Fixing Bugs

Never skip tracking or regression test. Use debugging-with-tools for investigation, test-driven-development for fix.

FORBIDDEN: Fix without bd issue, fix without regression test REQUIRED: Every bug gets tracked, tested, verified before closing

Production emergencies: Abbreviated workflow OK (hotfix first), but still create bd issue and add regression tests afterward.

1. Create bd Bug Issue

Track from the start:

bd create "Bug: [Clear description]" --type bug --priority P1
# Returns: bd-123

Document:

bd edit bd-123 --design "
## Bug Description
[What's wrong]

## Reproduction Steps
1. Step one
2. Step two

## Expected Behavior
[What should happen]

## Actual Behavior
[What actually happens]

## Environment
[Version, OS, etc.]"

2. Debug Systematically

REQUIRED: Use debugging-with-tools skill

Use Skill tool: hyperpowers:debugging-with-tools

debugging-with-tools will:

• Use internet-researcher to search for error
• Recommend debugger or instrumentation
• Use codebase-investigator to understand context
• Guide to root cause (not symptom)

Update bd issue with findings:

bd edit bd-123 --design "[previous content]

## Investigation
[Root cause found via debugging]
[Tools used: debugger, internet search, etc.]"

3. Write Failing Test (RED Phase)

REQUIRED: Use test-driven-development skill

Write test that reproduces the bug:

def test_rejects_empty_email():
    """Regression test for bd-123: Empty email accepted"""
    with pytest.raises(ValidationError):
        create_user(email="")  # Should fail, currently passes

Run test, verify it FAILS:

pytest tests/test_user.py::test_rejects_empty_email
# Expected: PASS (bug exists)
# Should fail AFTER fix

Why critical: If test passes before fix, it doesn't test the bug.

4. Implement Fix (GREEN Phase)

Fix the root cause (not symptom):

def create_user(email: str):
    if not email or not email.strip():  # Fix
        raise ValidationError("Email required")
    # ... rest

Run test, verify it now FAILS (test was written backwards by mistake earlier - fix this):

Actually write the test to FAIL first:

def test_rejects_empty_email():
    with pytest.raises(ValidationError):
        create_user(email="")

Run:

pytest tests/test_user.py::test_rejects_empty_email
# Should FAIL before fix (no validation)
# Should PASS after fix (validation added)

5. Verify Complete Fix

REQUIRED: Use verification-before-completion skill

# Run full test suite (via test-runner agent)
"Run: pytest"

# Agent returns: All tests pass (including regression test)

Verify:

• Regression test passes
• All other tests still pass
• No new warnings or errors
• Pre-commit hooks pass

6. Close bd Issue

Update with fix details:

bd edit bd-123 --design "[previous content]

## Fix Implemented
[Description of fix]
[File changed: src/auth/user.py:23]

## Regression Test
[Test added: tests/test_user.py::test_rejects_empty_email]

## Verification
[All tests pass: 145/145]"

bd close bd-123

Commit with bd reference:

git commit -m "fix(bd-123): Reject empty email in user creation

Adds validation to prevent empty strings.
Regression test: test_rejects_empty_email

Closes bd-123"

"Fixes" immediately:

def create_user(email: str):
    if not email:  # Quick fix
        raise ValidationError("Email required")

Commits: "fix: validate email"

[No bd issue, no regression test]

No regression test:

• Bug could come back in future
• Can't prove fix works
• No protection against breaking this again

Incomplete fix:

• Doesn't handle email=" " (whitespace)
• Didn't debug to understand full issue

Result: Bug returns when someone changes validation logic.

# 1. Track
bd create "Bug: Empty email accepted" --type bug
# Returns: bd-123

# 2. Debug (use debugging-with-tools)
# Investigation reveals: Email validation missing entirely
# Also: Whitespace emails like " " also accepted

# 3. Write failing test (RED)
def test_rejects_empty_email():
    with pytest.raises(ValidationError):
        create_user(email="")

def test_rejects_whitespace_email():
    with pytest.raises(ValidationError):
        create_user(email="   ")

# Run: Both PASS (bug exists) - WAIT, test should FAIL before fix!

Actually:

# Test currently PASSES (bug exists - no validation)
# We expect test to FAIL after we add validation

# 4. Fix
def create_user(email: str):
    if not email or not email.strip():
        raise ValidationError("Email required")

# 5. Verify
pytest  # All tests pass now, including regression tests

# 6. Close
bd close bd-123
git commit -m "fix(bd-123): Reject empty/whitespace email"

Result: Bug fixed, tracked, tested, won't regress.

