High-Level Orchestration Flow

Why 3 Subagents Instead of One?

Running all 5 modernization phases inside a single AI context causes catastrophic context bloat. Thousands of lines of JaCoCo coverage matrices, Surefire XML outputs, PITest bytecode mutant logs, and terminal stack traces saturate token windows — causing hallucinations and degraded code quality. SpringPhoenix isolates each phase into a dedicated Bob 2.0 subagent with a clean, sandboxed workspace and focused tool access.

What Runs Here

Baseline Snapshot — reports/baseline.json

{
  "module":          "com.example.legacy",
  "captured_at":     "2026-09-27T05:44:00Z",
  "java_version":    "8",
  "spring_boot":     "2.7.x",
  "line_coverage":   0.0,
  "branch_coverage": 0.0,
  "lines_covered":   0,
  "lines_total":     46,
  "tests_count":     0,
  "tests_passed":    0,
  "build_status":    "compiled_no_tests"
}

Legacy Source — UserService.java (before modernization)

@Service
public class UserService {
    private final UserRepository userRepository;

    // No @Transactional annotations — any method can corrupt state
    // No updateUser() method — update requires delete + create

    public List<User> getAllUsers() {
        return userRepository.findAll();
    }
    public Optional<User> getUserById(Long id) {
        return userRepository.findById(id);
    }
    public User saveUser(User user) {
        return userRepository.save(user);
    }
    public void deleteUser(Long id) {
        // No existence check — silent no-op on missing ID (flagged as gap)
        userRepository.deleteById(id);
    }
}

Coverage Gate — Line Coverage

Gate threshold marker at 80%. Pipeline would have looped back to Qwen for additional tests if below threshold.

Generated Test Files

Sample Generated Test — UserServiceTest.java

@ExtendWith(MockitoExtension.class)
class UserServiceTest {

    @Mock UserRepository userRepository;
    @InjectMocks UserService userService;

    @Test
    void getAllUsers_returnsAllFromRepository() {
        var users = List.of(new User(1L, "alice", "alice@example.com"));
        when(userRepository.findAll()).thenReturn(users);

        assertThat(userService.getAllUsers()).isEqualTo(users);
        verify(userRepository).findAll();
    }

    @Test
    void deleteUser_callsDeleteById() {
        userService.deleteUser(42L);
        verify(userRepository).deleteById(42L);
    }
}

Proposed Modernization Plan

Flagged Items for Review

Modernization Changes Applied

Self-Heal Loop — Status

If mvn test had failed, the self-heal loop would parse the terminal stack trace, call Qwen with the error context, patch the failing file, and re-run tests — up to 3 times per file before escalating to a human flag.

Modernized UserService.java

@Service
@Transactional(readOnly = true)   // ← added: disables dirty-check on reads
public class UserService {

    private final UserRepository userRepository;

    public UserService(UserRepository userRepository) {
        this.userRepository = userRepository;
    }

    public List<User> getAllUsers()             { return userRepository.findAll(); }
    public Optional<User> getUserById(Long id) { return userRepository.findById(id); }

    @Transactional
    public User saveUser(User user)             { return userRepository.save(user); }

    @Transactional                              // ← new method: was missing in legacy
    public Optional<User> updateUser(Long id, User userDetails) {
        return userRepository.findById(id).map(existing -> {
            existing.setUsername(userDetails.getUsername());
            existing.setEmail(userDetails.getEmail());
            return userRepository.save(existing);
        });
    }

    @Transactional
    public void deleteUser(Long id)             { userRepository.deleteById(id); }
}

Mutation Score Progress

Mutation Categories Tested

Sample Assertion — Mutation-Proof

@Test
void deleteUser_callsDeleteById() {
    userService.deleteUser(42L);
    // assertThat strengthened to verify exact argument — catches void-removal mutants
    verify(userRepository, times(1)).deleteById(42L);
    verifyNoMoreInteractions(userRepository);
}

@Test
void updateUser_returnsEmpty_whenNotFound() {
    when(userRepository.findById(99L)).thenReturn(Optional.empty());
    Optional<User> result = userService.updateUser(99L, new User());
    // assertThat value assertion — catches return-value mutants
    assertThat(result).isEmpty();
    verify(userRepository).findById(99L);
    verify(userRepository, never()).save(any());
}

Before vs. After — Three-State Comparison

Test Suite Breakdown

How Gaps Are Detected

All three subagents run a parallel business-logic scanner alongside their primary task. If a method's purpose isn't inferable from its name, comments, or surrounding context, it is flagged — never silently documented with a guess. Flagged items land in reports/knowledge_gaps.md with file path, method name, why it's unclear, and a suggested reviewer question.