pdd-cli 0.0.45__py3-none-any.whl → 0.0.90__py3-none-any.whl

pdd/sync_orchestration.py

@@ -12,21 +12,32 @@ import subprocess
 import re
 import os
 from pathlib import Path
-from typing import Dict, Any, Optional, List
-from dataclasses import asdict
+from typing import Dict, Any, Optional, List, Callable
+from dataclasses import asdict, dataclass, field
+import tempfile
+import sys
 
 import click
+import logging
+
+# --- Constants ---
+MAX_CONSECUTIVE_TESTS = 3  # Allow up to 3 consecutive test attempts
+MAX_TEST_EXTEND_ATTEMPTS = 2  # Allow up to 2 attempts to extend tests for coverage
+MAX_CONSECUTIVE_CRASHES = 3  # Allow up to 3 consecutive crash attempts (Bug #157 fix)
 
 # --- Real PDD Component Imports ---
-from .sync_animation import sync_animation
+from .sync_tui import SyncApp
 from .sync_determine_operation import (
     sync_determine_operation,
     get_pdd_file_paths,
     RunReport,
+    SyncDecision,
     PDD_DIR,
     META_DIR,
     SyncLock,
     read_run_report,
+    calculate_sha256,
+    calculate_current_hashes,
 )
 from .auto_deps_main import auto_deps_main
 from .code_generator_main import code_generator_main
@@ -37,6 +48,104 @@ from .cmd_test_main import cmd_test_main
 from .fix_main import fix_main
 from .update_main import update_main
 from .python_env_detector import detect_host_python_executable
+from .get_run_command import get_run_command_for_file
+from .pytest_output import extract_failing_files_from_output
+from . import DEFAULT_STRENGTH
+
+
+# --- Atomic State Update (Issue #159 Fix) ---
+
+@dataclass
+class PendingStateUpdate:
+    """Holds pending state updates for atomic commit."""
+    run_report: Optional[Dict[str, Any]] = None
+    fingerprint: Optional[Dict[str, Any]] = None
+    run_report_path: Optional[Path] = None
+    fingerprint_path: Optional[Path] = None
+
+
+class AtomicStateUpdate:
+    """
+    Context manager for atomic state updates.
+
+    Ensures run_report and fingerprint are both written or neither is written.
+    This fixes Issue #159 where non-atomic writes caused state desynchronization.
+
+    Usage:
+        with AtomicStateUpdate(basename, language) as state:
+            state.set_run_report(report_dict, report_path)
+            state.set_fingerprint(fingerprint_dict, fp_path)
+        # On successful exit, both files are written atomically
+        # On exception, neither file is written (rollback)
+    """
+
+    def __init__(self, basename: str, language: str):
+        self.basename = basename
+        self.language = language
+        self.pending = PendingStateUpdate()
+        self._temp_files: List[str] = []
+
+    def __enter__(self):
+        return self
+
+    def __exit__(self, exc_type, exc_val, exc_tb):
+        if exc_type is None:
+            self._commit()
+        else:
+            self._rollback()
+        return False  # Don't suppress exceptions
+
+    def set_run_report(self, report: Dict[str, Any], path: Path):
+        """Buffer a run report for atomic write."""
+        self.pending.run_report = report
+        self.pending.run_report_path = path
+
+    def set_fingerprint(self, fingerprint: Dict[str, Any], path: Path):
+        """Buffer a fingerprint for atomic write."""
+        self.pending.fingerprint = fingerprint
+        self.pending.fingerprint_path = path
+
+    def _atomic_write(self, data: Dict[str, Any], target_path: Path) -> None:
+        """Write data to file atomically using temp file + rename pattern."""
+        target_path.parent.mkdir(parents=True, exist_ok=True)
+
+        # Write to temp file in same directory (required for atomic rename)
+        fd, temp_path = tempfile.mkstemp(
+            dir=target_path.parent,
+            prefix=f".{target_path.stem}_",
+            suffix=".tmp"
+        )
+        self._temp_files.append(temp_path)
+
+        try:
+            with os.fdopen(fd, 'w') as f:
+                json.dump(data, f, indent=2, default=str)
+
+            # Atomic rename - guaranteed atomic on POSIX systems
+            os.replace(temp_path, target_path)
+            self._temp_files.remove(temp_path)  # Successfully moved, stop tracking
+        except Exception:
+            # Leave temp file for rollback to clean up
+            raise
+
+    def _commit(self):
+        """Commit all pending state updates atomically."""
+        # Write fingerprint first (checkpoint), then run_report
+        if self.pending.fingerprint and self.pending.fingerprint_path:
+            self._atomic_write(self.pending.fingerprint, self.pending.fingerprint_path)
+        if self.pending.run_report and self.pending.run_report_path:
+            self._atomic_write(self.pending.run_report, self.pending.run_report_path)
+
+    def _rollback(self):
+        """Clean up any temp files without committing changes."""
+        for temp_path in self._temp_files:
+            try:
+                if os.path.exists(temp_path):
+                    os.unlink(temp_path)
+            except OSError:
+                pass  # Best effort cleanup
+        self._temp_files.clear()
+
 
 # --- Mock Helper Functions ---
 
@@ -98,20 +207,44 @@ def log_sync_event(basename: str, language: str, event: str, details: Dict[str,
     }
     append_sync_log(basename, language, entry)
 
-def save_run_report(report: Dict[str, Any], basename: str, language: str):
-    """Save a run report to the metadata directory."""
+def save_run_report(report: Dict[str, Any], basename: str, language: str,
+                    atomic_state: Optional['AtomicStateUpdate'] = None):
+    """Save a run report to the metadata directory.
+
+    Args:
+        report: The run report dictionary to save.
+        basename: The module basename.
+        language: The programming language.
+        atomic_state: Optional AtomicStateUpdate for atomic writes (Issue #159 fix).
+    """
     report_file = META_DIR / f"{basename}_{language}_run.json"
-    META_DIR.mkdir(parents=True, exist_ok=True)
-    with open(report_file, 'w') as f:
-        json.dump(report, f, indent=2, default=str)
+    if atomic_state:
+        # Buffer for atomic write
+        atomic_state.set_run_report(report, report_file)
+    else:
+        # Legacy direct write
+        META_DIR.mkdir(parents=True, exist_ok=True)
+        with open(report_file, 'w') as f:
+            json.dump(report, f, indent=2, default=str)
+
+def _save_operation_fingerprint(basename: str, language: str, operation: str,
+                               paths: Dict[str, Path], cost: float, model: str,
+                               atomic_state: Optional['AtomicStateUpdate'] = None):
+    """Save fingerprint state after successful operation.
 
-def _save_operation_fingerprint(basename: str, language: str, operation: str, 
-                               paths: Dict[str, Path], cost: float, model: str):
-    """Save fingerprint state after successful operation."""
+    Args:
+        basename: The module basename.
+        language: The programming language.
+        operation: The operation that was performed.
+        paths: Dictionary of PDD file paths.
+        cost: The cost of the operation.
+        model: The model used.
+        atomic_state: Optional AtomicStateUpdate for atomic writes (Issue #159 fix).
+    """
     from datetime import datetime, timezone
     from .sync_determine_operation import calculate_current_hashes, Fingerprint
     from . import __version__
-    
+
     current_hashes = calculate_current_hashes(paths)
     fingerprint = Fingerprint(
         pdd_version=__version__,
@@ -120,103 +253,544 @@ def _save_operation_fingerprint(basename: str, language: str, operation: str,
         prompt_hash=current_hashes.get('prompt_hash'),
         code_hash=current_hashes.get('code_hash'),
         example_hash=current_hashes.get('example_hash'),
-        test_hash=current_hashes.get('test_hash')
+        test_hash=current_hashes.get('test_hash'),
+        test_files=current_hashes.get('test_files'),  # Bug #156
     )
-    
-    META_DIR.mkdir(parents=True, exist_ok=True)
+
     fingerprint_file = META_DIR / f"{basename}_{language}.json"
-    with open(fingerprint_file, 'w') as f:
-        json.dump(asdict(fingerprint), f, indent=2, default=str)
+    if atomic_state:
+        # Buffer for atomic write
+        atomic_state.set_fingerprint(asdict(fingerprint), fingerprint_file)
+    else:
+        # Legacy direct write
+        META_DIR.mkdir(parents=True, exist_ok=True)
+        with open(fingerprint_file, 'w') as f:
+            json.dump(asdict(fingerprint), f, indent=2, default=str)
 
-# SyncLock class now imported from sync_determine_operation module
+def _python_cov_target_for_code_file(code_file: Path) -> str:
+    """Return a `pytest-cov` `--cov` target for a Python code file.
+
+    - If the file is inside a Python package (directories with `__init__.py`),
+      returns a dotted module path (e.g., `pdd.sync_orchestration`).
+    - Otherwise falls back to the filename stem (e.g., `admin_get_users`).
+    """
+    if code_file.suffix != ".py":
+        return code_file.stem
+
+    package_dir: Optional[Path] = None
+    current = code_file.parent
+    while (current / "__init__.py").exists():
+        package_dir = current
+        parent = current.parent
+        if parent == current:
+            break
+        current = parent
+
+    if package_dir:
+        relative_module = code_file.relative_to(package_dir.parent).with_suffix("")
+        return str(relative_module).replace(os.sep, ".")
+
+    return code_file.stem
+
+
+def _python_cov_target_for_test_and_code(test_file: Path, code_file: Path, fallback: str) -> str:
+    """Choose the best `--cov` target based on how tests import the code.
+
+    In some repos, tests add a directory to `sys.path` and import modules by their
+    filename stem (e.g., `from admin_get_users import ...`) even when the code
+    also lives under a package (e.g., `backend.functions.admin_get_users`).
+
+    Heuristic:
+    - Prefer the code file stem when the test file imports it directly.
+    - Otherwise, prefer the dotted module path derived from the package layout.
+    - Fall back to the provided fallback (usually the basename).
+    """
+
+    def _imports_module(source: str, module: str) -> bool:
+        escaped = re.escape(module)
+        return bool(
+            re.search(rf"^\s*import\s+{escaped}\b", source, re.MULTILINE)
+            or re.search(rf"^\s*from\s+{escaped}\b", source, re.MULTILINE)
+        )
+
+    stem = code_file.stem
+    dotted = _python_cov_target_for_code_file(code_file)
 
