# -*- coding: utf-8 -*-
"""Comitari - Offline, read-only HardeningKitty audit package for WAPT.

All endpoint operations use WAPT setuphelpers where an appropriate helper exists.
WAPT verifies the package signature; vendor-manifest.json additionally checks the
pinned HardeningKitty payload, including after it is installed.

Build requires update_package.py to prepare vendor/ BEFORE WAPT build/sign/upload.
"""

from setuphelpers import *
import collections
import csv
import datetime
import io
import json
import locale
import os
import uuid

uninstallkey = []

PACKAGE_NAME = 'comi-audit-hardeningkitty'
AUDIT_SECTION = 'hardeningkitty'
AUDIT_KEY = 'hardeningkitty'
MAX_AUDIT_JSON_BYTES = 2 * 1024 * 1024
UPSTREAM_COMMIT = 'acf9ecee3d044a9337df7aad80e730cbd746e8de'
PROGRAM_DATA = os.environ.get('ProgramData', r'C:\ProgramData')
INSTALL_ROOT = makepath(PROGRAM_DATA, 'Comitari', 'Audit-HardeningKitty')
REPORT_ROOT = makepath(PROGRAM_DATA, 'Comitari', 'Audit-HardeningKitty-Reports')
OWNER_MARKER = '.comi-audit-hardeningkitty-owner'

# Keep aligned with runner.ps1 supported auto-selected baselines.
REQUIRED_VENDOR_FILES = (
    'HardeningKitty.psm1', 'HardeningKitty.psd1', 'LICENSE',
    'lists/hardeningkitty_lists_manifest.psd1',
    'lists/hardeningkitty_lists_manifest.psd1.p7s',
    'lists/finding_list_0x6d69636b_machine.csv',
    'lists/finding_list_msft_security_baseline_windows_10_20h2_21h1_machine.csv',
    'lists/finding_list_msft_security_baseline_windows_10_21h2_machine.csv',
    'lists/finding_list_msft_security_baseline_windows_10_22h2_machine.csv',
    'lists/finding_list_msft_security_baseline_windows_11_21h2_machine.csv',
    'lists/finding_list_msft_security_baseline_windows_11_22h2_machine.csv',
    'lists/finding_list_msft_security_baseline_windows_11_23h2_machine.csv',
    'lists/finding_list_msft_security_baseline_windows_11_24h2_machine.csv',
    'lists/finding_list_msft_security_baseline_windows_11_25h2_machine.csv',
    'lists/finding_list_cis_microsoft_windows_server_2012_r2_3.0.0_machine.csv',
    'lists/finding_list_cis_microsoft_windows_server_2016_1607_3.0.0_machine.csv',
    'lists/finding_list_cis_microsoft_windows_server_2019_1809_3.0.0_machine.csv',
    'lists/finding_list_msft_security_baseline_windows_server_2022_21h2_dc_machine.csv',
    'lists/finding_list_msft_security_baseline_windows_server_2022_21h2_member_machine.csv',
    'lists/finding_list_msft_security_baseline_windows_server_2025_24h2_dc_machine.csv',
    'lists/finding_list_msft_security_baseline_windows_server_2025_24h2_member_machine.csv',
)

# A staged/installed directory stores the vendor files at its root (NOT /vendor).
RUNTIME_FILES = ('runner.ps1', 'audit_config.json')
# Old package versions included this file. It is no longer required or trusted.
LEGACY_FILES = ('package-manifest.json',)


def _package_source():
    """Use WAPT's extraction directory, not cwd or the launcher's path."""
    root = globals().get('basedir')
    if not root:
        root = os.path.dirname(os.path.abspath(__file__)) if globals().get('__file__') else None
    if not root:
        raise RuntimeError('WAPT basedir is not available.')
    root = os.path.abspath(root)
    if not isfile(makepath(root, 'runner.ps1')) or not isfile(makepath(root, 'audit_config.json')):
        raise RuntimeError('WAPT package source is missing runner.ps1 or audit_config.json: %s' % root)
    return root


def _load_json(path):
    return json_load_file(path)


def _save_json(path, payload):
    json_write_file(path, payload, indent=2, sort_keys=True)


def _safe_vendor_name(name):
    return (isinstance(name, str) and bool(name) and '\\' not in name and
            ':' not in name and not name.startswith('/') and
            all(part not in ('', '.', '..') for part in name.split('/')))


