test-fixer

SKILL.md

Test Fixer

Debug and fix failing Playwright tests by visually inspecting the actual page state.

⚠️ CRITICAL: Browser Session Behavior

Each MCP call = new browser session. Browser CLOSES after each call. You CANNOT navigate in one call and interact in another. Use browser_run_code for ALL test debugging. If you need to return to a specific state (e.g., after login), you MUST redo ALL steps from scratch.

Workflow

  1. Parse the error - Extract failing test file, line number, selector, and error type
  2. Capture page state - Use browser_run_code to navigate AND interact in one session
  3. Analyze the issue - Compare expected vs actual selectors/state
  4. Fix the code - Update page object and/or test spec
  5. Verify - Re-run the single failing test

Step 1: Parse Error

Extract from test output:

  • File path: e.g., tests/olx-landing.spec.ts:45
  • Error type: timeout, strict mode violation, element not found, assertion failed
  • Failing selector: the locator that failed
  • Expected vs received: for assertion errors

Step 2: Capture Page State

CRITICAL: Browser Session Behavior

Each MCP call creates a NEW browser session. For multi-step operations, use browser_run_code!

Option A: Single browser_run_code Call (Recommended)

Run all exploration steps in one session:

python .claude/skills/mcp-client/scripts/mcp_client.py call playwright browser_run_code '{
  "code": "
    // Navigate to page
    await page.goto(\"https://www.olx.ro\");

    // Handle cookies
    const acceptBtn = page.getByRole(\"button\", { name: \"Accept\" });
    if (await acceptBtn.isVisible({ timeout: 3000 }).catch(() => false)) {
      await acceptBtn.click();
      await page.waitForTimeout(500);
    }

    // Replicate test steps up to failure
    const categoryLink = page.getByRole(\"link\", { name: /Auto, moto/i }).first();
    await categoryLink.click();
    await page.waitForTimeout(1500);

    // Capture state at failure point
    const snapshot = await page.accessibility.snapshot();
    return JSON.stringify({
      url: page.url(),
      didNavigate: page.url().includes(\"auto\"),
      snapshot: snapshot
    }, null, 2);
  "
}'

Option B: Simple Navigate (When No Interactions Needed)

browser_navigate returns both page AND snapshot in one call:

python .claude/skills/mcp-client/scripts/mcp_client.py call playwright browser_navigate '{"url": "https://example.com"}'

Option C: Run Test in Debug Mode

For complex scenarios requiring visual debugging:

# Run the specific failing test with headed browser and pause on failure
npx playwright test "tests/example.spec.ts:45" --headed --debug

Option D: Use Test Artifacts

Check the test-results folder for screenshots and traces:

# View trace from failed test
npx playwright show-trace test-results/*/trace.zip

Step 3: Analyze Issue

Error Type Analysis Typical Fix
strict mode violation Multiple elements match Add .first(), use more specific selector
element not visible Element exists but hidden Wait for visibility, check cookie banners
timeout waiting for selector Selector outdated Update to match actual DOM
toHaveURL failed Navigation didn't happen Add waitForURL, check click target
toContainText failed Wrong assumed text Discover actual error message text
click opens submenu, not page Multi-step interaction needed Add click on "View all" or final nav link

Common Issue: Wrong Assumed Text

If assertion fails on toContainText or toHaveText, the test probably assumed the wrong error message.