-def _execute_tests_and_create_run_report(test_file: Path, basename: str, language: str, target_coverage: float = 90.0) -> RunReport:
-    """Execute tests and create a RunReport with actual results."""
-    timestamp = datetime.datetime.now(datetime.timezone.utc).isoformat()
-    
     try:
-        # Execute pytest with coverage reporting on the specific module
-        # Extract module name from test file (e.g., test_factorial.py -> factorial)
-        module_name = test_file.name.replace('test_', '').replace('.py', '')
-        
-        # Use the module import path rather than file path for coverage
-        # Use environment-aware Python executable for pytest execution
-        python_executable = detect_host_python_executable()
-        
-        # Determine coverage target based on module location
-        if base_package:
-            cov_target = f'{base_package}.{module_name}'
-        else:
-            # Dynamically discover package structure based on test file location
-            relative_path = test_file.parent.relative_to(Path.cwd())
-            package_path = str(relative_path).replace(os.sep, '.')
-            cov_target = f'{package_path}.{module_name}' if package_path else module_name
-        
-        result = subprocess.run([
-            python_executable, '-m', 'pytest', 
-            str(test_file), 
-            '-v', 
-            '--tb=short',
-            f'--cov={cov_target}',
-            '--cov-report=term-missing'
-        ], capture_output=True, text=True, timeout=300)
-        
-        exit_code = result.returncode
-        stdout = result.stdout
-        stderr = result.stderr
-        
-        # Parse test results from pytest output
-        tests_passed = 0
-        tests_failed = 0
-        coverage = 0.0
-        
-        # Parse passed/failed tests
-        if 'passed' in stdout:
-            passed_match = re.search(r'(\d+) passed', stdout)
+        test_source = test_file.read_text(encoding="utf-8", errors="ignore")
+    except Exception:
+        test_source = ""
+
+    if stem and _imports_module(test_source, stem):
+        return stem
+
+    if dotted and dotted != stem:
+        if _imports_module(test_source, dotted):
+            return dotted
+
+        if "." in dotted:
+            parent = dotted.rsplit(".", 1)[0]
+            # e.g. `from backend.functions import admin_get_users`
+            if re.search(
+                rf"^\s*from\s+{re.escape(parent)}\s+import\s+.*\b{re.escape(stem)}\b",
+                test_source,
+                re.MULTILINE,
+            ):
+                return dotted
+            # e.g. `import backend.functions.admin_get_users`
+            if re.search(
+                rf"^\s*import\s+{re.escape(parent)}\.{re.escape(stem)}\b",
+                test_source,
+                re.MULTILINE,
+            ):
+                return dotted
+
+        return dotted
+
+    return stem or fallback
+
+
+def _parse_test_output(output: str, language: str) -> tuple[int, int, float]:
+    """
+    Parse test output to extract passed/failed/coverage.
+
+    Args:
+        output: Combined stdout/stderr from test runner
+        language: Language name (e.g., 'python', 'typescript', 'go')
+
+    Returns:
+        (tests_passed, tests_failed, coverage)
+    """
+    tests_passed = 0
+    tests_failed = 0
+    coverage = 0.0
+
+    lang = language.lower()
+
+    # Python (pytest)
+    if lang == 'python':
+        if 'passed' in output:
+            passed_match = re.search(r'(\d+) passed', output)
             if passed_match:
                 tests_passed = int(passed_match.group(1))
-        
-        if 'failed' in stdout:
-            failed_match = re.search(r'(\d+) failed', stdout)
+        if 'failed' in output:
+            failed_match = re.search(r'(\d+) failed', output)
             if failed_match:
                 tests_failed = int(failed_match.group(1))
-        
-        # Parse coverage percentage - try multiple patterns
-        coverage_match = re.search(r'TOTAL.*?(\d+)%', stdout)
+        if 'error' in output:
+            error_match = re.search(r'(\d+) error', output)
+            if error_match:
+                tests_failed += int(error_match.group(1))
+        coverage_match = re.search(r'TOTAL.*?(\d+)%', output)
         if not coverage_match:
-            # Try alternative patterns for coverage output
-            coverage_match = re.search(r'(\d+)%\s*$', stdout, re.MULTILINE)
+            coverage_match = re.search(r'(\d+)%\s*$', output, re.MULTILINE)
         if not coverage_match:
-            # Try pattern with decimal
-            coverage_match = re.search(r'(\d+(?:\.\d+)?)%', stdout)
-        
+            coverage_match = re.search(r'(\d+(?:\.\d+)?)%', output)
         if coverage_match:
             coverage = float(coverage_match.group(1))
-        
-        # Create and save run report
+
+    # Jest/Vitest (JavaScript/TypeScript)
+    elif lang in ('javascript', 'typescript', 'typescriptreact'):
+        # "Tests: X passed, Y failed" or "Tests: X passed, Y failed, Z total"
+        match = re.search(r'Tests:\s*(\d+)\s+passed', output)
+        if match:
+            tests_passed = int(match.group(1))
+        match = re.search(r'Tests:.*?(\d+)\s+failed', output)
+        if match:
+            tests_failed = int(match.group(1))
+
+        # Alternative Mocha-style: "X passing, Y failing"
+        if tests_passed == 0:
+            pass_match = re.search(r'(\d+)\s+pass(?:ing)?', output, re.I)
+            if pass_match:
+                tests_passed = int(pass_match.group(1))
+        if tests_failed == 0:
+            fail_match = re.search(r'(\d+)\s+fail(?:ing)?', output, re.I)
+            if fail_match:
+                tests_failed = int(fail_match.group(1))
+
+        # Coverage: "All files | XX.XX |"
+        cov_match = re.search(r'All files[^|]*\|\s*(\d+\.?\d*)', output)
+        if cov_match:
+            coverage = float(cov_match.group(1))
+
+    # Go
+    elif lang == 'go':
+        # Count PASS and FAIL occurrences for individual tests
+        tests_passed = len(re.findall(r'--- PASS:', output))
+        tests_failed = len(re.findall(r'--- FAIL:', output))
+
+        # Fallback: check for overall PASS/FAIL
+        if tests_passed == 0 and 'PASS' in output and 'FAIL' not in output:
+            tests_passed = 1
+        if tests_failed == 0 and 'FAIL' in output:
+            tests_failed = 1
+
+        # coverage: XX.X% of statements
+        cov_match = re.search(r'coverage:\s*(\d+\.?\d*)%', output)
+        if cov_match:
+            coverage = float(cov_match.group(1))
+
+    # Rust (cargo test)
+    elif lang == 'rust':
+        # "test result: ok. X passed; Y failed;"
+        match = re.search(r'(\d+)\s+passed', output)
+        if match:
+            tests_passed = int(match.group(1))
+        match = re.search(r'(\d+)\s+failed', output)
+        if match:
+            tests_failed = int(match.group(1))
+
+    # Fallback: try generic patterns
+    else:
+        pass_match = re.search(r'(\d+)\s+(?:tests?\s+)?pass(?:ed)?', output, re.I)
+        fail_match = re.search(r'(\d+)\s+(?:tests?\s+)?fail(?:ed)?', output, re.I)
+        if pass_match:
+            tests_passed = int(pass_match.group(1))
+        if fail_match:
+            tests_failed = int(fail_match.group(1))
+
+    return tests_passed, tests_failed, coverage
+
+
+def _detect_example_errors(output: str) -> tuple[bool, str]:
+    """
+    Detect if example output contains error indicators.
+
+    Only detects true crashes/errors:
+    - Python tracebacks (catches ALL unhandled exceptions)
+    - ERROR level log messages
+
+    Intentionally does NOT detect:
+    - HTTP status codes (examples may test error responses)
+    - Individual exception type names (causes false positives, redundant with traceback)
+
+    Returns:
+        (has_errors, error_summary)
+    """
+    error_patterns = [
+        (r'Traceback \(most recent call last\):', 'Python traceback'),
+        (r' - ERROR - ', 'Error log message'),  # Python logging format
+    ]
+
+    errors_found = []
+    for pattern, description in error_patterns:
+        if re.search(pattern, output, re.MULTILINE):
+            errors_found.append(description)
+
+    if errors_found:
+        return True, '; '.join(errors_found)
+    return False, ''
+
+
+def _try_auto_fix_import_error(
+    error_output: str,
+    code_file: Path,
+    example_file: Path,
+) -> tuple[bool, str]:
+    """
+    Try to automatically fix common import errors before calling expensive agentic fix.
+
+    Returns:
+        (fixed, message): Whether a fix was attempted and what was done.
+    """
+    import re
+
+    # Check for ModuleNotFoundError or ImportError
+    module_not_found = re.search(r"ModuleNotFoundError: No module named ['\"]([^'\"]+)['\"]", error_output)
+    import_error = re.search(r"ImportError: cannot import name ['\"]([^'\"]+)['\"]", error_output)
+
+    if not module_not_found and not import_error:
+        return False, "No import error detected"
+
+    if module_not_found:
+        missing_module = module_not_found.group(1)
+        # Split by . to get the top-level package
+        top_level_package = missing_module.split('.')[0]
+
+        # Check if this is the module we're trying to import (local module)
+        code_module_name = code_file.stem  # e.g., "data_validator" from "data_validator.py"
+
+        if top_level_package == code_module_name:
+            # It's trying to import our own generated code - fix the example's sys.path
+            # Read the example and fix the path manipulation
+            try:
+                example_content = example_file.read_text(encoding='utf-8')
+                code_dir = str(code_file.parent.resolve())
+
+                # Look for existing sys.path manipulation
+                if 'sys.path' in example_content:
+                    # Try to fix the existing path manipulation
+                    # Common pattern: module_path = os.path.abspath(os.path.join(...))
+                    # Replace with correct path
+                    fixed_content = re.sub(
+                        r"module_path\s*=\s*os\.path\.abspath\([^)]+\)",
+                        f"module_path = '{code_dir}'",
+                        example_content
+                    )
+                    if fixed_content != example_content:
+                        example_file.write_text(fixed_content, encoding='utf-8')
+                        return True, f"Fixed sys.path to point to {code_dir}"
+
+                # If no existing sys.path, add one at the start after imports
+                lines = example_content.split('\n')
+                insert_pos = 0
+                for i, line in enumerate(lines):
+                    if line.startswith('import ') or line.startswith('from '):
+                        if 'sys' in line or 'os' in line:
+                            insert_pos = i + 1
+                            continue
+                    if line.strip() and not line.startswith('#') and not line.startswith('import') and not line.startswith('from'):
+                        insert_pos = i
+                        break
+
+                path_fix = f"\n# Auto-added by pdd to fix import\nimport sys\nsys.path.insert(0, '{code_dir}')\n"
+                lines.insert(insert_pos, path_fix)
+                example_file.write_text('\n'.join(lines), encoding='utf-8')
+                return True, f"Added sys.path.insert(0, '{code_dir}') to example"
+
+            except Exception as e:
+                return False, f"Failed to fix import path: {e}"
+
+        else:
+            # It's an external package - try pip install
+            try:
+                result = subprocess.run(
+                    [sys.executable, '-m', 'pip', 'install', top_level_package],
+                    capture_output=True,
+                    text=True,
+                    timeout=120
+                )
+                if result.returncode == 0:
+                    return True, f"Installed missing package: {top_level_package}"
+                else:
+                    return False, f"Failed to install {top_level_package}: {result.stderr}"
+            except Exception as e:
+                return False, f"Failed to run pip install: {e}"
+
+    return False, "Import error detected but no auto-fix available"
+
+
+def _run_example_with_error_detection(
+    cmd_parts: list[str],
+    env: dict,
+    cwd: str,
+    timeout: int = 60
+) -> tuple[int, str, str]:
+    """
+    Run example file, detecting errors from output.
+
+    For server-style examples that block, this runs until timeout
+    then analyzes output for errors. No errors = success.
+
+    Returns:
+        (returncode, stdout, stderr)
+        - returncode: 0 if no errors detected, positive if errors found or process failed
+    """
+    import threading
+
+    proc = subprocess.Popen(
+        cmd_parts,
+        stdout=subprocess.PIPE,
+        stderr=subprocess.PIPE,
+        stdin=subprocess.DEVNULL,
+        env=env,
+        cwd=cwd,
+        start_new_session=True,
+    )
+
+    stdout_chunks = []
+    stderr_chunks = []
+
+    def read_pipe(pipe, chunks):
+        try:
+            for line in iter(pipe.readline, b''):
+                chunks.append(line)
+        except Exception:
+            pass
+
+    t_out = threading.Thread(target=read_pipe, args=(proc.stdout, stdout_chunks), daemon=True)
+    t_err = threading.Thread(target=read_pipe, args=(proc.stderr, stderr_chunks), daemon=True)
+    t_out.start()
+    t_err.start()
+
+    # Wait for process or timeout
+    try:
+        proc.wait(timeout=timeout)
+    except subprocess.TimeoutExpired:
+        proc.terminate()
+        try:
+            proc.wait(timeout=5)
+        except subprocess.TimeoutExpired:
+            proc.kill()
+            proc.wait()
+
+    t_out.join(timeout=2)
+    t_err.join(timeout=2)
+
+    stdout = b''.join(stdout_chunks).decode('utf-8', errors='replace')
+    stderr = b''.join(stderr_chunks).decode('utf-8', errors='replace')
+    combined = stdout + '\n' + stderr
+
+    # Check for errors in output
+    has_errors, error_summary = _detect_example_errors(combined)
+
+    # Determine result:
+    # - Errors in output → failure
+    # - Positive exit code (process failed normally, e.g., sys.exit(1)) → failure
+    # - Negative exit code (killed by signal, e.g., -9 for SIGKILL) → check output
+    # - Zero exit code → success
+    #
+    # IMPORTANT: When we kill the process after timeout, returncode is negative
+    # (the signal number). This is NOT a failure if output has no errors.
+    if has_errors:
+        return 1, stdout, stderr  # Errors detected in output
+    elif proc.returncode is not None and proc.returncode > 0:
+        return proc.returncode, stdout, stderr  # Process exited with error
+    else:
+        # Success cases:
+        # - returncode == 0 (clean exit)
+        # - returncode < 0 (killed by signal, but no errors in output)
+        # - returncode is None (shouldn't happen after wait, but safe fallback)
+        return 0, stdout, stderr
+
+
+def _execute_tests_and_create_run_report(
+    test_file: Path,
+    basename: str,
+    language: str,
+    target_coverage: float = 90.0,
+    *,
+    code_file: Optional[Path] = None,
+    atomic_state: Optional['AtomicStateUpdate'] = None,
+    test_files: Optional[List[Path]] = None,  # Bug #156: Support multiple test files
+) -> RunReport:
+    """Execute tests and create a RunReport with actual results.
+
+    Now supports multiple languages by using get_test_command_for_file()
+    to determine the appropriate test runner.
+
+    Args:
+        test_file: Primary test file (for backward compat)
+        test_files: Optional list of all test files to run (Bug #156)
+    """
+    from .get_test_command import get_test_command_for_file
+
+    timestamp = datetime.datetime.now(datetime.timezone.utc).isoformat()
+
+    # Bug #156: Use test_files if provided, otherwise just the single test_file
+    all_test_files = test_files if test_files else [test_file]
+
+    # Calculate test file hash for staleness detection (primary file for backward compat)
+    test_hash = calculate_sha256(test_file) if test_file.exists() else None
+
+    # Bug #156: Calculate hashes for ALL test files
+    test_file_hashes = {
+        f.name: calculate_sha256(f)
+        for f in all_test_files
+        if f.exists()
+    } if all_test_files else None
+
+    # Use clean env without TUI-specific vars
+    clean_env = os.environ.copy()
+    for var in ['FORCE_COLOR', 'COLUMNS']:
+        clean_env.pop(var, None)
+
+    try:
+        lang_lower = language.lower()
+
+        # Python: use existing pytest logic with coverage
+        if lang_lower == "python":
+            module_name = test_file.name.replace('test_', '').replace('.py', '')
+            python_executable = detect_host_python_executable()
+
+            cov_target = None
+            if code_file is not None:
+                cov_target = _python_cov_target_for_test_and_code(test_file, code_file, basename or module_name)
+            else:
+                cov_target = basename or module_name
+
+            if not cov_target:
+                cov_target = basename or module_name
+
+            # Bug #156: Run pytest on ALL test files
+            pytest_args = [
+                python_executable, '-m', 'pytest',
+            ] + [str(f) for f in all_test_files] + [
+                '-v',
+                '--tb=short',
+                f'--cov={cov_target}',
+                '--cov-report=term-missing'
+            ]
+            result = subprocess.run(
+                pytest_args,
+                capture_output=True, text=True, timeout=300, stdin=subprocess.DEVNULL, env=clean_env, start_new_session=True
+            )
+
+            exit_code = result.returncode
+            stdout = result.stdout + (result.stderr or '')
+            tests_passed, tests_failed, coverage = _parse_test_output(stdout, language)
+
+        else:
+            # Non-Python: use language-appropriate test command
+            test_cmd = get_test_command_for_file(str(test_file), language)
+
+            if test_cmd is None:
+                # No test command available - return report indicating this
+                report = RunReport(
+                    timestamp=timestamp,
+                    exit_code=127,  # Command not found
+                    tests_passed=0,
+                    tests_failed=0,
+                    coverage=0.0,
+                    test_hash=test_hash,
+                    test_files=test_file_hashes,  # Bug #156
+                )
+                save_run_report(asdict(report), basename, language, atomic_state)
+                return report
+
+            # Run the test command
+            result = subprocess.run(
+                test_cmd,
+                shell=True,
+                capture_output=True,
+                text=True,
+                timeout=300,
+                env=clean_env,
+                cwd=str(test_file.parent),
+                stdin=subprocess.DEVNULL,
+                start_new_session=True
+            )
+
+            exit_code = result.returncode
+            stdout = (result.stdout or '') + '\n' + (result.stderr or '')
+
+            # Parse results based on language
+            tests_passed, tests_failed, coverage = _parse_test_output(stdout, language)
+
         report = RunReport(
             timestamp=timestamp,
             exit_code=exit_code,
             tests_passed=tests_passed,
             tests_failed=tests_failed,
-            coverage=coverage
+            coverage=coverage,
+            test_hash=test_hash,
+            test_files=test_file_hashes,  # Bug #156
         )
-        
+
     except (subprocess.TimeoutExpired, subprocess.CalledProcessError, Exception) as e:
-        # If test execution fails, create a report indicating failure
         report = RunReport(
             timestamp=timestamp,
             exit_code=1,
             tests_passed=0,
             tests_failed=1,
-            coverage=0.0
+            coverage=0.0,
+            test_hash=test_hash,
+            test_files=test_file_hashes,  # Bug #156
         )
-    
-    # Save the run report
-    save_run_report(asdict(report), basename, language)
-    return report
 