def _verify_vendor(root, runtime=False):
    """Verify all publisher payload hashes and reject unexpected files/symlinks."""
    manifest_file = makepath(root, 'vendor-manifest.json')
    if not isfile(manifest_file):
        raise RuntimeError('HardeningKitty vendor-manifest.json is absent from the signed WAPT package. Run Update Package, then BUILD/SIGN/UPLOAD the new .wapt.')
    manifest = _load_json(manifest_file)
    if not isinstance(manifest, dict) or manifest.get('upstream_commit') != UPSTREAM_COMMIT:
        raise RuntimeError('Unexpected HardeningKitty vendor revision.')
    files = manifest.get('files')
    if not isinstance(files, dict) or len(files) < 50:
        raise RuntimeError('HardeningKitty vendor file inventory is invalid/incomplete.')
    missing = set(REQUIRED_VENDOR_FILES).difference(files)
    if missing:
        raise RuntimeError('HardeningKitty vendor manifest missing: %s' % ', '.join(sorted(missing)))

    root_real = os.path.realpath(root)
    expected = {'vendor-manifest.json'}
    for relative, checksum in files.items():
        if not _safe_vendor_name(relative):
            raise RuntimeError('Unsafe vendor filename: %r' % relative)
        if not isinstance(checksum, str) or len(checksum) != 64 or any(c not in '0123456789abcdefABCDEF' for c in checksum):
            raise RuntimeError('Invalid vendor SHA-256 entry: %s' % relative)
        filename = os.path.realpath(makepath(root, *relative.split('/')))
        if not filename.startswith(root_real + os.sep):
            raise RuntimeError('Vendor filename escaped installation directory: %s' % relative)
        if not isfile(filename) or get_sha256(filename).lower() != checksum.lower():
            raise RuntimeError('Missing/altered vendor asset: %s' % relative)
        expected.add(relative)

    if runtime:
        for filename in RUNTIME_FILES:
            if not isfile(makepath(root, filename)):
                raise RuntimeError('Installed WAPT runtime file missing: %s' % filename)
        expected.update(RUNTIME_FILES)
        # Older installed versions may contain this obsolete manifest.
        expected.update(name for name in LEGACY_FILES if isfile(makepath(root, name)))
        if isfile(makepath(root, OWNER_MARKER)):
            expected.add(OWNER_MARKER)

    actual = set()
    for directory, folders, names in os.walk(root, followlinks=False):
        for name in folders + names:
            if os.path.islink(makepath(directory, name)):
                raise RuntimeError('HardeningKitty directory contains a symlink; refusing it.')
        for name in names:
            actual.add(os.path.relpath(makepath(directory, name), root).replace(os.sep, '/'))
    if actual != expected:
        raise RuntimeError('Unexpected/missing files in HardeningKitty directory: %s' %
                           ', '.join(sorted(actual.symmetric_difference(expected))[:10]))
    return manifest


def _owned_install(root):
    """Never replace/delete a directory belonging to someone else."""
    if os.path.islink(root) or not isdir(root):
        return False
    try:
        marker = makepath(root, OWNER_MARKER)
        if isfile(marker):
            with open(marker, encoding='ascii') as stream:
                if stream.read().strip() != PACKAGE_NAME:
                    return False
        # Legacy upgrade without owner marker: require intact original payload.
        _verify_vendor(root, runtime=True)
        return True
    except Exception:
        return False


def _secure_directory(path):
    """Protect this directory; let its children inherit SYSTEM/Admin rights.

    Do not use /inheritance:r /T on every file. That removes inherited
    permissions from each individual file and is unnecessary for a freshly
    staged directory whose children inherit from its protected parent.
    """
    if os.name != 'nt':
        return
    # Grant first: avoid losing the installer's access before the new ACEs exist.
    # OI/CI make newly created files and subdirectories inherit the grants.
    run(['icacls.exe', path, '/grant:r',
         '*S-1-5-18:(OI)(CI)F', '*S-1-5-32-544:(OI)(CI)F'],
        shell=False, timeout=120, accept_returncodes=[0])
    # Only protect the directory itself. Existing descendants keep inheriting
    # from it; unlike /T, this does not strip every file's inherited ACEs.
    run(['icacls.exe', path, '/inheritance:r'],
        shell=False, timeout=120, accept_returncodes=[0])


