weflow-cli 1.17.0 → 1.19.1

package/scripts/nt_decrypt.py

@@ -1,1954 +1,2023 @@
-#!/usr/bin/env python3
-"""
-WeChat NT (4.x) Database Access Tool
-Uses sqlcipher3 to decrypt and query NT-format databases.
-"""
-import sys
-import os
-import json
-import html
-import re
-import hmac
-import struct
-import hashlib
-import ctypes
-from ctypes import wintypes, c_void_p, c_size_t, create_string_buffer, byref, sizeof
-from pathlib import Path
-
-sqlcipher = None
-
-
-def require_sqlcipher():
-    """Load SQLCipher only for database operations, not path discovery."""
-    global sqlcipher
-    if sqlcipher is not None:
-        return sqlcipher
-    try:
-        from sqlcipher3 import dbapi2 as sqlcipher_module
-    except ImportError as error:
-        raise RuntimeError("需要 sqlcipher3: pip install sqlcipher3") from error
-    sqlcipher = sqlcipher_module
-    return sqlcipher
-
-# ========== Memory Scanner ==========
-PROCESS_VM_READ = 0x0010
-PROCESS_QUERY_INFORMATION = 0x0400
-MEM_COMMIT = 0x1000
-PAGE_NOACCESS = 0x01
-PAGE_GUARD = 0x100
-
-class MEMORY_BASIC_INFORMATION(ctypes.Structure):
-    _fields_ = [
-        ('BaseAddress', ctypes.c_void_p),
-        ('AllocationBase', ctypes.c_void_p),
-        ('AllocationProtect', wintypes.DWORD),
-        ('PartitionId', wintypes.WORD),
-        ('RegionSize', ctypes.c_size_t),
-        ('State', wintypes.DWORD),
-        ('Protect', wintypes.DWORD),
-        ('Type', wintypes.DWORD),
-    ]
-
-IS_WINDOWS = os.name == 'nt'
-
-if IS_WINDOWS:
-    kernel32 = ctypes.windll.kernel32
-    ReadProcessMemory = kernel32.ReadProcessMemory
-    ReadProcessMemory.argtypes = [wintypes.HANDLE, wintypes.LPCVOID, wintypes.LPVOID, ctypes.c_size_t, ctypes.POINTER(c_size_t)]
-    ReadProcessMemory.restype = wintypes.BOOL
-    VirtualQueryEx = kernel32.VirtualQueryEx
-    VirtualQueryEx.argtypes = [wintypes.HANDLE, wintypes.LPCVOID, ctypes.c_void_p, ctypes.c_size_t]
-    VirtualQueryEx.restype = ctypes.c_size_t
-else:
-    kernel32 = None
-
-
-def find_weixin_pid():
-    """Find WeChat process ID (Windows: pymem; Linux: /proc scan)."""
-    if not IS_WINDOWS:
-        return find_weixin_pid_linux()
-    try:
-        import pymem, pymem.process
-        for proc in pymem.process.list_processes():
-            try:
-                name = proc.szExeFile
-                if isinstance(name, bytes):
-                    name = name.decode('utf-8', errors='ignore')
-                if name.lower() == 'weixin.exe':
-                    return proc.th32ProcessID
-            except:
-                pass
-    except ImportError:
-        pass
-    return None
-
-
-_LINUX_WECHAT_COMMS = {'wechat', 'wechatappex', 'weixin'}
-_LINUX_EXE_PREFIX_DENY = ('python', 'bash', 'sh', 'zsh', 'node', 'perl', 'ruby', 'electron')
-
-
-def _is_linux_wechat_process(pid):
-    if pid == os.getpid():
-        return False
-    try:
-        with open(f'/proc/{pid}/comm', encoding='utf-8') as f:
-            comm = f.read().strip().lower()
-        if comm in _LINUX_WECHAT_COMMS:
-            return True
-        try:
-            exe = os.path.realpath(os.readlink(f'/proc/{pid}/exe'))
-        except OSError:
-            return False
-        name = os.path.basename(exe).lower()
-        if any(name.startswith(p) for p in _LINUX_EXE_PREFIX_DENY):
-            return False
-        return 'wechat' in name or 'weixin' in name
-    except (PermissionError, FileNotFoundError, ProcessLookupError):
-        return False
-
-
-def find_weixin_pid_linux():
-    """Find Linux WeChat main process (largest RSS among candidates)."""
