Bug Fix Workflow

Fix bugs with regression prevention.

Agents Used

1. test-writer - Write failing test for bug
2. code-fixer - Implement the fix
3. test-runner - Verify fix works
4. browser-qa-agent - Confirm in browser (if UI bug)

Trigger

• Manual: When bug is reported

Execution

claude "Run bug-fix workflow:

Bug: [DESCRIPTION]
Steps to reproduce:
1. [Step 1]
2. [Step 2]
Expected: [What should happen]
Actual: [What happens]

1. test-writer: Create test that reproduces the bug
2. code-fixer: Fix the bug
3. test-runner: Verify test passes (and no regressions)
4. browser-qa-agent: Confirm fix in browser"

Agent Chain

┌─────────────────────────────────────────────────┐
│  1. test-writer                                 │
│     - Create test that fails (reproduces bug)   │
│     - Test should pass after fix                │
└─────────────────────────────────────────────────┘
                      ↓
┌─────────────────────────────────────────────────┐
│  2. code-fixer                                  │
│     - Find root cause                           │
│     - Implement minimal fix                     │
│     - Don't refactor unrelated code             │
└─────────────────────────────────────────────────┘
                      ↓
┌─────────────────────────────────────────────────┐
│  3. test-runner                                 │
│     - New test passes                           │
│     - All existing tests pass                   │
│     - No regressions introduced                 │
└─────────────────────────────────────────────────┘
                      ↓
┌─────────────────────────────────────────────────┐
│  4. browser-qa-agent (if applicable)            │
│     - Reproduce original steps                  │
│     - Confirm bug is fixed                      │
│     - Check related features still work         │
└─────────────────────────────────────────────────┘

Example Usage

API Bug

claude "Run bug-fix workflow:

Bug: Login fails with valid credentials
Steps to reproduce:
1. Go to /login
2. Enter valid email/password
3. Click submit

Expected: Redirect to dashboard
Actual: 500 error, 'Cannot read property email of undefined'

Location: Likely in src/api/auth/login.ts"

UI Bug

claude "Run bug-fix workflow:

Bug: Modal doesn't close on backdrop click
Steps to reproduce:
1. Open any modal
2. Click outside modal (on backdrop)

Expected: Modal closes
Actual: Modal stays open

Location: src/components/Modal.tsx"

Expected Output

Phase 1: test-writer

Created test: src/api/auth/login.test.ts

it('should login with valid credentials', async () => {
  const { req, res } = createMocks({
    method: 'POST',
    body: { email: 'test@example.com', password: 'password123' },
  });

  await handler(req, res);

  expect(res._getStatusCode()).toBe(200);
});

Test status: FAILING (as expected - reproduces bug)

Phase 2: code-fixer

Root cause: user object destructured before null check

Fixed in: src/api/auth/login.ts:23

- const { email } = user;  // Fails if user is null
+ const email = user?.email;
+ if (!user) return res.status(401).json({ error: 'Invalid credentials' });

Phase 3: test-runner

Tests: 48 passed, 0 failed
New test: PASSING
No regressions detected

Phase 4: browser-qa-agent

Verification:
1. Navigated to /login
2. Entered valid credentials
3. Clicked submit
4. ✅ Redirected to dashboard
5. ✅ No console errors

Best Practices

1. Reproduce first - Confirm you can trigger the bug
2. Test captures bug - Test fails before fix, passes after
3. Minimal fix - Don't refactor, just fix
4. No regressions - All existing tests must pass
5. Verify in context - Test in real browser/environment

Skip browser-qa for

• Non-UI bugs (API, background jobs)
• Build/compilation issues
• Test infrastructure bugs