def install():
    """Install files from WAPT basedir, offline; stage then swap atomically."""
    source = _package_source()
    vendor = makepath(source, 'vendor')
    manifest = _verify_vendor(vendor)
    parent = os.path.dirname(INSTALL_ROOT)
    mkdirs(parent)
    staging = INSTALL_ROOT + '.staging-' + uuid.uuid4().hex
    previous = INSTALL_ROOT + '.previous-' + uuid.uuid4().hex
    old_moved = False
    try:
        if os.path.lexists(INSTALL_ROOT) and not _owned_install(INSTALL_ROOT):
            raise RuntimeError('Install directory already exists but is not owned by this WAPT package.')
        mkdirs(staging)
        _secure_directory(staging)
        copytree2(vendor, staging, follow_symlinks=False)
        for filename in RUNTIME_FILES:
            filecopyto(makepath(source, filename), makepath(staging, filename))
        _verify_vendor(staging, runtime=True)
        with open(makepath(staging, OWNER_MARKER), 'w', encoding='ascii') as stream:
            stream.write(PACKAGE_NAME + '\n')
        _secure_directory(staging)
        # A security operation can change ACLs. Verify readability and all
        # checksums AFTER the final ACL change, before replacing a working install.
        _verify_vendor(staging, runtime=True)
        mkdirs(REPORT_ROOT)
        _secure_directory(REPORT_ROOT)
        if isdir(INSTALL_ROOT):
            os.rename(INSTALL_ROOT, previous)
            old_moved = True
        try:
            os.rename(staging, INSTALL_ROOT)
        except Exception:
            if old_moved and isdir(previous) and not os.path.exists(INSTALL_ROOT):
                os.rename(previous, INSTALL_ROOT)
                old_moved = False
            raise
        if old_moved and isdir(previous):
            try:
                remove_tree(previous)
            except Exception as exc:
                print('Previous owned installation preserved for manual cleanup: %s' % type(exc).__name__)
    finally:
        if isdir(staging):
            remove_tree(staging, ignore_errors=True)
    print('HardeningKitty installed offline: %s (%s verified files)' %
          (manifest['upstream_commit'], len(manifest['files'])))


def uninstall():
    if os.path.lexists(INSTALL_ROOT):
        if not _owned_install(INSTALL_ROOT):
            raise RuntimeError('Uninstall refused: directory is unverified or not owned by this package.')
        remove_tree(INSTALL_ROOT)
    print('HardeningKitty removed. Historical reports kept: %s' % REPORT_ROOT)


def _powershell_executable():
    """Use native 64-bit PowerShell when WAPT is a 32-bit process."""
    system_root = os.environ.get('SystemRoot', r'C:\Windows')
    sysnative = makepath(system_root, 'Sysnative', 'WindowsPowerShell', 'v1.0', 'powershell.exe')
    if isfile(sysnative):
        return sysnative
    native = makepath(system32(), 'WindowsPowerShell', 'v1.0', 'powershell.exe')
    if isfile(native):
        return native
    raise RuntimeError('Windows PowerShell 5.1 is missing.')

def _read_report(path, publish_current_values=False):
    fieldnames, rows = _read_csv_rows(path)
    required = {'ID', 'Category', 'Name', 'Severity', 'Result', 'Recommended', 'TestResult', 'SeverityFinding'}
    if not fieldnames or not required.issubset(fieldnames):
        raise RuntimeError('Unexpected HardeningKitty CSV headers: %r' % (fieldnames,))
    if not rows:
        raise RuntimeError('HardeningKitty CSV is empty; refuse a false OK.')