Fix: Discover the actual text:

python .claude/skills/mcp-client/scripts/mcp_client.py call playwright browser_run_code '{
  "code": "
    await page.goto(\"https://example.com/login\");

    // Cookies
    const acceptBtn = page.getByRole(\"button\", { name: /accept/i });
    if (await acceptBtn.isVisible({ timeout: 3000 }).catch(() => false)) {
      await acceptBtn.click();
    }

    // Trigger the error condition
    await page.fill(\"input[type=email]\", \"wrong@test.com\");
    await page.fill(\"input[type=password]\", \"wrongpass\");
    await page.click(\"button[type=submit]\");
    await page.waitForTimeout(3000);

    // Capture ACTUAL error text
    const errors = await page.locator(\"[class*=error], [role=alert]\").evaluateAll(els =>
      els.filter(e => e.offsetParent !== null).map(e => ({
        text: e.textContent?.trim(),
        selector: e.className ? \".\" + e.className.split(\" \")[0] : \"[role=alert]\"
      }))
    );

    return JSON.stringify({ errors }, null, 2);
  "
}'

Then update test with actual text:

// Before (assumed)
await expect(page.locator('.error')).toContainText('Invalid credentials');

// After (discovered)
await expect(page.getByRole('alert')).toContainText('Email sau parolă incorectă');

Selector Priority (best to worst)

  1. getByRole() - Accessible, stable
  2. getByTestId() - Stable if devs maintain it
  3. getByText() - Readable, somewhat stable
  4. locator('[href="..."]') - For links
  5. CSS selectors - Last resort

Understanding Element Behavior

Use browser_run_code to test what clicking does:

python .claude/skills/mcp-client/scripts/mcp_client.py call playwright browser_run_code '{
  "code": "
    await page.goto(\"https://www.olx.ro\");

    // Dismiss cookies
    const acceptBtn = page.getByRole(\"button\", { name: \"Accept\" });
    if (await acceptBtn.isVisible({ timeout: 3000 }).catch(() => false)) {
      await acceptBtn.click();
    }

    // Record initial state
    const initialUrl = page.url();

    // Click the element that failed
    const element = page.getByRole(\"link\", { name: /Auto, moto/i }).first();
    await element.click();
    await page.waitForTimeout(1500);

    // Analyze what happened
    const finalUrl = page.url();
    const didNavigate = finalUrl !== initialUrl;

    // Look for submenus or dropdowns
    const submenuVisible = await page.locator(\"[class*=submenu], [class*=dropdown], [class*=menu]\").first().isVisible().catch(() => false);

    // Get visible links that might be \"View all\" type
    const navLinks = await page.getByRole(\"link\").filter({ hasText: /vezi|view|all|toate/i }).allTextContents();

    const snapshot = await page.accessibility.snapshot();

    return JSON.stringify({
      initialUrl,
      finalUrl,
      didNavigate,
      submenuVisible,
      navLinks,
      snapshot
    }, null, 2);
  "
}'

Step 4: Fix the Code

Page Object Updates

When selectors break, update the page object locator:

// Before (broken)
this.searchButton = page.getByRole('button', { name: 'Search' });

// After (from snapshot showing actual name)
this.searchButton = page.getByRole('button', { name: /Căutare/i });

Test Updates

When assertions fail, update test logic:

// Before (navigation not waiting)
await categoryLink.click();
await expect(page).toHaveURL(/category/);

// After (proper navigation wait)
await Promise.all([
  page.waitForURL(/category/),
  categoryLink.click()
]);

Multi-Step Navigation Fixes

When click opens submenu instead of navigating:

// Before (assumes direct navigation)
async navigateToCategory(categoryName: string) {
  await this.page.getByRole('link', { name: categoryName }).click();
}

// After (handles submenu)
async navigateToCategory(categoryName: string) {
  // Click category to open submenu
  const categoryLink = this.page.getByRole('link', { name: categoryName }).first();
  await categoryLink.click();

  // Click "View all" to actually navigate
  const viewAllLink = this.page.getByRole('link', { name: /Vezi toate anunturile/i });
  await viewAllLink.click();
}

Common Fixes Reference

Cookie Banner Blocking

// Add to page object
async acceptCookies() {
  const banner = this.page.getByRole('dialog').filter({ hasText: /cookie|privacy/i });
  if (await banner.isVisible({ timeout: 2000 }).catch(() => false)) {
    await this.page.getByRole('button', { name: /accept|agree/i }).click();
  }
}