-    best = None
-    best_rss = -1
-    try:
-        pids = os.listdir('/proc')
-    except OSError:
-        return None
-    for pid_str in pids:
-        if not pid_str.isdigit():
-            continue
-        pid = int(pid_str)
-        if not _is_linux_wechat_process(pid):
-            continue
-        try:
-            with open(f'/proc/{pid}/statm', encoding='utf-8') as f:
-                rss_kb = int(f.read().split()[1]) * 4
-        except (OSError, IndexError, ValueError):
-            rss_kb = 0
-        if rss_kb > best_rss:
-            best_rss = rss_kb
-            best = pid
-    return best
-
-
-def scan_memory_keys(pid):
-    """Scan process memory for x'<64hex_key><32hex_salt>' patterns.
-
-    Returns (keys, error): keys is a list of {"key","salt"} dicts,
-    error is None on success or 'permission' / 'gone' / 'not_windows'.
-    """
-    if not IS_WINDOWS:
-        return scan_memory_keys_linux(pid)
-
-    hProcess = kernel32.OpenProcess(PROCESS_VM_READ | PROCESS_QUERY_INFORMATION, False, pid)
-    if not hProcess:
-        return [], None
-
-    pattern = re.compile(rb"x'([0-9a-fA-F]{64})([0-9a-fA-F]{32})'")
-    keys_found = []
-    address = 0x10000
-
-    while address < 0x7FFFFFFFFFFF:
-        mbi = MEMORY_BASIC_INFORMATION()
-        result = VirtualQueryEx(hProcess, ctypes.c_void_p(address), ctypes.byref(mbi), sizeof(mbi))
-        if result == 0:
-            break
-
-        region_addr = mbi.BaseAddress or 0
-        region_size = mbi.RegionSize or 0
-
-        if (mbi.State == MEM_COMMIT and
-                region_size > 256 and region_size < 200 * 1024 * 1024 and
-                mbi.Protect not in (0, PAGE_NOACCESS, PAGE_GUARD)):
-
-            pos = region_addr
-            end = region_addr + region_size
-            while pos < end:
-                chunk_size = min(65536, end - pos)
-                buf = create_string_buffer(chunk_size)
-                bytesRead = c_size_t(0)
-                ok = ReadProcessMemory(hProcess, ctypes.c_void_p(pos), buf, chunk_size, byref(bytesRead))
-                if ok and bytesRead.value > 0:
-                    data = buf.raw[:bytesRead.value]
-                    for m in pattern.finditer(data):
-                        key_hex = m.group(1).decode()
-                        salt_hex = m.group(2).decode()
-                        keys_found.append((key_hex, salt_hex))
-                pos += chunk_size
-
-        address = region_addr + region_size
-
-    kernel32.CloseHandle(hProcess)
-
-    # Deduplicate
-    seen = set()
-    unique_keys = []
-    for k, s in keys_found:
-        pair = (k, s)
-        if pair not in seen:
-            seen.add(pair)
-            unique_keys.append({"key": k, "salt": s})
-
-    return unique_keys, None
-
-
-_LINUX_SKIP_MAPPINGS = {'[vdso]', '[vsyscall]', '[vvar]'}
-_LINUX_SKIP_PREFIXES = ('/usr/lib/', '/lib/', '/usr/share/')
-
-
-def scan_memory_keys_linux(pid):
-    """Scan /proc/<pid>/maps + /proc/<pid>/mem for the key pattern.
-
-    Requires root or CAP_SYS_PTRACE (or the target being a descendant
-    of this process when yama ptrace_scope=1).