    counts = {'passed': 0, 'high': 0, 'medium': 0, 'low': 0, 'unknown': 0,
              'ignored': 0, 'not_applicable': 0, 'excluded': 0, 'total': 0, 'executed': 0}
    findings = []
    seen = set()
    for item in rows:
        if None in item:
            raise RuntimeError('HardeningKitty CSV row has an unexpected number of columns.')
        control_id = str(item.get('ID') or '').strip()
        if not control_id or control_id in seen:
            raise RuntimeError('HardeningKitty CSV contains a blank or duplicate control ID: %r' % (control_id,))
        seen.add(control_id)
        level = (item.get('Severity') or '').strip().lower()
        outcome = (item.get('TestResult') or '').strip().lower()
        if outcome == 'passed' and level == 'passed':
            counts['passed'] += 1
            counts['total'] += 1
            counts['executed'] += 1
            continue
        if outcome in ('not applicable', 'not_applicable', 'n/a'):
            counts['not_applicable'] += 1
            continue
        if outcome in ('ignored', 'skipped', 'not evaluated', 'not_evaluated'):
            counts['ignored'] += 1
            counts['total'] += 1
            level = 'ignored'
        elif (outcome in ('failed', 'fail') and level in ('high', 'medium', 'low') and
              (item.get('SeverityFinding') or '').strip().lower() == level):
            counts[level] += 1
            counts['total'] += 1
            counts['executed'] += 1
        else:
            counts['unknown'] += 1
            counts['total'] += 1
            level = 'unknown'
        finding = {
            'id': str(item.get('ID') or ''),
            'category': str(item.get('Category') or ''),
            'name': str(item.get('Name') or ''),
            'severity': level,
            'result': str(item.get('TestResult') or ''),
        }
        # Avoid uploading potentially sensitive local configuration by default.
        if publish_current_values:
            finding['current'] = str(item.get('Result') or '')
            finding['recommended'] = str(item.get('Recommended') or '')
        findings.append(finding)
    counts['compliant_percent'] = round(100.0 * counts['passed'] / counts['total'], 2) if counts['total'] else 0.0
    return counts, findings


def _read_csv_rows(path):
    """Read PowerShell 5.1 output strictly across UTF encodings and local code pages."""
    with open(path, 'rb') as fh:
        data = fh.read()
    if data.startswith((b'\xff\xfe', b'\xfe\xff')):
        text = data.decode('utf-16')
    elif data.startswith(b'\xef\xbb\xbf'):
        text = data.decode('utf-8-sig')
    else:
        try:
            text = data.decode('utf-8')
        except UnicodeDecodeError:
            encoding = locale.getpreferredencoding(False)
            try:
                text = data.decode(encoding)
            except (UnicodeDecodeError, LookupError) as exc:
                raise RuntimeError('HardeningKitty CSV encoding is not supported.') from exc
    reader = csv.DictReader(io.StringIO(text, newline=''))
    return reader.fieldnames, list(reader)


def _check_coverage(baseline_path, report_path, counts, findings, exclusions=None, benchmark=None):
    """Mark tests skipped by HardeningKitty as unknown, never as Passed.

    HardeningKitty can skip a check (e.g. a tool is missing). Such an item
    may be absent from the output CSV. Counting only present rows could give
    a misleading green result, so compare to the original finding list.
    """
    fieldnames, expected_rows = _read_csv_rows(baseline_path)
    required = {'ID', 'Category', 'Name'}
    if not fieldnames or not required.issubset(fieldnames):
        raise RuntimeError('Selected finding list has unexpected headers: %r' % (fieldnames,))
    _, actual_rows = _read_csv_rows(report_path)
    if not expected_rows:
        raise RuntimeError('Finding list is empty.')

    def control_id(row):
        return str(row.get('ID') or '').strip()

    # IDs are stable across localizations; labels/categories may be translated by
    # the installed HardeningKitty language data and are not join keys.
    actual = collections.Counter(control_id(row) for row in actual_rows)
    if any(value > 1 for value in actual.values()):
        raise RuntimeError('HardeningKitty report contains duplicate control IDs.')
    expected = collections.Counter(control_id(row) for row in expected_rows)
    if any(not key for key in expected) or any(value > 1 for value in expected.values()):
        raise RuntimeError('Selected finding list contains duplicate control IDs.')
    if actual - expected:
        raise RuntimeError('HardeningKitty report contains controls outside the selected benchmark.')
    allowed = (exclusions or {}).get(benchmark or '', {})
    if (not isinstance(allowed, dict) or
            any(not isinstance(excluded_id, str) or excluded_id not in set(expected) for excluded_id in allowed) or
            any(not isinstance(reason, str) or not reason.strip() for reason in allowed.values())):
        raise RuntimeError('intentional_exclusions must map benchmark IDs to non-empty review reasons.')
    missing = []
    for row in expected_rows:
        key = control_id(row)
        if actual[key]:
            actual[key] -= 1
        else:
            if key in allowed:
                counts['excluded'] += 1
                findings.append({'id': key, 'name': str(row.get('Name') or ''),
                                 'category': str(row.get('Category') or ''),
                                 'severity': 'not_applicable', 'result': str(allowed[key])})
            else:
                missing.append({'id': key, 'name': str(row.get('Name') or ''),
                                'category': str(row.get('Category') or ''),
                                'severity': 'unknown', 'result': 'Skipped / no CSV result'})
    if missing:
        findings.extend(missing)
        counts['unknown'] += len(missing)
        counts['total'] += len(missing)
    counts['skipped'] = len(missing)
    counts['expected'] = len(expected_rows)
    counts['compliant_percent'] = round(100.0 * counts['passed'] / counts['total'], 2) if counts['total'] else 0.0
    return counts, findings