-# --- Helper for Click Context ---
+    save_run_report(asdict(report), basename, language, atomic_state)
+    return report
 
 def _create_mock_context(**kwargs) -> click.Context:
     """Creates a mock Click context object to pass parameters to command functions."""
@@ -242,7 +816,6 @@ def _display_sync_log(basename: str, language: str, verbose: bool = False) -> Di
     for entry in log_entries:
         timestamp = entry.get('timestamp', 'N/A')
         
-        # Handle special event entries
         if 'event' in entry:
             event = entry.get('event', 'N/A')
             print(f"[{timestamp[:19]}] EVENT: {event}")
@@ -251,7 +824,6 @@ def _display_sync_log(basename: str, language: str, verbose: bool = False) -> Di
                 print(f"  Details: {details_str}")
             continue
         
-        # Handle operation entries
         operation = entry.get('operation', 'N/A')
         reason = entry.get('reason', 'N/A')
         success = entry.get('success')
@@ -260,7 +832,6 @@ def _display_sync_log(basename: str, language: str, verbose: bool = False) -> Di
         duration = entry.get('duration')
         
         if verbose:
-            # Verbose format
             print(f"[{timestamp[:19]}] {operation:<12} | {reason}")
             decision_type = entry.get('decision_type', 'N/A')
             confidence = entry.get('confidence', 'N/A')
@@ -276,14 +847,12 @@ def _display_sync_log(basename: str, language: str, verbose: bool = False) -> Di
                 print(f"  Estimated Cost: ${estimated_cost:.2f}")
             
             if 'details' in entry and entry['details']:
-                # Show details without budget_remaining to avoid clutter
                 details_copy = entry['details'].copy()
                 details_copy.pop('budget_remaining', None)
                 if details_copy:
                     details_str = json.dumps(details_copy, indent=2)
                     print(f"  Details: {details_str}")
         else:
-            # Normal format: [timestamp] operation | reason | status cost | duration
             status_icon = "✓" if success else "✗" if success is False else "?"
             
             cost_info = ""
@@ -318,45 +887,60 @@ def sync_orchestration(
     budget: float = 10.0,
     skip_verify: bool = False,
     skip_tests: bool = False,
-    log: bool = False,
+    dry_run: bool = False,
     force: bool = False,
-    strength: float = 0.5,
+    strength: float = DEFAULT_STRENGTH,
     temperature: float = 0.0,
-    time_param: float = 0.25, # Renamed to avoid conflict with `time` module
+    time_param: float = 0.25,
     verbose: bool = False,
     quiet: bool = False,
     output_cost: Optional[str] = None,
     review_examples: bool = False,
     local: bool = False,
     context_config: Optional[Dict[str, str]] = None,
+    context_override: Optional[str] = None,
+    confirm_callback: Optional[Callable[[str, str], bool]] = None,
 ) -> Dict[str, Any]:
     """
     Orchestrates the complete PDD sync workflow with parallel animation.
-
-    If log=True, displays the sync log instead of running sync operations.
-    The verbose flag controls the detail level of the log output.
-
-    Returns a dictionary summarizing the outcome of the sync process.
     """
-    if log:
+    # Import get_extension at function scope
+    from .sync_determine_operation import get_extension
+    
+    if dry_run:
         return _display_sync_log(basename, language, verbose)
 
     # --- Initialize State and Paths ---
     try:
-        pdd_files = get_pdd_file_paths(basename, language, prompts_dir)
+        pdd_files = get_pdd_file_paths(basename, language, prompts_dir, context_override=context_override)
+    except FileNotFoundError as e:
+        if "test_config.py" in str(e) or "tests/test_" in str(e):
+            pdd_files = {
+                'prompt': Path(prompts_dir) / f"{basename}_{language}.prompt",
+                'code': Path(f"src/{basename}.{get_extension(language)}"),
+                'example': Path(f"context/{basename}_example.{get_extension(language)}"),
+                'test': Path(f"tests/test_{basename}.{get_extension(language)}")
+            }
+            if not quiet:
+                print(f"Note: Test file missing, continuing with sync workflow to generate it")
+        else:
+            print(f"Error constructing paths: {e}")
+            return {
+                "success": False,
+                "error": f"Failed to construct paths: {str(e)}",
+                "operations_completed": [],
+                "errors": [f"Path construction failed: {str(e)}"]
+            }
     except Exception as e:
-        # Log the error and return early with failure status
         print(f"Error constructing paths: {e}")
         return {
             "success": False,
-            "total_cost": 0.0,
-            "model_name": "",
             "error": f"Failed to construct paths: {str(e)}",
             "operations_completed": [],
             "errors": [f"Path construction failed: {str(e)}"]
         }
     
-    # Shared state for animation thread
+    # Shared state for animation (passed to App)
     current_function_name_ref = ["initializing"]
     stop_event = threading.Event()
     current_cost_ref = [0.0]
@@ -364,696 +948,695 @@ def sync_orchestration(
     code_path_ref = [str(pdd_files.get('code', 'N/A'))]
     example_path_ref = [str(pdd_files.get('example', 'N/A'))]
     tests_path_ref = [str(pdd_files.get('test', 'N/A'))]
-    prompt_box_color_ref, code_box_color_ref, example_box_color_ref, tests_box_color_ref = \
-        ["blue"], ["blue"], ["blue"], ["blue"]
-    
-    # Orchestration state
-    operations_completed: List[str] = []
-    skipped_operations: List[str] = []
-    errors: List[str] = []
-    start_time = time.time()
-    animation_thread = None
-    
-    # Track operation history for cycle detection
-    operation_history: List[str] = []
-    MAX_CYCLE_REPEATS = 2  # Maximum times to allow crash-verify cycle
+    prompt_box_color_ref = ["blue"]
+    code_box_color_ref = ["blue"]
+    example_box_color_ref = ["blue"]
+    tests_box_color_ref = ["blue"]
 