-    """
-    regions = []
-    try:
-        with open(f'/proc/{pid}/maps', encoding='utf-8') as f:
-            for line in f:
-                parts = line.split()
-                if len(parts) < 2 or 'r' not in parts[1]:
-                    continue
-                if len(parts) >= 6:
-                    name = parts[5]
-                    if name in _LINUX_SKIP_MAPPINGS:
-                        continue
-                    name_lower = name.lower()
-                    if name.startswith(_LINUX_SKIP_PREFIXES) and \
-                            'wcdb' not in name_lower and 'wechat' not in name_lower and 'weixin' not in name_lower:
-                        continue
-                try:
-                    start_s, end_s = parts[0].split('-')
-                    start = int(start_s, 16)
-                    size = int(end_s, 16) - start
-                except ValueError:
-                    continue
-                if 0 < size < 500 * 1024 * 1024:
-                    regions.append((start, size))
-    except PermissionError:
-        return [], 'permission'
-    except (FileNotFoundError, ProcessLookupError):
-        return [], 'gone'
-
-    pattern = re.compile(rb"x'([0-9a-fA-F]{64})([0-9a-fA-F]{32})'")
-    keys_found = []
-    try:
-        with open(f'/proc/{pid}/mem', 'rb') as mem:
-            for base, size in regions:
-                try:
-                    mem.seek(base)
-                    data = mem.read(size)
-                except (OSError, ValueError):
-                    continue
-                for m in pattern.finditer(data):
-                    keys_found.append((m.group(1).decode(), m.group(2).decode()))
-    except PermissionError:
-        return [], 'permission'
-    except (FileNotFoundError, ProcessLookupError):
-        return [], 'gone'
-
-    seen = set()
-    unique_keys = []
-    for k, s in keys_found:
-        pair = (k, s)
-        if pair not in seen:
-            seen.add(pair)
-            unique_keys.append({"key": k, "salt": s})
-
-    return unique_keys, None
-
-
-# ========== NT Database Discovery ==========
-
-def _is_nt_account_dir(path):
-    return os.path.isdir(os.path.join(path, 'db_storage')) or \
-        os.path.isdir(os.path.join(path, 'Msg'))
-
-
-def _normalize_nt_root(root):
-    if not root:
-        return None
-    path = os.path.abspath(os.path.expandvars(os.path.expanduser(root)))
-    if os.path.isfile(path):
-        path = os.path.dirname(path)
-    if not os.path.isdir(path):
-        return None
-
-    try:
-        for entry in os.listdir(path):
-            candidate = os.path.join(path, entry)
-            if os.path.isdir(candidate) and _is_nt_account_dir(candidate):
-                return path
-    except OSError:
-        return None
-
-    for _ in range(6):
-        if _is_nt_account_dir(path):
-            return os.path.dirname(path)
-        parent = os.path.dirname(path)
-        if parent == path:
-            break
-        path = parent
-
-    return None
-
-
-def find_nt_databases(root=None):
-    """Find all NT-format databases under xwechat_files (message + contact)."""
-    if root:
-        normalized_root = _normalize_nt_root(root)
-        candidates = [normalized_root] if normalized_root else []
-    elif IS_WINDOWS:
-        candidates = [
-            os.path.expandvars(r'%USERPROFILE%\xwechat_files'),
-            os.path.expandvars(r'%USERPROFILE%\Documents\xwechat_files'),
-        ]
-    else:
-        home = os.path.expanduser('~')
-        candidates = [
-            os.path.join(home, '.local', 'share', 'xwechat_files'),
-            os.path.join(home, 'xwechat_files'),
-            os.path.join(home, 'Documents', 'xwechat_files'),
-            os.path.join(home, '文档', 'xwechat_files'),
-        ]
-    xwechat = None
-    for c in candidates:
-        if os.path.isdir(c):
-            xwechat = c
-            break
-    if not xwechat:
-        return []
-
-    databases = []
-    for wxid_dir in os.listdir(xwechat):
-        # Scan message databases
-        msg_storage = os.path.join(xwechat, wxid_dir, 'db_storage', 'message')
-        if os.path.isdir(msg_storage):
-            for f in os.listdir(msg_storage):
-                if f.endswith('.db') and not any(x in f for x in ['-shm', '-wal']):
-                    full_path = os.path.join(msg_storage, f)
-                    try:
-                        with open(full_path, 'rb') as fh:
-                            salt = fh.read(16)
-                        databases.append({
-                            "path": full_path,
-                            "name": f"message/{f}",
-                            "salt": salt.hex(),
-                            "size": os.path.getsize(full_path),
-                            "wxid": wxid_dir,
-                        })
-                    except:
-                        pass
-
-        # Scan contact database
-        contact_db = os.path.join(xwechat, wxid_dir, 'db_storage', 'contact', 'contact.db')
-        if os.path.isfile(contact_db):
-            try:
-                with open(contact_db, 'rb') as fh:
-                    salt = fh.read(16)
-                databases.append({
-                    "path": contact_db,
-                    "name": "contact/contact.db",
-                    "salt": salt.hex(),
-                    "size": os.path.getsize(contact_db),
-                    "wxid": wxid_dir,
-                })
-            except:
-                pass
-
-        # Scan SNS (朋友圈) database
-        sns_db = os.path.join(xwechat, wxid_dir, 'db_storage', 'sns', 'sns.db')
-        if os.path.isfile(sns_db):
-            try:
-                with open(sns_db, 'rb') as fh:
-                    salt = fh.read(16)
-                databases.append({
-                    "path": sns_db,
-                    "name": "sns/sns.db",
-                    "salt": salt.hex(),
-                    "size": os.path.getsize(sns_db),
-                    "wxid": wxid_dir,
-                })
-            except:
-                pass
-
-        # Scan favorites (收藏) database
-        fav_db = os.path.join(xwechat, wxid_dir, 'db_storage', 'favorite', 'favorite.db')
-        if os.path.isfile(fav_db):
-            try:
-                with open(fav_db, 'rb') as fh:
-                    salt = fh.read(16)
-                databases.append({
-                    "path": fav_db,
-                    "name": "favorite/favorite.db",
-                    "salt": salt.hex(),
-                    "size": os.path.getsize(fav_db),
-                    "wxid": wxid_dir,
-                })
-            except:
-                pass
-
-    return databases
-
-
-def find_contact_db_path(message_db_path):
-    """Derive contact.db path from message_0.db path.