Multiple Elements Match

// Use .first() for first match
await page.getByRole('link', { name: 'Category' }).first().click();

// Or use more specific parent context
await page.locator('nav').getByRole('link', { name: 'Category' }).click();

// Or use getByTestId if available
await page.getByTestId('login-submit-button').click();

Element Not Ready

// Wait for element to be actionable
await expect(element).toBeVisible();
await element.click();

// Or wait for network idle
await page.waitForLoadState('networkidle');

Navigation Not Completing

// Wait for URL change explicitly
await Promise.all([
  page.waitForURL(/expected-path/),
  triggerElement.click()
]);

Step 5: Verify

Re-run only the failing test:

npx playwright test "tests/olx-landing.spec.ts" -g "Category links navigate correctly"

Or run by line number:

npx playwright test tests/olx-landing.spec.ts:45

Checklist Before Fixing

  • Read the failing test file
  • Navigate to the URL the test uses (with browser_run_code)
  • Accept cookies/dismiss popups if present
  • Replicate test steps up to failure
  • Capture fresh DOM snapshot
  • Understand actual UI behavior (dropdowns, submenus, multi-step flows)
  • Find the actual element in snapshot
  • Update selector AND flow to match reality
  • Re-run the single failing test to verify

Common Misunderstandings

Assumption Reality Fix
Click navigates directly Opens submenu first Add step to click "View all" or similar
Element is immediately visible Requires scroll or hover Add scrollIntoViewIfNeeded() or hover action
Form submits on button click Requires Enter key or specific trigger Use correct submission method
Modal closes on outside click Requires explicit close button Click the close/X button
Page loads immediately Redirect to different domain Add waitForURL with correct domain pattern

References

See references/error-patterns.md for detailed diagnosis and fixes for:

  • Timeout errors (selector, URL)
  • Strict mode violations
  • Assertion failures (text, URL)
  • Element state errors (not visible, disabled)
  • Navigation issues

Quick Debug Script

For complex debugging, use this exploration script:

python .claude/skills/mcp-client/scripts/mcp_client.py call playwright browser_run_code '{
  "code": "
    // ====== CUSTOMIZE THIS SECTION ======
    const URL = \"https://www.olx.ro/cont/\";
    const WAIT_FOR_REDIRECT = /login\\.olx\\.ro/;
    // =====================================

    await page.goto(URL);

    // Handle cookies
    const acceptBtn = page.getByRole(\"button\", { name: /accept/i });
    if (await acceptBtn.isVisible({ timeout: 3000 }).catch(() => false)) {
      await acceptBtn.click();
      await page.waitForTimeout(500);
    }

    // Wait for redirect if specified
    if (WAIT_FOR_REDIRECT) {
      await page.waitForURL(WAIT_FOR_REDIRECT, { timeout: 10000 });
    }

    // Get comprehensive element info
    const inputs = await page.locator(\"input\").evaluateAll(els =>
      els.map(e => ({
        type: e.type,
        name: e.name,
        placeholder: e.placeholder,
        testid: e.dataset.testid
      }))
    );

    const buttons = await page.locator(\"button\").evaluateAll(els =>
      els.map(e => ({
        text: e.textContent?.trim(),
        testid: e.dataset.testid,
        type: e.type
      }))
    );

    const links = await page.getByRole(\"link\").evaluateAll(els =>
      els.slice(0, 20).map(e => ({
        text: e.textContent?.trim()?.substring(0, 50),
        href: e.href
      }))
    );

    const snapshot = await page.accessibility.snapshot();

    return JSON.stringify({
      url: page.url(),
      inputs,
      buttons,
      links,
      snapshot
    }, null, 2);
  "
}'
Weekly Installs
3
First Seen
Feb 26, 2026
Installed on
opencode3
gemini-cli3
github-copilot3
codex3
amp3
cline3