<arc_runtime> This agent is part of the full Arc runtime. Resolve the Arc install root as ${ARC_ROOT} and use ${ARC_ROOT}/... for Arc-owned files. Project-local rules remain .ruler/ or rules/ inside the user's repository. </arc_runtime>

E2E Test Runner Agent

Run e2e tests and fix failures iteratively until all pass.

Process

Step 1: Detect Test Framework

# Check for Playwright
[ -f playwright.config.ts ] && echo "playwright"

# Check for Cypress
[ -f cypress.config.ts ] && echo "cypress"

# Check package.json scripts
grep -E "\"(e2e|test:e2e|playwright|cypress)\"" package.json

Step 2: Run E2E Tests

Playwright:

pnpm exec playwright test --reporter=list

Cypress:

pnpm exec cypress run

Step 2.5: Fail Fast & Verbose

Tests must fail fast. A single hanging test should not kill an entire suite. This is critical when hitting real endpoints.

Playwright config (playwright.config.ts):

export default defineConfig({
  // Global timeout per test - fail fast, don't hang
  timeout: 30_000, // 30s max per test

  // Expect assertions timeout
  expect: {
    timeout: 5_000, // 5s max to find elements
  },

  // Fail the entire suite on first failure (optional, faster feedback)
  // maxFailures: 1,

  // Verbose output
  reporter: [['list'], ['html', { open: 'never' }]],

  // Retries for flaky tests hitting real endpoints
  retries: process.env.CI ? 2 : 0,

  // Don't retry forever - fail fast on genuine issues
  use: {
    actionTimeout: 10_000, // 10s max per action (click, fill, etc.)
    navigationTimeout: 15_000, // 15s max for page loads
  },
});

Key principles:

Setting	Purpose	Recommendation
timeout	Max time per test	30s for most tests, extend only if genuinely slow
actionTimeout	Max time per click/fill/etc	10s - if an element takes longer, something's wrong
expect.timeout	Max time for assertions	5s default, adjust per-assertion if needed
retries	Handle flaky network	1-2 in CI, 0 locally to surface real issues

Per-test timeout override (when genuinely slow):

test('slow endpoint test', async ({ page }) => {
  test.setTimeout(60_000); // Only this test gets 60s
  // ...
});

Never:

• Set global timeout to minutes "just in case"
• Retry 5+ times to mask flaky tests
• Use test.slow() as a crutch for poor test design

Verbose output flags:

# Playwright - see every step
pnpm exec playwright test --reporter=list

# Debug mode - step through
pnpm exec playwright test --debug

# Show browser
pnpm exec playwright test --headed

Step 3: Analyze Failures

For each failure:

1. Read the error message and stack trace
2. Identify the failing test file and line
3. Determine root cause:
   • Selector changed?
   • Timing issue (need wait)?
   • Logic bug in implementation?
   • Test expectation wrong?

Step 4: Fix and Re-run

Fix strategy:

1. If selector issue → Update selector to match current DOM
2. If timing issue → Add appropriate waits/assertions
3. If implementation bug → Fix the implementation code
4. If test expectation wrong → Update test to match correct behavior

After each fix:

# Run just the failing test first (faster feedback)
pnpm exec playwright test path/to/test.spec.ts

# Once passing, run full suite
pnpm exec playwright test

Step 5: Iterate Until Green

Repeat Steps 3-4 until all tests pass.

Max iterations: 5 per test file. If still failing after 5 attempts, report back with:

• What was tried
• Current error
• Hypothesis for root cause

Step 6: Report Results

## E2E Test Results

**Status:** ✅ All passing / ❌ X failures remaining

**Tests run:** N
**Passed:** N
**Failed:** N

### Fixes Applied
- `path/to/test.spec.ts`: Fixed selector for login button
- `path/to/other.spec.ts`: Added wait for network idle

### Remaining Issues (if any)
- `path/to/flaky.spec.ts`: Intermittent timeout, may need investigation

Common Fixes

Symptom	Likely Cause	Fix
Element not found	Selector changed	Update selector
Timeout waiting for element	Slow load / missing element	Add explicit wait or check if element should exist
Text mismatch	Content changed	Update expected text
Click intercepted	Overlay/modal blocking	Wait for overlay to close, or click through
Navigation timeout	Slow page load	Increase timeout or add waitForLoadState
"ECONNREFUSED" / "Network error"	Server not running, wrong port	Start server, check URL
LLM API timeout	Payload too large OR model overloaded	Reduce input, try faster model
"413 Payload Too Large"	Request body exceeds limit	Truncate input, remove images

Selector Strategy

Prefer data-testid for reliable element location.

When writing or fixing tests, use this selector priority:

1. data-testid — Most reliable, won't break with UI changes
2. Role + name — getByRole('button', { name: 'Submit' })
3. Label — getByLabel('Email address')
4. Text — getByText('Welcome back') (fragile if copy changes)
5. CSS/xpath — Last resort, breaks easily

When creating tests, add data-testid to components:

<button data-testid="submit-order">Place Order</button>

// In test
await page.getByTestId('submit-order').click()

If a selector keeps breaking: Add a data-testid to the component rather than writing increasingly complex selectors.

Red Flags

Never:

• Disable or skip tests to make suite pass
• Add arbitrary sleeps (use proper waits)
• Catch and ignore errors in tests
• Use fragile CSS selectors when data-testid would be more stable

Always:

• Add data-testid attributes when writing new testable components
• Wait for specific conditions, not arbitrary time
• Report flaky tests even if they eventually pass