def _status_from_counts(counts, config):
    if counts['total'] == 0 or counts['unknown'] or counts.get('ignored') or counts.get('skipped'):
        return 'ERROR'
    if counts['high']:
        return 'ERROR' if config['error_on_high'] else 'WARNING'
    if counts['medium'] or (counts['low'] and config['warn_on_low']):
        return 'WARNING'
    return 'OK'


def _build_audit_document(summary, findings, report_file, baseline_file, config):
    """One native JSON object, with all benchmark controls addressable by ID.

    The detailed CSV stays local. Observed values are excluded by default
    because some settings can reveal accounts or environment-specific details.
    """
    _, baseline_rows = _read_csv_rows(baseline_file)
    _, report_rows = _read_csv_rows(report_file)
    reported = {str(row.get('ID') or '').strip(): row for row in report_rows}
    problems = {str(row['id']): row for row in findings}
    category_stats = {}
    controls = {}
    source = summary['source']
    status_labels = {
        'passed': 'Conforme',
        'failed': 'Non conforme',
        'skipped': 'Non exécuté',
        'ignored': 'Ignoré',
        'excluded': 'Exclu (justifié)',
        'not_applicable': 'Non applicable',
        'unknown': 'Inconnu',
    }
    severity_labels = {
        'high': 'Élevée',
        'medium': 'Moyenne',
        'low': 'Faible',
        'unknown': 'Indéterminée',
    }
    explanations = {
        'passed': 'Le réglage satisfait au benchmark sélectionné.',
        'failed': 'Le réglage ne satisfait pas au benchmark sélectionné. Vérifier la valeur cible et la politique appliquée avant toute correction.',
        'skipped': 'HardeningKitty n\'a pas retourné de résultat pour ce contrôle. Il ne peut pas etre considéré conforme.',
        'ignored': 'Le contrôle n\'a pas été évalué lors de cet audit.',
        'excluded': 'Contrôle exclu intentionnellement. La justification doit être examinée dans la politique d\'audit.',
        'not_applicable': 'Contrôle marqué comme non applicable.',
        'unknown': 'Résultat insuffisant pour déterminer la conformité.',
    }

    for baseline in baseline_rows:
        control_id = str(baseline.get('ID') or '').strip()
        recorded = reported.get(control_id)
        issue = problems.get(control_id)
        category = str(baseline.get('Category') or '')
        risk = str(baseline.get('Severity') or '').strip().lower()
        if risk not in ('high', 'medium', 'low'):
            risk = 'unknown'

        if recorded is None:
            outcome = 'excluded' if issue and issue.get('severity') == 'not_applicable' else 'skipped'
        elif str(recorded.get('TestResult') or '').strip().lower() in ('not applicable', 'not_applicable', 'n/a'):
            outcome = 'not_applicable'
        elif issue is None:
            outcome = 'passed'
        elif issue['severity'] in ('high', 'medium', 'low'):
            outcome = 'failed'
        elif issue['severity'] in ('ignored', 'not_applicable', 'unknown'):
            outcome = issue['severity']
        else:
            outcome = 'unknown'

        control = {
            'id': control_id,
            'name': str(baseline.get('Name') or ''),
            'category': category,
            'status': outcome,
            'status_label': status_labels.get(outcome, outcome),
            'severity': risk,
            'severity_label': severity_labels.get(risk, risk),
            'explanation': explanations.get(outcome, explanations['unknown']),
        }
        # Only copy a description when the selected benchmark actually carries it.
        description = str(baseline.get('Description') or
                          (recorded.get('Description') if recorded else '') or '').strip()
        if description:
            control['description'] = description
        method = str(baseline.get('Method') or '').strip()
        if method:
            control['method'] = method
        argument = str(baseline.get('MethodArgument') or '').strip()
        if argument:
            control['method_argument'] = argument
        operator_field = ('OperatorIntune' if source == 'Intune' and
                          method.lower() in ('registry', 'registrylist') else 'Operator')
        operator = str(baseline.get(operator_field) or '').strip()
        if operator:
            control['operator'] = operator
        path_field = ('RegistryPathIntune' if source == 'Intune' and
                      method.lower() in ('registry', 'registrylist') else 'RegistryPath')
        registry_path = str(baseline.get(path_field) or '').strip()
        if registry_path:
            control['registry_path'] = registry_path
        registry_item = str(baseline.get('RegistryItem') or '').strip()
        if registry_item:
            control['registry_item'] = registry_item