-
-    message_0.db:  <xwechat_files>/<wxid>/db_storage/message/message_0.db
-    contact.db:    <xwechat_files>/<wxid>/db_storage/contact/contact.db
-    """
-    msg_dir = os.path.dirname(message_db_path)
-    wxid_dir = os.path.dirname(msg_dir)     # .../db_storage
-    xwechat_dir = os.path.dirname(wxid_dir) # .../<wxid>
-    contact_db = os.path.join(xwechat_dir, 'db_storage', 'contact', 'contact.db')
-    if os.path.isfile(contact_db):
-        return contact_db
-    return None
-
-
-def load_contact_names(contact_db_path, contact_key_hex, contact_salt_hex):
-    """Load wxid -> {remark, nick_name} map from contact.db.
-
-    Returns dict: {wxid: display_name}
-    display_name priority: remark > nick_name > alias > wxid
-
-    取不到就返回空表（名字是装饰性的，调用方会退回 wxid）。**连接必须在 finally
-    里关**：原先 `conn.close()` 写在 try 末尾，异常路径上泄漏——而在 Windows 上
-    一个没关的连接会把文件锁住（同一个坑早先在这个文件的分片读取里踩过一次）。
-    读了一半也保留已经拿到的名字，而不是整份丢掉。
-    """
-    if not contact_db_path or not contact_key_hex or not contact_salt_hex:
-        return {}
-    if not os.path.isfile(contact_db_path):
-        return {}
-
-    name_map = {}
-    conn = None
-    try:
-        raw_key = f"x'{contact_key_hex}{contact_salt_hex}'"
-        conn = require_sqlcipher().connect(contact_db_path)
-        c = conn.cursor()
-        c.execute(f'PRAGMA key = "{raw_key}";')
-
-        # contact.db schema: username, alias, remark, nick_name, ...
-        c.execute("SELECT username, COALESCE(NULLIF(remark,''), NULLIF(nick_name,''), NULLIF(alias,''), username) FROM contact")
-        for username, display in c.fetchall():
-            if username:
-                name_map[username] = display
-    except Exception:
-        pass
-    finally:
-        if conn is not None:
-            try:
-                conn.close()
-            except Exception:
-                pass
-    return name_map
-
-
-def filter_contacts(contacts, keyword, limit=0):
-    """按关键字过滤联系人，**然后**才截断到 `limit`。
-
-    顺序是这里的全部要点：先截断会让第 `limit` 个之后的人搜不到（实测 500 人的通讯录里，
-    按备注名找 10 个只命中 3 个）。关键字比对 `username + displayName + remark + nickname`，
-    大小写不敏感。
-    """
-    if not keyword:
-        return contacts[:limit] if limit else contacts
-    needle = keyword.lower()
-    hits = [
-        c for c in contacts
-        if needle in (c.get('username', '') + c.get('displayName', '')
-                      + c.get('remark', '') + c.get('nickname', '')).lower()
-    ]
-    return hits[:limit] if limit else hits
-
-
-def apply_contact_names(sessions, name_map):
-    """Apply contact names to session list, replacing bare wxid displayNames."""
-    if not name_map:
-        return sessions
-    for s in sessions:
-        username = s.get('username', '')
-        if username in name_map:
-            s['displayName'] = name_map[username]
-    return sessions
-
-
-def match_keys_to_databases(keys, databases):
-    """Match memory keys to databases by comparing salts."""
-    salt_to_key = {}
-    for k in keys:
-        salt_to_key[k["salt"]] = k["key"]
-
-    matched = []
-    for db in databases:
-        if db["salt"] in salt_to_key:
-            db["key"] = salt_to_key[db["salt"]]
-            matched.append(db)
-
-    return matched
-
-
-# ========== Database Operations ==========
-
-def connect_nt_db(db_path, key_hex, salt_hex):
-    """Connect to an NT database using sqlcipher3."""
