Status:** ✅ PASS

**Diagnostics:**
- Console Errors: 0
- Network Requests: 3 (all to localhost)
- Render Time: 142 ms

**Generated Regression Test:**
Committed to: `tests/adrenal-ct-washout.spec.js`

---

✅ **QA: All tests passed. Safe to merge to test1.**

Troubleshooting

Common Issues

Issue: Calculator not found in sidebar Fix: Check calculator name matches calcDefs array in App.jsx

Issue: Input field selectors not working Fix: Inspect actual input field names/IDs, update selectors

Issue: Results not rendering Fix: Check for console errors, verify Calculate button click registered

Issue: Math precision errors Fix: Use .toBeCloseTo() with appropriate precision tolerance

Debugging Tools

// Get all input fields
await page.$$eval('input', els => els.map(e => ({name: e.name, value: e.value})))

// Get results container HTML
await page.locator('.results').innerHTML()

// Monitor all console messages
page.on('console', msg => console.log('Browser:', msg.text()))

Best Practices

1. Verify MCP connection before starting tests:

   docker mcp client ls  # Check connection status
   docker mcp tools ls   # Verify tools available
   

2. Use Docker Desktop MCP Toolkit for team consistency and CI/CD readiness
3. Commit .mcp.json to version control for team collaboration
4. Run tests in headless mode for CI, headful for debugging
5. Take screenshots after every major step for documentation
6. Keep test data in sync with medical literature references
7. Generate Playwright tests for all calculators to build regression suite
8. Document deviations clearly with formula references and code snippets
9. Test on multiple browsers (Chrome, Firefox, Safari) using Playwright
10. Monitor token usage - simplify Playwright interactions if needed

Resources

• Test Cases: references/test_cases.md
• Workflow Guide: references/workflow.md
• MCP Setup: references/mcp_setup.md
• Verification Scripts: scripts/verify_calculators.py
• Test Generator: scripts/generate_playwright_test.py

Notes

• Recommended for Claude Code CLI users: This skill uses Docker Desktop MCP Toolkit
• Docker MCP Gateway provides team collaboration and CI/CD-ready configuration
• For CI/CD integration, see references/workflow.md GitHub Actions section
• Playwright runs identically on macOS, Linux, and Windows via Docker
• All test data based on peer-reviewed medical literature
• Generated Playwright tests can run independently in CI pipelines
• .mcp.json should be committed to git for team sharing