-    try:
-        with SyncLock(basename, language):
-            # Log lock acquisition
-            log_sync_event(basename, language, "lock_acquired", {"pid": os.getpid()})
-            
-            # --- Start Animation Thread ---
-            animation_thread = threading.Thread(
-                target=sync_animation,
-                args=(
-                    current_function_name_ref, stop_event, basename, current_cost_ref, budget,
-                    prompt_box_color_ref, code_box_color_ref, example_box_color_ref, tests_box_color_ref,
-                    prompt_path_ref, code_path_ref, example_path_ref, tests_path_ref
-                ),
-                daemon=True
-            )
-            animation_thread.start()
-
-            # --- Main Workflow Loop ---
-            while True:
-                budget_remaining = budget - current_cost_ref[0]
-                if current_cost_ref[0] >= budget:
-                    errors.append(f"Budget of ${budget:.2f} exceeded.")
-                    log_sync_event(basename, language, "budget_exceeded", {
-                        "total_cost": current_cost_ref[0], 
-                        "budget": budget
-                    })
-                    break
-
-                # Log budget warning when running low
-                if budget_remaining < budget * 0.2 and budget_remaining > 0:
-                    log_sync_event(basename, language, "budget_warning", {
-                        "remaining": budget_remaining,
-                        "percentage": (budget_remaining / budget) * 100
-                    })
-
-                decision = sync_determine_operation(basename, language, target_coverage, budget_remaining, False, prompts_dir, skip_tests, skip_verify)
-                operation = decision.operation
-                
-                # Create log entry with decision info
-                log_entry = create_sync_log_entry(decision, budget_remaining)
-                
-                # Track operation history
-                operation_history.append(operation)
-                
-                # Detect crash-verify cycles
-                if len(operation_history) >= 4:
-                    # Check for repeating crash-verify pattern
-                    recent_ops = operation_history[-4:]
-                    if (recent_ops == ['crash', 'verify', 'crash', 'verify'] or
-                        recent_ops == ['verify', 'crash', 'verify', 'crash']):
-                        # Count how many times this cycle has occurred
-                        cycle_count = 0
-                        for i in range(0, len(operation_history) - 1, 2):
-                            if i + 1 < len(operation_history):
-                                if ((operation_history[i] == 'crash' and operation_history[i+1] == 'verify') or
-                                    (operation_history[i] == 'verify' and operation_history[i+1] == 'crash')):
-                                    cycle_count += 1
-                        
-                        if cycle_count >= MAX_CYCLE_REPEATS:
-                            errors.append(f"Detected crash-verify cycle repeated {cycle_count} times. Breaking cycle.")
-                            errors.append("The example file may have syntax errors that couldn't be automatically fixed.")
-                            log_sync_event(basename, language, "cycle_detected", {
-                                "cycle_type": "crash-verify",
-                                "cycle_count": cycle_count,
-                                "operation_history": operation_history[-10:]  # Last 10 operations
-                            })
-                            break
+    # Mutable container for the app reference (set after app creation)
+    # This allows the worker to access app.request_confirmation()
+    app_ref: List[Optional['SyncApp']] = [None]
 
-                # Detect consecutive fix operations (infinite fix loop protection)
-                if operation == 'fix':
-                    # Count consecutive fix operations
-                    consecutive_fixes = 0
-                    for i in range(len(operation_history) - 1, -1, -1):
-                        if operation_history[i] == 'fix':
-                            consecutive_fixes += 1
-                        else:
-                            break
-                    
-                    MAX_CONSECUTIVE_FIXES = 5  # Allow up to 5 consecutive fix attempts
-                    if consecutive_fixes >= MAX_CONSECUTIVE_FIXES:
-                        errors.append(f"Detected {consecutive_fixes} consecutive fix operations. Breaking infinite fix loop.")
-                        errors.append("The test failures may not be resolvable by automated fixes in this environment.")
-                        log_sync_event(basename, language, "cycle_detected", {
-                            "cycle_type": "consecutive-fix",
-                            "consecutive_count": consecutive_fixes,
-                            "operation_history": operation_history[-10:]  # Last 10 operations
+    # Progress callback ref for TUI ProgressBar updates during auto-deps
+    progress_callback_ref: List[Optional[Callable[[int, int], None]]] = [None]
+
+    # Track if user has already confirmed overwrite (to avoid asking multiple times)
+    user_confirmed_overwrite: List[bool] = [False]
+
+    def get_confirm_callback() -> Optional[Callable[[str, str], bool]]:
+        """Get the confirmation callback from the app if available.
+
+        Once user confirms, we remember it so subsequent operations don't ask again.
+        """
+        if user_confirmed_overwrite[0]:
+            # User already confirmed, return a callback that always returns True
+            return lambda msg, title: True
+
+        if app_ref[0] is not None:
+            def confirming_callback(msg: str, title: str) -> bool:
+                result = app_ref[0].request_confirmation(msg, title)
+                if result:
+                    user_confirmed_overwrite[0] = True
+                return result
+            return confirming_callback
+        return confirm_callback  # Fall back to provided callback
+
+    def sync_worker_logic():
+        """
+        The main loop of sync logic, run in a worker thread by Textual App.
+        """
+        operations_completed: List[str] = []
+        skipped_operations: List[str] = []
+        errors: List[str] = []
+        start_time = time.time()
+        last_model_name: str = ""
+        operation_history: List[str] = []
+        MAX_CYCLE_REPEATS = 2
+        
+        # Helper function to print inside worker (goes to RichLog via redirection)
+        # print() will work if sys.stdout is redirected.
+        
+        try:
+            with SyncLock(basename, language):
+                log_sync_event(basename, language, "lock_acquired", {"pid": os.getpid()})
+                
+                while True:
+                    budget_remaining = budget - current_cost_ref[0]
+                    if current_cost_ref[0] >= budget:
+                        errors.append(f"Budget of ${budget:.2f} exceeded.")
+                        log_sync_event(basename, language, "budget_exceeded", {
+                            "total_cost": current_cost_ref[0], 
+                            "budget": budget
                         })
                         break
 
-                if operation in ['all_synced', 'nothing', 'fail_and_request_manual_merge', 'error', 'analyze_conflict']:
-                    current_function_name_ref[0] = "synced" if operation in ['all_synced', 'nothing'] else "conflict"
-                    
-                    # Log these final operations
-                    success = operation in ['all_synced', 'nothing']
-                    error_msg = None
-                    if operation == 'fail_and_request_manual_merge':
-                        errors.append(f"Manual merge required: {decision.reason}")
-                        error_msg = f"Manual merge required: {decision.reason}"
-                    elif operation == 'error':
-                        errors.append(f"Error determining operation: {decision.reason}")
-                        error_msg = f"Error determining operation: {decision.reason}"
-                    elif operation == 'analyze_conflict':
-                        errors.append(f"Conflict detected: {decision.reason}")
-                        error_msg = f"Conflict detected: {decision.reason}"
-                    
-                    # Update log entry for final operation
-                    update_sync_log_entry(log_entry, {
-                        'success': success,
-                        'cost': 0.0,
-                        'model': 'none',
-                        'error': error_msg
-                    }, 0.0)
-                    append_sync_log(basename, language, log_entry)
-                    
-                    break
-                
-                # Handle skips
-                if operation == 'verify' and (skip_verify or skip_tests):
-                    # Skip verification if explicitly requested OR if tests are skipped (can't verify without tests)
-                    skipped_operations.append('verify')
-                    skip_reason = 'skip_verify' if skip_verify else 'skip_tests_implies_skip_verify'
-                    
-                    # Update log entry for skipped operation
-                    update_sync_log_entry(log_entry, {
-                        'success': True,
-                        'cost': 0.0,
-                        'model': 'skipped',
-                        'error': None
-                    }, 0.0)
-                    log_entry['details']['skip_reason'] = skip_reason
-                    append_sync_log(basename, language, log_entry)
+                    if budget_remaining < budget * 0.2 and budget_remaining > 0:
+                        log_sync_event(basename, language, "budget_warning", {
+                            "remaining": budget_remaining,
+                            "percentage": (budget_remaining / budget) * 100
+                        })
+
+                    decision = sync_determine_operation(basename, language, target_coverage, budget_remaining, False, prompts_dir, skip_tests, skip_verify, context_override)
+                    operation = decision.operation
                     
-                    report_data = RunReport(
-                        timestamp=datetime.datetime.now(datetime.timezone.utc).isoformat(),
-                        exit_code=0, tests_passed=0, tests_failed=0, coverage=0.0
-                    )
-                    save_run_report(asdict(report_data), basename, language)
-                    _save_operation_fingerprint(basename, language, 'verify', pdd_files, 0.0, skip_reason)
-                    continue
-                if operation == 'test' and skip_tests:
-                    skipped_operations.append('test')
+                    log_entry = create_sync_log_entry(decision, budget_remaining)
+                    operation_history.append(operation)
                     
-                    # Update log entry for skipped operation
-                    update_sync_log_entry(log_entry, {
-                        'success': True,
-                        'cost': 0.0,
-                        'model': 'skipped',
-                        'error': None
-                    }, 0.0)
-                    log_entry['details']['skip_reason'] = 'skip_tests'
-                    append_sync_log(basename, language, log_entry)
+                    # Cycle detection logic
+                    if len(operation_history) >= 3:
+                        recent_auto_deps = [op for op in operation_history[-3:] if op == 'auto-deps']
+                        if len(recent_auto_deps) >= 2:
+                            errors.append("Detected auto-deps infinite loop. Force advancing to generate operation.")
+                            log_sync_event(basename, language, "cycle_detected", {"cycle_type": "auto-deps-infinite"})
+                            operation = 'generate'
+                            decision.operation = 'generate' # Update decision too
+
+                    # Bug #4 fix: Detect crash-verify cycle pattern
+                    # The pattern [crash, verify, crash, verify] or [verify, crash, verify, crash]
+                    # represents 2 iterations of the alternating cycle, so break immediately
+                    if len(operation_history) >= 4:
+                        recent_ops = operation_history[-4:]
+                        if (recent_ops == ['crash', 'verify', 'crash', 'verify'] or
+                            recent_ops == ['verify', 'crash', 'verify', 'crash']):
+                            # Pattern detected - this represents MAX_CYCLE_REPEATS iterations
+                            errors.append(f"Detected crash-verify cycle repeated {MAX_CYCLE_REPEATS} times. Breaking cycle.")
+                            log_sync_event(basename, language, "cycle_detected", {"cycle_type": "crash-verify", "count": MAX_CYCLE_REPEATS})
+                            break
+
+                    # Bug #4 fix: Detect test-fix cycle pattern
+                    # The pattern [test, fix, test, fix] or [fix, test, fix, test]
+                    # represents 2 iterations of the alternating cycle, so break immediately
+                    if len(operation_history) >= 4:
+                        recent_ops = operation_history[-4:]
+                        if (recent_ops == ['test', 'fix', 'test', 'fix'] or
+                            recent_ops == ['fix', 'test', 'fix', 'test']):
+                            # Pattern detected - this represents MAX_CYCLE_REPEATS iterations
+                            errors.append(f"Detected test-fix cycle repeated {MAX_CYCLE_REPEATS} times. Breaking cycle.")
+                            log_sync_event(basename, language, "cycle_detected", {"cycle_type": "test-fix", "count": MAX_CYCLE_REPEATS})
+                            break
+                                
+                    if operation == 'fix':
+                        consecutive_fixes = 0
+                        for i in range(len(operation_history) - 1, -1, -1):
+                            if operation_history[i] == 'fix':
+                                consecutive_fixes += 1
+                            else:
+                                break
+                        if consecutive_fixes >= 5:
+                            errors.append(f"Detected {consecutive_fixes} consecutive fix operations. Breaking infinite fix loop.")
+                            break
+
+                    if operation == 'test':
+                        consecutive_tests = 0
+                        for i in range(len(operation_history) - 1, -1, -1):
+                            if operation_history[i] == 'test':
+                                consecutive_tests += 1
+                            else:
+                                break
+                        if consecutive_tests >= MAX_CONSECUTIVE_TESTS:
+                            errors.append(f"Detected {consecutive_tests} consecutive test operations. Breaking infinite test loop.")
+                            break
+
+                    # Bug #157 fix: Prevent infinite crash retry loops
+                    if operation == 'crash':
+                        consecutive_crashes = 0
+                        for i in range(len(operation_history) - 1, -1, -1):
+                            if operation_history[i] == 'crash':
+                                consecutive_crashes += 1
+                            else:
+                                break
+                        if consecutive_crashes >= MAX_CONSECUTIVE_CRASHES:
+                            errors.append(f"Detected {consecutive_crashes} consecutive crash operations. Breaking infinite crash loop.")
+                            break
+
+                    if operation == 'test_extend':
+                        # Count test_extend attempts to prevent infinite loop
+                        extend_attempts = sum(1 for op in operation_history if op == 'test_extend')
+                        if extend_attempts >= MAX_TEST_EXTEND_ATTEMPTS:
+                            # Accept current coverage after max attempts
+                            log_sync_event(basename, language, "test_extend_limit", {
+                                "attempts": extend_attempts,
+                                "max_attempts": MAX_TEST_EXTEND_ATTEMPTS,
+                                "reason": "Accepting current coverage after max extend attempts"
+                            })
+                            success = True
+                            break
+
+                    if operation in ['all_synced', 'nothing', 'fail_and_request_manual_merge', 'error', 'analyze_conflict']:
+                        current_function_name_ref[0] = "synced" if operation in ['all_synced', 'nothing'] else "conflict"
+                        success = operation in ['all_synced', 'nothing']
+                        error_msg = None
+                        if operation == 'fail_and_request_manual_merge':
+                            errors.append(f"Manual merge required: {decision.reason}")
+                            error_msg = decision.reason
+                        elif operation == 'error':
+                            errors.append(f"Error determining operation: {decision.reason}")
+                            error_msg = decision.reason
+                        elif operation == 'analyze_conflict':
+                            errors.append(f"Conflict detected: {decision.reason}")
+                            error_msg = decision.reason
+                        
+                        update_sync_log_entry(log_entry, {'success': success, 'cost': 0.0, 'model': 'none', 'error': error_msg}, 0.0)
+                        append_sync_log(basename, language, log_entry)
+                        break
                     