        recommended = ''
        if recorded is not None:
            recommended = str(recorded.get('Recommended') or '')
        if not recommended:
            # The recommendation is policy data, not an observed host value.
            field = ('RecommendedValueIntune' if source == 'Intune' and
                     method.lower() in ('registry', 'registrylist') else 'RecommendedValue')
            recommended = str(baseline.get(field) or '')
        if recommended:
            control['recommended'] = recommended
        if recorded is not None and config.get('publish_current_values', False):
            control['current'] = str(recorded.get('Result') or '')
        if outcome in ('excluded', 'skipped', 'ignored', 'unknown'):
            control['reason'] = (str(issue.get('result') or '') if issue else 'Not evaluated')

        controls[control_id] = control
        cat = category_stats.setdefault(category, {
            'expected': 0, 'passed': 0, 'high': 0, 'medium': 0, 'low': 0,
            'unknown': 0, 'skipped': 0, 'ignored': 0, 'excluded': 0,
            'not_applicable': 0,
        })
        cat['expected'] += 1
        if outcome == 'failed':
            actual_risk = str(issue.get('severity') or risk).lower()
            if actual_risk not in ('high', 'medium', 'low'):
                actual_risk = 'unknown'
            cat[actual_risk] += 1
        elif outcome == 'passed':
            cat['passed'] += 1
        elif outcome in ('skipped', 'unknown', 'ignored'):
            cat[outcome] += 1
            if outcome == 'skipped':
                cat['unknown'] += 1  # A skipped test is not a pass.
        elif outcome == 'excluded':
            cat['excluded'] += 1
        elif outcome == 'not_applicable':
            cat['not_applicable'] += 1

    for cat in category_stats.values():
        denominator = cat['expected'] - cat['excluded'] - cat['not_applicable']
        cat['compliant_percent'] = (round(100.0 * cat['passed'] / denominator, 2)
                                    if denominator else 0.0)

    evaluated = summary['high'] + summary['medium'] + summary['low']
    if summary['unknown'] or summary['ignored'] or summary.get('skipped'):
        compliance = 'incomplete'
    elif evaluated:
        compliance = 'non_compliant'
    else:
        compliance = 'compliant'

    # Keep only a compact index here; the full control object below is canonical.
    issue_controls = [c for c in controls.values() if c['status'] != 'passed'
                      and c['status'] not in ('excluded', 'not_applicable')]
    level_sort = {'high': 0, 'medium': 1, 'low': 2, 'unknown': 3}
    issue_controls.sort(key=lambda c: (0 if c['status'] == 'failed' else 1,
                                       level_sort.get(c['severity'], 3),
                                       c['category'], c['id']))
    priority_fields = ('id', 'category', 'status', 'severity')
    priority_findings = [
        {key: control[key] for key in priority_fields if key in control}
        for control in issue_controls[:25]
    ]

    doc = {
        'schema_version': 2,
        'status': summary['status'],
        'compliance': compliance,
        'scan': {
            'date_utc': summary['last_run_utc'],
            'engine': 'HardeningKitty',
            'engine_commit': summary['upstream_commit'],
            'profile': config.get('profile', 'auto'),
            'source': source,
        },
        'host': {
            'os_caption': summary['os_caption'],
            'build': summary['os_build'],
            'display_version': summary['display_version'],
        },
        'benchmark': {
            'name': summary['benchmark'],
            'expected': summary['expected'],
        },
        'summary': {
            'expected': summary['expected'],
            'checked': summary['total'],
            'executed': summary['executed'],
            'passed': summary['passed'],
            'failed': evaluated,
            'high': summary['high'],
            'medium': summary['medium'],
            'low': summary['low'],
            'unknown': summary['unknown'],
            'skipped': summary['skipped'],
            'ignored': summary['ignored'],
            'excluded': summary['excluded'],
            'not_applicable': summary['not_applicable'],
            'compliant_percent': summary['compliant_percent'],
        },
        'categories': category_stats,
        'controls': controls,
        'priority_findings': priority_findings,
        'priority_total': len(issue_controls),
        'privacy': {
            'current_values_published': bool(config.get('publish_current_values', False)),
        },
    }
    if len(controls) != summary['expected']:
        raise RuntimeError('Structured audit JSON does not cover the complete benchmark.')
    # This is a size check only: the dict itself is sent to WAPT, not a string.
    json_size = len(json.dumps(doc, ensure_ascii=False).encode('utf-8'))
    if json_size > MAX_AUDIT_JSON_BYTES:
        raise RuntimeError('Structured audit data too large (%s bytes).' % json_size)
    print('HardeningKitty WAPT JSON: %s controls, %s bytes' % (len(controls), json_size))
    return doc