# Fix first
def validate_email(email):
    return "@" in email and len(email) > 0

# Then test
def test_validate_email():
    assert validate_email("user@example.com") == True

Test runs: PASS

Commits both together.

Later, someone changes validation:

def validate_email(email):
    return True  # Breaks validation!

Test still PASSES (only checks happy path).

Why it happens: Skipping TDD RED phase.

# 1. Write test FIRST that reproduces the bug
def test_validate_email():
    # Happy path
    assert validate_email("user@example.com") == True
    # Bug case (empty email was accepted)
    assert validate_email("") == False
    # Edge case discovered during debugging
    assert validate_email("@@") == False

# 2. Run test - should FAIL (bug exists)
pytest test_validate_email
# FAIL: validate_email("") returned True, expected False

# 3. Implement fix
def validate_email(email):
    if not email or len(email) == 0:
        return False
    return "@" in email and email.count("@") == 1

# 4. Run test - should PASS
pytest test_validate_email
# PASS: All assertions pass

Later regression:

def validate_email(email):
    return True  # Someone breaks it

Test catches it:

FAIL: assert validate_email("") == False
Expected False, got True

Result: Regression test actually prevents bug from returning.

Error:

NullPointerException at UserService.java:45

Developer sees line 45:

String email = user.getEmail().toLowerCase();  // Line 45

"Obvious fix":

String email = user.getEmail() != null ? user.getEmail().toLowerCase() : "";

Bug "fixed"... but crashes continue with different data.

Actual root cause: User object created without email in registration flow. Email is null for all users created via broken endpoint.

Result: Null-check applied everywhere, root cause (broken registration) unfixed.

# Dispatch internet-researcher
"Search for: NullPointerException UserService getEmail
- Common causes of null email in user objects
- User registration validation patterns"

# Dispatch codebase-investigator
"Investigate:
- How is User object created?
- Where is email set?
- Are there paths where email can be null?
- Which endpoints create users?"

# Agent reports:
"Found: POST /register endpoint creates User without validating email field.
Email is optional in UserDTO but required in User domain object."

Root cause found: Registration doesn't validate email.

Proper fix:

// In registration endpoint
@PostMapping("/register")
public User register(@RequestBody UserDTO dto) {
    if (dto.getEmail() == null || dto.getEmail().isEmpty()) {
        throw new ValidationException("Email required");
    }
    return userService.create(dto);
}

Regression test:

@Test
void registrationRequiresEmail() {
    assertThrows(ValidationException.class, () ->
        register(new UserDTO(null, "password")));
}

Result: Root cause fixed, no more null emails created.

Rules That Have No Exceptions

1. Every bug gets a bd issue → Track from discovery to closure

   • Create bd issue before fixing
   • Document reproduction steps
   • Update with investigation findings
   • Close only after verified

2. Use debugging-with-tools skill → Systematic investigation required

   • Never guess at fixes
   • Use internet-researcher for errors
   • Use debugger/instrumentation for state
   • Find root cause, not symptom

3. Write failing test first (RED) → Regression prevention

   • Test must fail before fix
   • Test must reproduce the bug
   • Test must pass after fix
   • If test passes immediately, it doesn't test the bug

4. Verify complete fix → Use verification-before-completion

   • Regression test passes
   • Full test suite passes
   • No new warnings
   • Pre-commit hooks pass

Common Excuses

All of these mean: Stop, follow complete workflow:

• "Quick fix, no need for bd issue"
• "Obvious bug, no need to debug"
• "I'll add test later"
• "Test passes, must be fixed"
• "Just one line change"

Before claiming bug fixed:

• bd issue created with reproduction steps
• Used debugging-with-tools to find root cause
• Wrote test that reproduces bug (RED phase)
• Verified test FAILS before fix
• Implemented fix addressing root cause
• Verified test PASSES after fix
• Ran full test suite (all pass)
• Updated bd issue with fix details
• Closed bd issue
• Committed with bd reference

This skill calls:

• debugging-with-tools (systematic investigation)
• test-driven-development (RED-GREEN-REFACTOR cycle)
• verification-before-completion (verify complete fix)

This skill is called by:

• When bugs discovered during development
• When test failures need fixing
• When user reports bugs

Agents used:

• hyperpowers:internet-researcher (via debugging-with-tools)
• hyperpowers:codebase-investigator (via debugging-with-tools)
• hyperpowers:test-runner (run tests without output pollution)

When stuck:

• Don't understand bug → Use debugging-with-tools skill
• Tempted to skip tracking → Create bd issue first, always
• Test passes immediately → Not testing the bug, rewrite test
• Fix doesn't work → Return to debugging-with-tools, find actual root cause