screenshot-eval

name	screenshot-eval
description	Evaluate browser screenshots and page state for errors. Use when verifying features work correctly, detecting error conditions, or validating UI state during browser testing.

Screenshot Evaluation Skill

Activation: screenshot evaluation, error detection, browser testing, UI verification, feature validation, console errors, network failures

Overview

This skill provides a systematic approach to evaluating browser state during testing. It ensures features actually work by checking for console errors, network failures, and visual error indicators before declaring success.

Critical Rule: A feature is NOT complete until browser testing passes with no errors.

Evaluation Checklist

Execute these steps in order after navigating to the page or completing an action:

Step 1: Check Console for Errors

Use mcp__playwright__browser_console_messages with level="error"

Pass criteria:

• No console errors related to your feature
• React errors (component crashes) = FAIL
• Network/fetch errors = FAIL
• TypeScript/undefined errors = FAIL

Acceptable errors (can ignore):

• Third-party script warnings (analytics, etc.)
• Deprecated API warnings not related to your code

Step 2: Check Network Requests

Use mcp__playwright__browser_network_requests

Pass criteria:

• All API requests return 2xx status codes
• No 4xx errors (except expected 401/403 for auth tests)
• No 5xx server errors
• All requests complete (no pending/cancelled)

Check specifically:

• Your API endpoints return expected data
• Static assets load successfully
• WebSocket connections establish if needed

Step 3: Take Accessibility Snapshot

Use mcp__playwright__browser_snapshot

Error patterns to look for in snapshot:

Pattern	Indicates
"404" or "Not Found"	Missing route or resource
"500" or "Internal Server Error"	Backend crash
"Error" in heading/alert	Application error state
"Something went wrong"	Error boundary triggered
"Loading..." persisting	Infinite loading state
Empty content where data expected	Failed data fetch
"undefined" or "null" in text	Missing data handling

Step 4: Take Screenshot Proof

Use mcp__playwright__browser_take_screenshot

This screenshot is REQUIRED evidence that the feature works:

• Shows the expected UI state
• Proves the feature rendered correctly
• Attached to PR or issue as verification

Pass/Fail Criteria

PASS - All conditions met:

• No console errors (Step 1)
• All network requests succeeded (Step 2)
• No error patterns in snapshot (Step 3)
• Screenshot shows expected state (Step 4)

FAIL - Any condition:

• Console errors present
• Network requests failed (4xx/5xx)
• Error patterns detected in snapshot
• UI shows error state or missing content

Example Evaluation Workflow

Testing a New Feature

## Feature: Create Shootout Form

1. Navigate to /builder
   → mcp__playwright__browser_navigate url="/builder"

2. Check for errors after page load
   → mcp__playwright__browser_console_messages level="error"
   ✓ No console errors

3. Check network requests
   → mcp__playwright__browser_network_requests
   ✓ All 200 OK

4. Take snapshot, verify form elements
   → mcp__playwright__browser_snapshot
   ✓ Form visible with expected fields

5. Fill form and submit
   → mcp__playwright__browser_fill_form
   → mcp__playwright__browser_click (submit button)

6. Check for errors after submission
   → mcp__playwright__browser_console_messages level="error"
   ✓ No console errors

7. Check API call succeeded
   → mcp__playwright__browser_network_requests
   ✓ POST /api/v1/shootouts returned 201

8. Verify success state
   → mcp__playwright__browser_snapshot
   ✓ Success message visible OR redirected to new shootout

9. Take final screenshot proof
   → mcp__playwright__browser_take_screenshot
   ✓ Screenshot captured

Debugging a Failure

When evaluation fails:

1. Console error? → Check the error message, fix the code
2. Network failure? → Check backend logs, verify endpoint exists
3. Error in snapshot? → Identify the component showing error, trace the cause
4. Infinite loading? → Check if API call returned, verify data handling

MCP Tools Reference

Playwright (PR Verification)

Tool	Purpose	When to Use
mcp__playwright__browser_navigate	Go to URL	Start of test
mcp__playwright__browser_console_messages	Get console output	After page load, after actions
mcp__playwright__browser_network_requests	Get network activity	Verify API calls
mcp__playwright__browser_snapshot	Accessibility tree	Verify page content
mcp__playwright__browser_take_screenshot	Visual capture	Final proof
mcp__playwright__browser_click	Click element	User interactions
mcp__playwright__browser_fill_form	Fill form fields	Form testing

Chrome DevTools (Deep Debugging)

When Playwright checks show errors, use Chrome DevTools for deeper investigation:

Tool	Purpose	When to Use
mcp__chrome-devtools__get_console_logs	Full console output	Debug error details
mcp__chrome-devtools__get_network_logs	Request/response details	Debug API failures
mcp__chrome-devtools__get_computed_style	CSS debugging	Fix styling issues
mcp__chrome-devtools__get_element_info	Element properties	Debug layout problems
mcp__chrome-devtools__evaluate	Run JS in context	Debug state issues

Workflow: Playwright finds issue → Chrome DevTools investigates → Fix code → Playwright verifies

Integration with /pr Command

The /pr command enforces this evaluation before PR creation:

1. Runs quality gates (tests, lint, build)
2. Requires browser testing using this skill
3. Checks console for errors
4. Checks network for failures
5. Captures screenshot proof
6. Includes evidence in PR body

Never bypass browser testing - unit tests passing does NOT mean the feature works.

Common Issues and Solutions

React Error Boundary Triggered

• Check for unhandled promises
• Verify data shape matches component expectations
• Check for null/undefined access

API Returns 404

• Verify endpoint URL is correct
• Check if backend route exists
• Verify path parameters are valid

API Returns 500

• Check backend container logs
• Verify database migrations ran
• Check for missing environment variables

Infinite Loading

• API might be hanging - check backend
• Data might not trigger render - check state updates
• WebSocket might not connect - check connection setup

Empty Content

• API might return empty array - verify test data exists
• Conditional rendering might hide content - check conditions
• Data might be undefined - check fetch success