-                    report_data = RunReport(
-                        timestamp=datetime.datetime.now(datetime.timezone.utc).isoformat(),
-                        exit_code=0, tests_passed=0, tests_failed=0, coverage=1.0
+                    # Handle skips - save fingerprint with 'skip:' prefix to distinguish from actual execution
+                    # Bug #11 fix: Use 'skip:' prefix so _is_workflow_complete() knows the op was skipped
+                    if operation == 'verify' and (skip_verify or skip_tests):
+                        skipped_operations.append('verify')
+                        update_sync_log_entry(log_entry, {'success': True, 'cost': 0.0, 'model': 'skipped', 'error': None}, 0.0)
+                        append_sync_log(basename, language, log_entry)
+                        # Save fingerprint with 'skip:' prefix to indicate operation was skipped, not executed
+                        _save_operation_fingerprint(basename, language, 'skip:verify', pdd_files, 0.0, 'skipped')
+                        continue
+                    if operation == 'test' and skip_tests:
+                        skipped_operations.append('test')
+                        update_sync_log_entry(log_entry, {'success': True, 'cost': 0.0, 'model': 'skipped', 'error': None}, 0.0)
+                        append_sync_log(basename, language, log_entry)
+                        # Save fingerprint with 'skip:' prefix to indicate operation was skipped, not executed
+                        _save_operation_fingerprint(basename, language, 'skip:test', pdd_files, 0.0, 'skipped')
+                        continue
+                    if operation == 'crash' and (skip_tests or skip_verify):
+                        skipped_operations.append('crash')
+                        update_sync_log_entry(log_entry, {'success': True, 'cost': 0.0, 'model': 'skipped', 'error': None}, 0.0)
+                        append_sync_log(basename, language, log_entry)
+                        # Save fingerprint with 'skip:' prefix to indicate operation was skipped, not executed
+                        _save_operation_fingerprint(basename, language, 'skip:crash', pdd_files, 0.0, 'skipped')
+                        # FIX: Create a synthetic run_report to prevent infinite loop when crash is skipped
+                        # Without this, sync_determine_operation keeps returning 'crash' because no run_report exists
+                        current_hashes = calculate_current_hashes(pdd_files)
+                        synthetic_report = RunReport(
+                            timestamp=datetime.datetime.now(datetime.timezone.utc).isoformat(),
+                            exit_code=0,  # Assume success since we're skipping validation
+                            tests_passed=0,
+                            tests_failed=0,
+                            coverage=0.0,
+                            test_hash=current_hashes.get('test_hash')
+                        )
+                        save_run_report(asdict(synthetic_report), basename, language)
+                        continue
+
+                    current_function_name_ref[0] = operation
+                    ctx = _create_mock_context(
+                        force=force, strength=strength, temperature=temperature, time=time_param,
+                        verbose=verbose, quiet=quiet, output_cost=output_cost,
+                        review_examples=review_examples, local=local, budget=budget - current_cost_ref[0],
+                        max_attempts=max_attempts, target_coverage=target_coverage,
+                        confirm_callback=get_confirm_callback(),
+                        context=context_override
                     )
-                    save_run_report(asdict(report_data), basename, language)
-                    _save_operation_fingerprint(basename, language, 'test', pdd_files, 0.0, 'skipped')
-                    continue
-                if operation == 'crash' and skip_tests:
-                    # Skip crash operations when tests are skipped since crash fixes usually require test execution
-                    skipped_operations.append('crash')
                     
-                    # Update log entry for skipped operation
-                    update_sync_log_entry(log_entry, {
-                        'success': True,
-                        'cost': 0.0,
-                        'model': 'skipped',
-                        'error': None
-                    }, 0.0)
-                    log_entry['details']['skip_reason'] = 'skip_tests'
-                    append_sync_log(basename, language, log_entry)
-                    
-                    # Create a dummy run report indicating crash was skipped
-                    report_data = RunReport(
-                        timestamp=datetime.datetime.now(datetime.timezone.utc).isoformat(),
-                        exit_code=0, tests_passed=0, tests_failed=0, coverage=0.0
-                    )
-                    save_run_report(asdict(report_data), basename, language)
-                    _save_operation_fingerprint(basename, language, 'crash', pdd_files, 0.0, 'skipped')
-                    continue
-
-                current_function_name_ref[0] = operation
-                ctx = _create_mock_context(
-                    force=force, strength=strength, temperature=temperature, time=time_param,
-                    verbose=verbose, quiet=quiet, output_cost=output_cost,
-                    review_examples=review_examples, local=local, budget=budget - current_cost_ref[0],
-                    max_attempts=max_attempts, target_coverage=target_coverage
-                )
-                
-                result = {}
-                success = False
-                start_time = time.time()  # Track execution time
-
-                # --- Execute Operation ---
-                try:
-                    if operation == 'auto-deps':
-                        # Save the modified prompt to a temporary location
-                        temp_output = str(pdd_files['prompt']).replace('.prompt', '_with_deps.prompt')
-                        
-                        # Read original prompt content to compare later
-                        original_content = pdd_files['prompt'].read_text(encoding='utf-8')
-                        
-                        result = auto_deps_main(
-                            ctx, 
-                            prompt_file=str(pdd_files['prompt']), 
-                            directory_path=examples_dir,
-                            auto_deps_csv_path="project_dependencies.csv",
-                            output=temp_output,
-                            force_scan=False  # Don't force scan every time
-                        )
-                        
-                        # Only move the temp file back if content actually changed
-                        if Path(temp_output).exists():
-                            import shutil
-                            new_content = Path(temp_output).read_text(encoding='utf-8')
-                            if new_content != original_content:
-                                shutil.move(temp_output, str(pdd_files['prompt']))
-                            else:
-                                # No changes needed, remove temp file
-                                Path(temp_output).unlink()
-                                # Mark as successful with no changes
-                                result = (new_content, 0.0, 'no-changes')
-                    elif operation == 'generate':
-                        result = code_generator_main(
-                            ctx, 
-                            prompt_file=str(pdd_files['prompt']), 
-                            output=str(pdd_files['code']),
-                            original_prompt_file_path=None,
-                            force_incremental_flag=False
-                        )
-                    elif operation == 'example':
-                        print(f"DEBUG SYNC: pdd_files['example'] = {pdd_files['example']}")
-                        print(f"DEBUG SYNC: str(pdd_files['example']) = {str(pdd_files['example'])}")
-                        result = context_generator_main(
-                            ctx, 
-                            prompt_file=str(pdd_files['prompt']), 
-                            code_file=str(pdd_files['code']), 
-                            output=str(pdd_files['example'])
-                        )
-                    elif operation == 'crash':
-                        # Validate required files exist before attempting crash operation
-                        required_files = [pdd_files['code'], pdd_files['example']]
-                        missing_files = [f for f in required_files if not f.exists()]
-                        
-                        if missing_files:
-                            # Skip crash operation if required files are missing
-                            print(f"Skipping crash operation - missing files: {[f.name for f in missing_files]}")
-                            skipped_operations.append('crash')
-                            
-                            # Update log entry for skipped operation
-                            update_sync_log_entry(log_entry, {
-                                'success': True,
-                                'cost': 0.0,
-                                'model': 'skipped',
-                                'error': None
-                            }, 0.0)
-                            log_entry['details']['skip_reason'] = 'missing_files'
-                            log_entry['details']['missing_files'] = [f.name for f in missing_files]
-                            append_sync_log(basename, language, log_entry)
+                    result = {}
+                    success = False
+                    op_start_time = time.time()
+
+                    # Issue #159 fix: Use atomic state for consistent run_report + fingerprint writes
+                    with AtomicStateUpdate(basename, language) as atomic_state:
+
+                        # --- Execute Operation ---
+                        try:
+                            if operation == 'auto-deps':
+                                temp_output = str(pdd_files['prompt']).replace('.prompt', '_with_deps.prompt')
+                                original_content = pdd_files['prompt'].read_text(encoding='utf-8')
+                                result = auto_deps_main(
+                                    ctx,
+                                    prompt_file=str(pdd_files['prompt']),
+                                    directory_path=examples_dir,
+                                    auto_deps_csv_path="project_dependencies.csv",
+                                    output=temp_output,
+                                    force_scan=False,
+                                    progress_callback=progress_callback_ref[0]
+                                )
+                                if Path(temp_output).exists():
+                                    import shutil
+                                    new_content = Path(temp_output).read_text(encoding='utf-8')
+                                    if new_content != original_content:
+                                        shutil.move(temp_output, str(pdd_files['prompt']))
+                                    else:
+                                        Path(temp_output).unlink()
+                                        result = (new_content, 0.0, 'no-changes')
+                            elif operation == 'generate':
+                                result = code_generator_main(ctx, prompt_file=str(pdd_files['prompt']), output=str(pdd_files['code']), original_prompt_file_path=None, force_incremental_flag=False)
+                                # Clear stale run_report so crash/verify is required for newly generated code
+                                run_report_file = META_DIR / f"{basename}_{language}_run.json"
+                                run_report_file.unlink(missing_ok=True)
+                            elif operation == 'example':
+                                result = context_generator_main(ctx, prompt_file=str(pdd_files['prompt']), code_file=str(pdd_files['code']), output=str(pdd_files['example']))
+                            elif operation == 'crash':
+                                required_files = [pdd_files['code'], pdd_files['example']]
+                                missing_files = [f for f in required_files if not f.exists()]
+                                if missing_files:
+                                    skipped_operations.append('crash')
+                                    continue
                             
-                            # Create a dummy run report indicating crash was skipped due to missing files
-                            report_data = RunReport(
-                                timestamp=datetime.datetime.now(datetime.timezone.utc).isoformat(),
-                                exit_code=0, tests_passed=0, tests_failed=0, coverage=0.0
-                            )
-                            save_run_report(asdict(report_data), basename, language)
-                            _save_operation_fingerprint(basename, language, 'crash', pdd_files, 0.0, 'skipped_missing_files')
-                            continue
-                        else:
-                            # Check if we have a run report indicating failures that need crash fixing
-                            current_run_report = read_run_report(basename, language)
-                            crash_log_content = ""
+                                # Crash handling logic (simplified copy from original)
+                                current_run_report = read_run_report(basename, language)
+                                crash_log_content = ""
                             