def _publish_audit(document, config):
    """Write exactly one native JSON value, as in WAPT's Wazuh example."""
    if not config.get('publish_to_wapt', True):
        return
    wapt = globals().get('WAPT')
    write = getattr(wapt, 'write_audit_data_if_changed', None)
    if not callable(write):
        raise RuntimeError('WAPT.write_audit_data_if_changed is unavailable.')
    write(AUDIT_SECTION, AUDIT_KEY, document, max_count=2, keep_days=120)


def _publish_failure(exc):
    """Replace a previous successful snapshot with an explicit error snapshot."""
    wapt = globals().get('WAPT')
    write = getattr(wapt, 'write_audit_data_if_changed', None)
    if not callable(write):
        return
    document = {
        'schema_version': 2,
        'status': 'ERROR',
        'compliance': 'incomplete',
        'scan': {
            'date_utc': datetime.datetime.now(datetime.timezone.utc).isoformat(),
            'engine': 'HardeningKitty',
        },
        'error': {
            'code': 'AUDIT_FAILED',
            'type': type(exc).__name__,
            'description': 'Audit failed; no current compliance assessment is available. Check the local WAPT audit log.',
        },
    }
    write(AUDIT_SECTION, AUDIT_KEY, document, max_count=2, keep_days=120)


def _trim_reports(keep):
    if not isdir(REPORT_ROOT):
        return
    paths = [makepath(REPORT_ROOT, name) for name in os.listdir(REPORT_ROOT)
             if name.startswith('run-') and isdir(makepath(REPORT_ROOT, name))
             and not os.path.islink(makepath(REPORT_ROOT, name))]
    paths.sort(key=lambda p: os.path.getmtime(p), reverse=True)
    for path in paths[max(1, int(keep)):]:
        remove_tree(path)


def _audit_impl():
    manifest = _verify_vendor(INSTALL_ROOT, runtime=True)
    config = _load_json(makepath(INSTALL_ROOT, 'audit_config.json'))
    if config.get('source') not in ('GPO', 'Intune'):
        raise RuntimeError('audit_config.json: source must be GPO or Intune.')
    profile = str(config.get('profile', 'auto'))
    if not isinstance(config.get('error_on_high'), bool) or not isinstance(config.get('warn_on_low'), bool):
        raise RuntimeError('Invalid audit threshold settings.')
    if not isinstance(config.get('publish_to_wapt', True), bool) or not isinstance(config.get('publish_current_values', False), bool):
        raise RuntimeError('Invalid WAPT publication/privacy settings.')
    retain = config.get('retain_audit_runs', 8)
    max_log = config.get('max_findings_in_log', 15)
    if isinstance(retain, bool) or not isinstance(retain, int) or not 1 <= retain <= 100:
        raise RuntimeError('retain_audit_runs must be 1..100.')
    if isinstance(max_log, bool) or not isinstance(max_log, int) or not 0 <= max_log <= 100:
        raise RuntimeError('max_findings_in_log must be 0..100.')
    exclusions = config.get('intentional_exclusions', {})
    if not isinstance(exclusions, dict):
        raise RuntimeError('intentional_exclusions must be an object.')
    timeout = config.get('timeout_seconds', 900)
    if isinstance(timeout, bool) or not isinstance(timeout, int) or not 30 <= timeout <= 7200:
        raise RuntimeError('timeout_seconds must be 30..7200.')

    mkdirs(REPORT_ROOT)
    _secure_directory(REPORT_ROOT)
    run_id = datetime.datetime.now(datetime.timezone.utc).strftime('%Y%m%d-%H%M%S') + '-' + uuid.uuid4().hex[:8]
    run_dir = makepath(REPORT_ROOT, 'run-' + run_id)
    mkdirs(run_dir)
    _secure_directory(run_dir)
    csv_file = makepath(run_dir, 'hardeningkitty.csv')
    log_file = makepath(run_dir, 'powershell.log')
    metadata_file = makepath(run_dir, 'host.json')