-                            # If we have a run report with exit_code != 0, that indicates a crash that needs fixing
-                            if current_run_report and current_run_report.exit_code != 0:
-                                # We have a crash to fix based on the run report
-                                crash_log_content = f"Test execution failed with exit code: {current_run_report.exit_code}\n\n"
-                                
-                                # Try to run the example program to get additional error details
-                                try:
-                                    example_result = subprocess.run(
-                                        ['python', str(pdd_files['example'])],
-                                        capture_output=True,
-                                        text=True,
-                                        timeout=60,
-                                        env=os.environ.copy(),
-                                        cwd=str(pdd_files['example'].parent)
+                                # Check for crash condition (either run report says so, or we check manually)
+                                has_crash = False
+                                if current_run_report and current_run_report.exit_code != 0:
+                                    has_crash = True
+                                    crash_log_content = f"Test execution failed exit code: {current_run_report.exit_code}\n"
+                                else:
+                                    # Manual check - run the example to see if it crashes
+                                    env = os.environ.copy()
+                                    src_dir = Path.cwd() / 'src'
+                                    env['PYTHONPATH'] = f"{src_dir}:{env.get('PYTHONPATH', '')}"
+                                    # Remove TUI-specific env vars that might contaminate subprocess
+                                    for var in ['FORCE_COLOR', 'COLUMNS']:
+                                        env.pop(var, None)
+                                    # Get language-appropriate run command from language_format.csv
+                                    example_path = str(pdd_files['example'])
+                                    run_cmd = get_run_command_for_file(example_path)
+                                    if run_cmd:
+                                        # Use the language-specific interpreter (e.g., node for .js)
+                                        cmd_parts = run_cmd.split()
+                                    else:
+                                        # Fallback to Python if no run command found
+                                        cmd_parts = ['python', example_path]
+                                    # Use error-detection runner that handles server-style examples
+                                    returncode, stdout, stderr = _run_example_with_error_detection(
+                                        cmd_parts,
+                                        env=env,
+                                        cwd=str(pdd_files['example'].parent),
+                                        timeout=60
                                     )
-                                    
-                                    if example_result.returncode != 0:
-                                        crash_log_content += f"Example program also failed with exit code: {example_result.returncode}\n\n"
-                                        if example_result.stdout:
-                                            crash_log_content += f"STDOUT:\n{example_result.stdout}\n\n"
-                                        if example_result.stderr:
-                                            crash_log_content += f"STDERR:\n{example_result.stderr}\n"
-                                        
-                                        # Check for syntax errors specifically
-                                        if "SyntaxError" in example_result.stderr:
-                                            crash_log_content = f"SYNTAX ERROR DETECTED:\n\n{crash_log_content}"
+
+                                    class ExampleResult:
+                                        def __init__(self, rc, out, err):
+                                            self.returncode = rc
+                                            self.stdout = out
+                                            self.stderr = err
+
+                                    ex_res = ExampleResult(returncode, stdout, stderr)
+                                    if ex_res.returncode != 0:
+                                        has_crash = True
+                                        crash_log_content = f"Example failed exit code: {ex_res.returncode}\nSTDOUT:\n{ex_res.stdout}\nSTDERR:\n{ex_res.stderr}\n"
+                                        if "SyntaxError" in ex_res.stderr:
+                                             crash_log_content = "SYNTAX ERROR DETECTED:\n" + crash_log_content
                                     else:
-                                        crash_log_content += "Example program runs successfully, but tests are failing.\n"
-                                        crash_log_content += "This may indicate issues with test execution or test file syntax.\n"
-                                        
-                                except subprocess.TimeoutExpired:
-                                    crash_log_content += "Example program execution timed out after 60 seconds\n"
-                                    crash_log_content += "This may indicate an infinite loop or the program is waiting for input.\n"
-                                except Exception as e:
-                                    crash_log_content += f"Error running example program: {str(e)}\n"
-                                    crash_log_content += f"Program path: {pdd_files['example']}\n"
-                            else:
-                                # No crash detected, skip crash operation
-                                print("No crash detected in run report, skipping crash fix")
-                                skipped_operations.append('crash')
-                                
-                                # Update log entry for skipped operation
-                                update_sync_log_entry(log_entry, {
-                                    'success': True,
-                                    'cost': 0.0,
-                                    'model': 'skipped',
-                                    'error': None
-                                }, time.time() - start_time)
-                                log_entry['details']['skip_reason'] = 'no_crash'
-                                append_sync_log(basename, language, log_entry)
-                                
-                                report_data = RunReport(
-                                    timestamp=datetime.datetime.now(datetime.timezone.utc).isoformat(),
-                                    exit_code=0, tests_passed=0, tests_failed=0, coverage=0.0
-                                )
-                                save_run_report(asdict(report_data), basename, language)
-                                _save_operation_fingerprint(basename, language, 'crash', pdd_files, 0.0, 'no_crash')
-                                continue
-                            
-                            # Write actual error content or fallback
-                            if not crash_log_content:
-                                crash_log_content = "Unknown crash error - program failed but no error output captured"
-                            
-                            Path("crash.log").write_text(crash_log_content)
-                            
-                            try:
-                                result = crash_main(
-                                    ctx, 
-                                    prompt_file=str(pdd_files['prompt']), 
-                                    code_file=str(pdd_files['code']), 
-                                    program_file=str(pdd_files['example']), 
-                                    error_file="crash.log",
-                                    output=str(pdd_files['code']),
-                                    output_program=str(pdd_files['example']),
-                                    loop=True,
-                                    max_attempts=max_attempts,
-                                    budget=budget - current_cost_ref[0]
-                                )
-                            except (RuntimeError, Exception) as e:
-                                error_str = str(e)
-                                if ("LLM returned None" in error_str or 
-                                    "LLM failed to analyze errors" in error_str):
-                                    # Skip crash operation for LLM failures
-                                    print(f"Skipping crash operation due to LLM error: {e}")
-                                    skipped_operations.append('crash')
-                                    
-                                    # Update log entry for skipped operation
-                                    update_sync_log_entry(log_entry, {
-                                        'success': False,
-                                        'cost': 0.0,
-                                        'model': 'skipped',
-                                        'error': f"LLM error: {str(e)}"
-                                    }, time.time() - start_time)
-                                    log_entry['details']['skip_reason'] = 'llm_error'
-                                    append_sync_log(basename, language, log_entry)
+                                        # No crash - save run report with exit_code=0 so sync_determine_operation
+                                        # knows the example was tested and passed (prevents infinite loop)
+                                        # Include test_hash for staleness detection
+                                        test_hash = calculate_sha256(pdd_files['test']) if pdd_files['test'].exists() else None
+                                        report = RunReport(
+                                            datetime.datetime.now(datetime.timezone.utc).isoformat(),
+                                            exit_code=0,
+                                            tests_passed=1,
+                                            tests_failed=0,
+                                            coverage=0.0,
+                                            test_hash=test_hash
+                                        )
+                                        save_run_report(asdict(report), basename, language)
+                                        skipped_operations.append('crash')
+                                        continue
                                     