    # run() is WAPT's process helper and handles timeouts and exit statuses.
    # No subprocess, shell interpolation, or endpoint Internet access.
    command = [_powershell_executable(), '-NoLogo', '-NoProfile', '-NonInteractive',
               '-ExecutionPolicy', 'Bypass', '-File', makepath(INSTALL_ROOT, 'runner.ps1'),
               '-InstallRoot', INSTALL_ROOT, '-CsvFile', csv_file,
               '-MetadataFile', metadata_file, '-Source', config['source'], '-Profile', profile]
    with open(log_file, 'w', encoding='utf-8') as log:
        def record(line):
            text = str(line)
            log.write(text if text.endswith('\n') else text + '\n')
            log.flush()
        try:
            result = run(command, shell=False, timeout=timeout,
                         accept_returncodes=[0], on_write=record)
            if log.tell() == 0 and result:
                record(result)
        except Exception as exc:
            log.write('Execution failed: %s\n' % type(exc).__name__)
            raise RuntimeError('HardeningKitty runner failed (%s). See %s' %
                               (type(exc).__name__, log_file)) from exc

    if not isfile(csv_file) or not isfile(metadata_file):
        raise RuntimeError('HardeningKitty runner did not produce CSV and host metadata.')
    metadata = _load_json(metadata_file)
    benchmark = str(metadata.get('benchmark', ''))
    if (not benchmark.startswith('finding_list_') or not benchmark.endswith('.csv') or
            '/' in benchmark or '\\' in benchmark or '..' in benchmark):
        raise RuntimeError('Unsafe or invalid benchmark from PowerShell runner.')
    baseline = makepath(INSTALL_ROOT, 'lists', benchmark)
    if not isfile(baseline):
        raise RuntimeError('Runner selected a benchmark absent from vendor assets: %s' % benchmark)

    counts, findings = _read_report(csv_file, config.get('publish_current_values', False))
    counts, findings = _check_coverage(baseline, csv_file, counts, findings, exclusions, benchmark)
    status = _status_from_counts(counts, config)
    summary = dict(counts)
    summary.update({
        'status': status,
        'benchmark': benchmark,
        'os_build': str(metadata['build']),
        'os_caption': str(metadata.get('os_caption', '')),
        'display_version': str(metadata.get('display_version', '')),
        'source': metadata['source'],
        'upstream_commit': manifest['upstream_commit'],
        'last_error': '',
        'last_run_utc': datetime.datetime.now(datetime.timezone.utc).isoformat(),
        'report_csv': csv_file,
        'report_log': log_file,
    })
    _save_json(makepath(run_dir, 'summary.json'), summary)
    document = _build_audit_document(summary, findings, csv_file, baseline, config)
    _save_json(makepath(run_dir, 'wapt-audit.json'), document)
    _publish_audit(document, config)
    print('HardeningKitty: %s | %s | checked=%s passed=%s high=%s medium=%s low=%s unknown=%s skipped=%s | compliant=%s%%' %
          (status, benchmark, counts['total'], counts['passed'], counts['high'],
           counts['medium'], counts['low'], counts['unknown'], counts['skipped'], counts['compliant_percent']))
    priority = {'high': 0, 'medium': 1, 'low': 2, 'unknown': 3}
    for finding in sorted(findings, key=lambda f: (priority.get(f['severity'], 9), f['id']))[:max_log]:
        print('[%s] %s | %s | %s' %
              (finding['severity'].upper(), finding['id'], finding['category'], finding['name']))
    print('Full CSV report: %s' % csv_file)
    _trim_reports(retain)
    return status


def audit():
    try:
        return _audit_impl()
    except Exception as exc:
        # Log enough detail locally, but publish only a sanitized technical error.
        print('HardeningKitty audit ERROR: %s' % exc)
        try:
            _publish_failure(exc)
        except Exception as publish_error:
            print('HardeningKitty: failed to publish ERROR JSON (%s).' % type(publish_error).__name__)
        try:
            config = _load_json(makepath(INSTALL_ROOT, 'audit_config.json'))
            _trim_reports(config.get('retain_audit_runs', 8))
        except Exception:
            pass
        return 'ERROR'