-                                    report_data = RunReport(
-                                        timestamp=datetime.datetime.now(datetime.timezone.utc).isoformat(),
-                                        exit_code=0, tests_passed=0, tests_failed=0, coverage=0.0
+                                if has_crash:
+                                    # Try auto-fix for common import errors before expensive agentic call
+                                    auto_fixed, auto_fix_msg = _try_auto_fix_import_error(
+                                        crash_log_content,
+                                        pdd_files['code'],
+                                        pdd_files['example']
                                     )
-                                    save_run_report(asdict(report_data), basename, language)
-                                    _save_operation_fingerprint(basename, language, 'crash', pdd_files, 0.0, 'skipped_llm_error')
+                                    if auto_fixed:
+                                        log_sync_event(basename, language, "auto_fix_attempted", {"message": auto_fix_msg})
+                                        # Retry running the example after auto-fix
+                                        retry_returncode, retry_stdout, retry_stderr = _run_example_with_error_detection(
+                                            cmd_parts,
+                                            env=env,
+                                            cwd=str(pdd_files['example'].parent),
+                                            timeout=60
+                                        )
+                                        if retry_returncode == 0:
+                                            # Auto-fix worked! Save run report and continue
+                                            log_sync_event(basename, language, "auto_fix_success", {"message": auto_fix_msg})
+                                            test_hash = calculate_sha256(pdd_files['test']) if pdd_files['test'].exists() else None
+                                            report = RunReport(
+                                                datetime.datetime.now(datetime.timezone.utc).isoformat(),
+                                                exit_code=0,
+                                                tests_passed=1,
+                                                tests_failed=0,
+                                                coverage=0.0,
+                                                test_hash=test_hash
+                                            )
+                                            save_run_report(asdict(report), basename, language)
+                                            result = (True, 0.0, 'auto-fix')
+                                            success = True
+                                            actual_cost = 0.0
+                                            model_name = 'auto-fix'
+                                            # Update crash_log_content for logging
+                                            crash_log_content = f"Auto-fixed: {auto_fix_msg}"
+                                            continue  # Skip crash_main, move to next operation
+                                        else:
+                                            # Auto-fix didn't fully work, update error log and proceed
+                                            crash_log_content = f"Auto-fix attempted ({auto_fix_msg}) but still failing:\nRETRY STDOUT:\n{retry_stdout}\nRETRY STDERR:\n{retry_stderr}\n"
+
+                                    Path("crash.log").write_text(crash_log_content)
+                                    try:
+                                        result = crash_main(ctx, prompt_file=str(pdd_files['prompt']), code_file=str(pdd_files['code']), program_file=str(pdd_files['example']), error_file="crash.log", output=str(pdd_files['code']), output_program=str(pdd_files['example']), loop=True, max_attempts=max_attempts, budget=budget - current_cost_ref[0], strength=strength, temperature=temperature)
+                                    except Exception as e:
+                                        print(f"Crash fix failed: {e}")
+                                        skipped_operations.append('crash')
+                                        continue
+
+                            elif operation == 'verify':
+                                if not pdd_files['example'].exists():
+                                    skipped_operations.append('verify')
                                     continue
-                                else:
-                                    # Re-raise other exceptions
-                                    raise
-                    elif operation == 'verify':
-                        result = fix_verification_main(
-                            ctx, 
-                            prompt_file=str(pdd_files['prompt']), 
-                            code_file=str(pdd_files['code']), 
-                            program_file=str(pdd_files['example']),
-                            output_results=f"{basename}_verify_results.log",
-                            output_code=str(pdd_files['code']),
-                            output_program=str(pdd_files['example']),
-                            loop=True,
-                            verification_program=str(pdd_files['example']),
-                            max_attempts=max_attempts,
-                            budget=budget - current_cost_ref[0]
-                        )
-                    elif operation == 'test':
-                        # First, generate the test file
-                        result = cmd_test_main(
-                            ctx, 
-                            prompt_file=str(pdd_files['prompt']), 
-                            code_file=str(pdd_files['code']), 
-                            output=str(pdd_files['test']),
-                            language=language,
-                            coverage_report=None,
-                            existing_tests=None,
-                            target_coverage=target_coverage,
-                            merge=False
-                        )
-                        
-                        # After successful test generation, execute the tests and create run report
-                        # This enables the next sync iteration to detect test failures and trigger fix
-                        if isinstance(result, dict) and result.get('success', False):
-                            try:
-                                test_file = pdd_files['test']
-                                if test_file.exists():
+                                result = fix_verification_main(ctx, prompt_file=str(pdd_files['prompt']), code_file=str(pdd_files['code']), program_file=str(pdd_files['example']), output_results=f"{basename}_verify_results.log", output_code=str(pdd_files['code']), output_program=str(pdd_files['example']), loop=True, verification_program=str(pdd_files['example']), max_attempts=max_attempts, budget=budget - current_cost_ref[0], strength=strength, temperature=temperature)
+                            elif operation == 'test':
+                                pdd_files['test'].parent.mkdir(parents=True, exist_ok=True)
+                                # Use merge=True when test file exists to preserve fixes and append new tests
+                                # instead of regenerating from scratch (which would overwrite fixes)
+                                test_file_exists = pdd_files['test'].exists()
+                                result = cmd_test_main(ctx, prompt_file=str(pdd_files['prompt']), code_file=str(pdd_files['code']), output=str(pdd_files['test']), language=language, coverage_report=None, existing_tests=[str(pdd_files['test'])] if test_file_exists else None, target_coverage=target_coverage, merge=test_file_exists, strength=strength, temperature=temperature)
+                                if pdd_files['test'].exists():
                                     _execute_tests_and_create_run_report(
-                                        test_file, basename, language, target_coverage
+                                        pdd_files['test'],
+                                        basename,
+                                        language,
+                                        target_coverage,
+                                        code_file=pdd_files.get("code"),
+                                        atomic_state=atomic_state,
+                                        test_files=pdd_files.get('test_files'),  # Bug #156
+                                    )
+                            elif operation == 'test_extend':
+                                # Extend existing tests to improve coverage
+                                # Uses existing_tests and merge=True to add more test cases
+                                pdd_files['test'].parent.mkdir(parents=True, exist_ok=True)
+                                if pdd_files['test'].exists():
+                                    existing_test_path = str(pdd_files['test'])
+                                    result = cmd_test_main(
+                                        ctx,
+                                        prompt_file=str(pdd_files['prompt']),
+                                        code_file=str(pdd_files['code']),
+                                        output=str(pdd_files['test']),
+                                        language=language,
+                                        coverage_report=None,
+                                        existing_tests=[existing_test_path],
+                                        target_coverage=target_coverage,
+                                        merge=True,
+                                        strength=strength,
+                                        temperature=temperature
                                     )
-                            except Exception as e:
-                                # Don't fail the entire operation if test execution fails
-                                # Just log it - the test file generation was successful
-                                print(f"Warning: Test execution failed: {e}")
-                        elif isinstance(result, tuple) and len(result) >= 3:
-                            # Handle tuple return format - assume success and execute tests
-                            try:
-                                test_file = pdd_files['test']
-                                if test_file.exists():
                                     _execute_tests_and_create_run_report(
-                                        test_file, basename, language, target_coverage
+                                        pdd_files['test'],
+                                        basename,
+                                        language,
+                                        target_coverage,
+                                        code_file=pdd_files.get("code"),
+                                        atomic_state=atomic_state,
+                                        test_files=pdd_files.get('test_files'),  # Bug #156
                                     )
-                            except Exception as e:
-                                print(f"Warning: Test execution failed: {e}")
-                    elif operation == 'fix':
-                        # Create error file with actual test failure information
-                        error_file_path = Path("fix_errors.log")
-                        
-                        # Try to get actual test failure details from latest run
-                        try:
-                            run_report = read_run_report(basename, language)
-                            if run_report and run_report.tests_failed > 0:
-                                # Run the tests again to capture actual error output
-                                # Use environment-aware Python executable for pytest execution
-                                python_executable = detect_host_python_executable()
-                                test_result = subprocess.run([
-                                    python_executable, '-m', 'pytest', 
-                                    str(pdd_files['test']), 
-                                    '-v', '--tb=short'
-                                ], capture_output=True, text=True, timeout=300)
-                                
-                                error_content = f"Test failures detected ({run_report.tests_failed} failed tests):\n\n"
-                                error_content += "STDOUT:\n" + test_result.stdout + "\n\n"
-                                error_content += "STDERR:\n" + test_result.stderr
+                                else:
+                                    # No existing test file, fall back to regular test generation
+                                    result = cmd_test_main(ctx, prompt_file=str(pdd_files['prompt']), code_file=str(pdd_files['code']), output=str(pdd_files['test']), language=language, coverage_report=None, existing_tests=None, target_coverage=target_coverage, merge=False, strength=strength, temperature=temperature)
+                                    if pdd_files['test'].exists():
+                                        _execute_tests_and_create_run_report(
+                                            pdd_files['test'],
+                                            basename,
+                                            language,
+                                            target_coverage,
+                                            code_file=pdd_files.get("code"),
+                                            atomic_state=atomic_state,
+                                            test_files=pdd_files.get('test_files'),  # Bug #156
+                                        )
+                            elif operation == 'fix':
+                                error_file_path = Path("fix_errors.log")
+                                # Capture errors using language-appropriate test command
+                                try:
+                                    from .get_test_command import get_test_command_for_file
+                                    test_cmd = get_test_command_for_file(str(pdd_files['test']), language)
+
+                                    # Use clean env without TUI-specific vars
+                                    clean_env = os.environ.copy()
+                                    for var in ['FORCE_COLOR', 'COLUMNS']:
+                                        clean_env.pop(var, None)
+
+                                    if test_cmd:
+                                        # Run language-appropriate test command
+                                        if language.lower() == 'python':
+                                            # Use pytest directly for Python
+                                            python_executable = detect_host_python_executable()
+                                            # Bug #156: Run pytest on ALL matching test files
+                                            test_files = pdd_files.get('test_files', [pdd_files['test']])
+                                            pytest_args = [python_executable, '-m', 'pytest'] + [str(f) for f in test_files] + ['-v', '--tb=short']
+                                            test_result = subprocess.run(
+                                                pytest_args,
+                                                capture_output=True, text=True, timeout=300,
+                                                stdin=subprocess.DEVNULL, env=clean_env, start_new_session=True,
+                                                cwd=str(pdd_files['test'].parent)
+                                            )
+                                        else:
+                                            # Use shell command for non-Python
+                                            test_result = subprocess.run(
+                                                test_cmd,
+                                                shell=True,
+                                                capture_output=True, text=True, timeout=300,
+                                                stdin=subprocess.DEVNULL, env=clean_env,
+                                                cwd=str(pdd_files['test'].parent),
+                                                start_new_session=True
+                                            )
+                                        error_content = f"Test output:\n{test_result.stdout}\n{test_result.stderr}"
+                                    else:
+                                        # No test command available - trigger agentic fallback with context
+                                        error_content = f"No test command available for {language}. Please run tests manually and provide error output."
+                                except Exception as e:
+                                    error_content = f"Test execution error: {e}"
+                                error_file_path.write_text(error_content)
+
+                                # Bug #156 fix: Parse pytest output to find actual failing files
+                                # and pass the correct file to fix_main
+                                failing_files = extract_failing_files_from_output(error_content)
+                                unit_test_file_for_fix = str(pdd_files['test'])  # Default to tracked file
+
+                                if failing_files:
+                                    # Try to resolve the failing file paths
+                                    test_dir = pdd_files['test'].parent
+                                    tracked_file_name = pdd_files['test'].name
+
+                                    # Check if the tracked file is among the failures
+                                    tracked_in_failures = any(
+                                        Path(ff).name == tracked_file_name for ff in failing_files
+                                    )
+
+                                    if not tracked_in_failures:
+                                        # Failures are in a different file - use the first failing file
+                                        for ff in failing_files:
+                                            # Try to resolve the path relative to test directory
+                                            ff_path = Path(ff)
+                                            if ff_path.is_absolute() and ff_path.exists():
+                                                unit_test_file_for_fix = str(ff_path)
+                                                break
+                                            else:
+                                                # Try to find it in the test directory
+                                                candidate = test_dir / ff_path.name
+                                                if candidate.exists():
+                                                    unit_test_file_for_fix = str(candidate)
+                                                    break
+                                                # Also try the path as-is relative to cwd
+                                                if ff_path.exists():
+                                                    unit_test_file_for_fix = str(ff_path.resolve())
+                                                    break
+
+                                result = fix_main(ctx, prompt_file=str(pdd_files['prompt']), code_file=str(pdd_files['code']), unit_test_file=unit_test_file_for_fix, error_file=str(error_file_path), output_test=str(pdd_files['test']), output_code=str(pdd_files['code']), output_results=f"{basename}_fix_results.log", loop=True, verification_program=str(pdd_files['example']), max_attempts=max_attempts, budget=budget - current_cost_ref[0], auto_submit=True, strength=strength, temperature=temperature)
+                            elif operation == 'update':
+                                result = update_main(ctx, input_prompt_file=str(pdd_files['prompt']), modified_code_file=str(pdd_files['code']), input_code_file=None, output=str(pdd_files['prompt']), use_git=True, strength=strength, temperature=temperature)
                             else:
-                                error_content = "Simulated test failures"
+                                errors.append(f"Unknown operation {operation}")
+                                result = {'success': False}
+
+                            # Result parsing
+                            if isinstance(result, dict):
+                                success = result.get('success', False)
+                                current_cost_ref[0] += result.get('cost', 0.0)
+                            elif isinstance(result, tuple) and len(result) >= 3:
+                                if operation == 'test': success = pdd_files['test'].exists()
+                                else: success = bool(result[0])
+                                cost = result[-2] if len(result) >= 2 and isinstance(result[-2], (int, float)) else 0.0
+                                current_cost_ref[0] += cost
+                            else:
+                                success = result is not None
+
                         except Exception as e:
-                            error_content = f"Could not capture test failures: {e}\nUsing simulated test failures"
-                        
-                        error_file_path.write_text(error_content)
-                        
-                        result = fix_main(
-                            ctx, 
-                            prompt_file=str(pdd_files['prompt']), 
-                            code_file=str(pdd_files['code']), 
-                            unit_test_file=str(pdd_files['test']), 
-                            error_file=str(error_file_path),
-                            output_test=str(pdd_files['test']),
-                            output_code=str(pdd_files['code']),
-                            output_results=f"{basename}_fix_results.log",
-                            loop=True,
-                            verification_program=str(pdd_files['example']),
-                            max_attempts=max_attempts,
-                            budget=budget - current_cost_ref[0],
-                            auto_submit=True
-                        )
-                    elif operation == 'update':
-                        result = update_main(
-                            ctx, 
-                            input_prompt_file=str(pdd_files['prompt']), 
-                            modified_code_file=str(pdd_files['code']),
-                            input_code_file=None,
-                            output=str(pdd_files['prompt']),
-                            git=True
-                        )
-                    else:
-                        errors.append(f"Unknown operation '{operation}' requested.")
-                        result = {'success': False, 'cost': 0.0}
+                            errors.append(f"Exception during '{operation}': {e}")
+                            success = False
                     
-                    # Handle different return formats from command functions
-                    if isinstance(result, dict):
-                        # Dictionary return (e.g., from some commands)
-                        success = result.get('success', False)
-                        current_cost_ref[0] += result.get('cost', 0.0)
-                    elif isinstance(result, tuple) and len(result) >= 3:
-                        # Tuple return (e.g., from code_generator_main, context_generator_main)
-                        # For tuples, success is determined by no exceptions and valid return content
-                        # Check if the first element (generated content) is None, which indicates failure
-                        success = result[0] is not None
-                        # Extract cost from tuple (usually second-to-last element)
-                        cost = result[-2] if len(result) >= 2 and isinstance(result[-2], (int, float)) else 0.0
-                        current_cost_ref[0] += cost
-                    else:
-                        # Unknown return format
-                        success = result is not None
-                        current_cost_ref[0] += 0.0
-
-                except Exception as e:
-                    errors.append(f"Exception during '{operation}': {e}")
-                    success = False
+                        # Log update
+                        duration = time.time() - op_start_time
+                        actual_cost = 0.0
+                        model_name = "unknown"
+                        if success:
+                            if isinstance(result, dict):
+                                 actual_cost = result.get('cost', 0.0)
+                                 model_name = result.get('model', 'unknown')
+                            elif isinstance(result, tuple) and len(result) >= 3:
+                                 actual_cost = result[-2] if len(result) >= 2 else 0.0
+                                 model_name = result[-1] if len(result) >= 1 else 'unknown'
+                            last_model_name = str(model_name)
+                            operations_completed.append(operation)
+                            _save_operation_fingerprint(basename, language, operation, pdd_files, actual_cost, str(model_name), atomic_state=atomic_state)
 
-                # Calculate execution duration
-                duration = time.time() - start_time
+                        update_sync_log_entry(log_entry, {'success': success, 'cost': actual_cost, 'model': model_name, 'error': errors[-1] if errors and not success else None}, duration)
+                        append_sync_log(basename, language, log_entry)
 
-                # Extract cost and model from result for logging
-                actual_cost = 0.0
-                model_name = "unknown"
-                error_message = None
-                
-                if success:
-                    if isinstance(result, dict):
-                        actual_cost = result.get('cost', 0.0)
-                        model_name = result.get('model', 'unknown')
-                    elif isinstance(result, tuple) and len(result) >= 3:
-                        actual_cost = result[-2] if len(result) >= 2 and isinstance(result[-2], (int, float)) else 0.0
-                        model_name = result[-1] if len(result) >= 1 and isinstance(result[-1], str) else 'unknown'
-                else:
-                    error_message = errors[-1] if errors else "Operation failed"
-
-                # Update and save log entry with execution results
-                update_sync_log_entry(log_entry, {
-                    'success': success,
-                    'cost': actual_cost,
-                    'model': model_name,
-                    'error': error_message
-                }, duration)
-                append_sync_log(basename, language, log_entry)
-
-                if success:
-                    operations_completed.append(operation)
-                    # Extract cost and model from result based on format
-                    if isinstance(result, dict):
-                        cost = result.get('cost', 0.0)
-                        model = result.get('model', '')
-                    elif isinstance(result, tuple) and len(result) >= 3:
-                        cost = result[-2] if len(result) >= 2 and isinstance(result[-2], (int, float)) else 0.0
-                        model = result[-1] if len(result) >= 1 and isinstance(result[-1], str) else ''
-                    else:
-                        cost = 0.0
-                        model = ''
-                    _save_operation_fingerprint(basename, language, operation, pdd_files, cost, model)
-                    
-                    # After successful crash operation, re-run the example to generate fresh run report
-                    if operation == 'crash':
-                        try:
-                            example_file = pdd_files['example']
-                            if example_file.exists():
-                                # Run the example program to check if crash is actually fixed
-                                try:
-                                    example_result = subprocess.run(
-                                        ['python', str(example_file)],
-                                        capture_output=True,
-                                        text=True,
-                                        timeout=60,
-                                        env=os.environ.copy(),
-                                        cwd=str(example_file.parent)
-                                    )
-                                    
-                                    # Create fresh run report based on actual execution
-                                    report_data = RunReport(
-                                        timestamp=datetime.datetime.now(datetime.timezone.utc).isoformat(),
-                                        exit_code=example_result.returncode,
-                                        tests_passed=1 if example_result.returncode == 0 else 0,
-                                        tests_failed=0 if example_result.returncode == 0 else 1,
-                                        coverage=100.0 if example_result.returncode == 0 else 0.0
-                                    )
-                                    save_run_report(asdict(report_data), basename, language)
-                                    print(f"Re-ran example after crash fix: exit_code={example_result.returncode}")
-                                    
-                                except subprocess.TimeoutExpired:
-                                    # Example timed out - still considered a failure
-                                    report_data = RunReport(
-                                        timestamp=datetime.datetime.now(datetime.timezone.utc).isoformat(),
-                                        exit_code=124,  # Standard timeout exit code
-                                        tests_passed=0, tests_failed=1, coverage=0.0
-                                    )
-                                    save_run_report(asdict(report_data), basename, language)
-                                    print("Example timed out after crash fix - created failure run report")
-                                    
-                        except Exception as e:
-                            # Don't fail the entire operation if example re-execution fails
-                            print(f"Warning: Post-crash example re-execution failed: {e}")
+                        # Post-operation checks (simplified)
+                        if success and operation == 'crash':
+                            # Re-run example to verify crash fix worked
+                            try:
+                                 # Use clean env without TUI-specific vars
+                                 clean_env = os.environ.copy()
+                                 for var in ['FORCE_COLOR', 'COLUMNS']:
+                                     clean_env.pop(var, None)
+                                 # Get language-appropriate run command
+                                 example_path = str(pdd_files['example'])
+                                 run_cmd = get_run_command_for_file(example_path)
+                                 if run_cmd:
+                                     cmd_parts = run_cmd.split()
+                                 else:
+                                     cmd_parts = ['python', example_path]
+                                 # Use error-detection runner that handles server-style examples
+                                 returncode, stdout, stderr = _run_example_with_error_detection(
+                                     cmd_parts,
+                                     env=clean_env,
+                                     cwd=str(pdd_files['example'].parent),
+                                     timeout=60
+                                 )
+                                 # Include test_hash for staleness detection
+                                 test_hash = calculate_sha256(pdd_files['test']) if pdd_files['test'].exists() else None
+                                 report = RunReport(datetime.datetime.now(datetime.timezone.utc).isoformat(), returncode, 1 if returncode==0 else 0, 0 if returncode==0 else 1, 100.0 if returncode==0 else 0.0, test_hash=test_hash)
+                                 save_run_report(asdict(report), basename, language)
+                            except Exception as e:
+                                 # Bug #8 fix: Don't silently swallow exceptions - log them and mark as error
+                                 error_msg = f"Post-crash verification failed: {e}"
+                                 errors.append(error_msg)
+                                 log_sync_event(basename, language, "post_crash_verification_failed", {"error": str(e)})
                     
-                    # After successful fix operation, execute tests to update run report
-                    if operation == 'fix':
-                        try:
-                            test_file = pdd_files['test']
-                            if test_file.exists():
+                        if success and operation == 'fix':
+                            # Re-run tests to update run_report after successful fix
+                            # This prevents infinite loop by updating the state machine
+                            if pdd_files['test'].exists():
                                 _execute_tests_and_create_run_report(
-                                    test_file, basename, language, target_coverage
+                                    pdd_files['test'],
+                                    basename,
+                                    language,
+                                    target_coverage,
+                                    code_file=pdd_files.get("code"),
+                                    atomic_state=atomic_state,
+                                    test_files=pdd_files.get('test_files'),  # Bug #156
                                 )
-                        except Exception as e:
-                            # Don't fail the entire operation if test execution fails
-                            print(f"Warning: Post-fix test execution failed: {e}")
-                else:
-                    errors.append(f"Operation '{operation}' failed.")
-                    break
+                    
+                        if not success:
+                            errors.append(f"Operation '{operation}' failed.")
+                            break
 
-    except TimeoutError:
-        errors.append(f"Could not acquire lock for '{basename}'. Another sync process may be running.")
-    except Exception as e:
-        errors.append(f"An unexpected error occurred in the orchestrator: {e}")
-    finally:
-        # Log lock release
-        try:
-            log_sync_event(basename, language, "lock_released", {
-                "pid": os.getpid(),
-                "total_operations": len(operations_completed) if 'operations_completed' in locals() else 0,
-                "total_cost": current_cost_ref[0] if 'current_cost_ref' in locals() else 0.0
-            })
-        except Exception:
-            pass  # Don't fail if logging fails
+        except BaseException as e:
+            errors.append(f"An unexpected error occurred in the orchestrator: {type(e).__name__}: {e}")
+            # Log the full traceback for debugging
+            import traceback
+            traceback.print_exc()
+        finally:
+            try:
+                log_sync_event(basename, language, "lock_released", {"pid": os.getpid(), "total_cost": current_cost_ref[0]})
+            except: pass
             
-        if stop_event:
-            stop_event.set()
-        if animation_thread and animation_thread.is_alive():
-            animation_thread.join(timeout=5)
+        # Return result dict
+        return {
+            'success': not errors,
+            'operations_completed': operations_completed,
+            'skipped_operations': skipped_operations,
+            'total_cost': current_cost_ref[0],
+            'total_time': time.time() - start_time,
+            'final_state': {p: {'exists': f.exists(), 'path': str(f)} for p, f in pdd_files.items() if p != 'test_files'},
+            'errors': errors,
+            'error': "; ".join(errors) if errors else None,  # Add this line
+            'model_name': last_model_name,
+        }
+
+    # Instantiate and run Textual App
+    app = SyncApp(
+        basename=basename,
+        budget=budget,
+        worker_func=sync_worker_logic,
+        function_name_ref=current_function_name_ref,
+        cost_ref=current_cost_ref,
+        prompt_path_ref=prompt_path_ref,
+        code_path_ref=code_path_ref,
+        example_path_ref=example_path_ref,
+        tests_path_ref=tests_path_ref,
+        prompt_color_ref=prompt_box_color_ref,
+        code_color_ref=code_box_color_ref,
+        example_color_ref=example_box_color_ref,
+        tests_color_ref=tests_box_color_ref,
+        stop_event=stop_event,
+        progress_callback_ref=progress_callback_ref
+    )
+
+    # Store app reference so worker can access request_confirmation
+    app_ref[0] = app
+
+    result = app.run()
+    
+    # Show exit animation if not quiet
+    if not quiet:
+        from .sync_tui import show_exit_animation
+        show_exit_animation()
+    
+    # Check for worker exception that might have caused a crash
+    if app.worker_exception:
+        print(f"\n[Error] Worker thread crashed with exception: {app.worker_exception}", file=sys.stderr)
         
-    total_time = time.time() - start_time
-    final_state = {
-        p_name: {'exists': p_path.exists(), 'path': str(p_path)} 
-        for p_name, p_path in pdd_files.items()
-    }
+        if hasattr(app, 'captured_logs') and app.captured_logs:
+             print("\n[Captured Logs (last 20 lines)]", file=sys.stderr)
+             for line in app.captured_logs[-20:]: # Print last 20 lines
+                 print(f"  {line}", file=sys.stderr)
+        
+        import traceback
+        # Use trace module to print the stored exception's traceback if available
+        if hasattr(app.worker_exception, '__traceback__'):
+            traceback.print_exception(type(app.worker_exception), app.worker_exception, app.worker_exception.__traceback__, file=sys.stderr)
+
+    if result is None:
+        return {
+            "success": False,
+            "total_cost": current_cost_ref[0],
+            "model_name": "",
+            "error": "Sync process interrupted or returned no result.",
+            "operations_completed": [],
+            "errors": ["App exited without result"]
+        }
     
-    return {
-        'success': not errors,
-        'operations_completed': operations_completed,
-        'skipped_operations': skipped_operations,
-        'total_cost': current_cost_ref[0],
-        'total_time': total_time,
-        'final_state': final_state,
-        'errors': errors,
-    }
+    return result
 
 if __name__ == '__main__':
-    # Example usage of the sync_orchestration module.
-    # This simulates running `pdd sync my_calculator` from the command line.
-    
-    print("--- Running Basic Sync Orchestration Example ---")
-    
-    # Setup a dummy project structure
+    # Example usage
     Path("./prompts").mkdir(exist_ok=True)
     Path("./src").mkdir(exist_ok=True)
     Path("./examples").mkdir(exist_ok=True)
     Path("./tests").mkdir(exist_ok=True)
     Path("./prompts/my_calculator_python.prompt").write_text("Create a calculator.")
-    
-    # Ensure PDD meta directory exists for logs and locks
     PDD_DIR.mkdir(exist_ok=True)
     META_DIR.mkdir(exist_ok=True)
-
-    result = sync_orchestration(
-        basename="my_calculator",
-        language="python",
-        quiet=True # Suppress mock command output for cleaner example run
-    )
-    
-    print("\n--- Sync Orchestration Finished ---")
+    result = sync_orchestration(basename="my_calculator", language="python", quiet=True)
     print(json.dumps(result, indent=2))
-
-    if result['success']:
-        print("\n✅ Sync completed successfully.")
-    else:
-        print(f"\n❌ Sync failed. Errors: {result['errors']}")
-
-    print("\n--- Running Sync Log Example ---")
-    # This will now show the log from the run we just completed.
-    log_result = sync_orchestration(
-        basename="my_calculator",
-        language="python",
-        log=True
-    )