-    raw_key = f"x'{key_hex}{salt_hex}'"
-    conn = require_sqlcipher().connect(db_path)
-    c = conn.cursor()
-    c.execute(f'PRAGMA key = "{raw_key}";')
-    return conn, c
-
-
-# WeChat rolls a conversation into a new shard over time, and every shard is
-# encrypted with its own key. Reading only the configured database therefore
-# shows a transcript that stops wherever the first shard's last write left off
-# - which is silent, because the query still succeeds. Only the HTML exporter
-# used to merge shards, so the same chat exported two different histories
-# depending on the format asked for.
-
-# 分片发现、密钥派生、列探测都在 nt_common 里——**只此一份**。
-# 原先 `nt_decrypt` 与 `export_chat_html` 各有一份，而契约并不相同。
-#
-# 这一行不能省：本文件被 `spec_from_file_location` 加载时（测试就是这么加的），
-# 解释器**不会**把脚本目录放进 sys.path——只有直接运行才会。少了它，单独跑
-# `nt_decrypt_shards_test.py` 会以 ModuleNotFoundError 失败，而全套一起跑却能过
-# （别的测试文件先插了路径）。那种"碰巧能过"正是要避免的。
-sys.path.insert(0, os.path.dirname(os.path.abspath(__file__)))
-from nt_common import (discover_message_shards, derive_database_key,
-                       table_columns, MESSAGE_ANCHOR_COLUMNS)
-
-
-def verify_passphrase_native(passphrase_hex, db_path, internal_key_hex=''):
-    """Check a passphrase against a database without sqlcipher3.
-
-    Verifies the SQLCipher page-1 HMAC directly with stdlib primitives:
-
-        mac_salt = header_salt XOR 0x3a
-        key      = PBKDF2-HMAC-SHA512(passphrase, header_salt, 256000, 32)
-        mac_key  = PBKDF2-HMAC-SHA512(key, mac_salt, 2, 32)
-        compare HMAC-SHA512(mac_key, page1_body || page_number=1) with the
-        digest stored at the end of page 1
-
-    Why this exists: the normal way to check a key is to open the database,
-    which needs the native library. When sqlcipher3 is missing or misbehaving,
-    that route says nothing about whether the key itself is right - and the
-    two failures look identical. This separates them.
-
-    Returns True (correct), False (wrong), or None (cannot tell - bad input,
-    file too small, or a passphrase that is not hex).
-    """
-    try:
-        raw_passphrase = bytes.fromhex(passphrase_hex)
-    except (TypeError, ValueError):
-        return None
-    if internal_key_hex:
-        # Some WeChat builds XOR the passphrase with a key embedded in
-        # Weixin.dll before deriving. Optional: the installs seen so far do
-        # not need it, so an empty value must stay the default.
-        try:
-            internal = bytes.fromhex(internal_key_hex)
-        except (TypeError, ValueError):
-            return None
-        if len(internal) == len(raw_passphrase):
-            raw_passphrase = bytes(a ^ b for a, b in zip(raw_passphrase, internal))
-
-    try:
-        with open(db_path, 'rb') as handle:
-            page = handle.read(4096)
-    except OSError:
-        return None
-    if len(page) < 4096 or page[:16] == b'\x00' * 16:
-        return None
-
-    salt = page[:16]
-    mac_salt = bytes(byte ^ 0x3a for byte in salt)
-    key = hashlib.pbkdf2_hmac('sha512', raw_passphrase, salt, 256000, 32)
-    mac_key = hashlib.pbkdf2_hmac('sha512', key, mac_salt, 2, 32)
-
-    reserve = 16 + 64                      # IV + HMAC-SHA512
-    reserve = ((reserve + 15) // 16) * 16  # rounded up to the AES block size
-    body_end = 4096 - reserve + 16
-    mac = hmac.new(mac_key, page[16:body_end], hashlib.sha512)
-    mac.update(struct.pack('<I', 1))       # page number
-    return hmac.compare_digest(mac.digest(), page[body_end:body_end + 64])
-
-
-def connect_message_shards_detailed(db_path, key_hex, salt_hex, passphrase=''):
-    """Open every shard and report what happened to each one.
-
-    Returns (pairs, failures):
-      pairs    [(shard_path, conn)] for the shards that opened
-      failures [{"name": basename, "reason": "KEY_REJECTED" | "OPEN_FAILED"}]
-
-    A shard that fails is still skipped rather than fatal - a partially
-    readable transcript beats a command that refuses to run at all - but the
-    failure stops being invisible. `shardsFailed` is the difference between
-    "read everything" and "read what happened to be reachable", which is the
-    whole point of the coverage report.
-
-    Only the basename is reported: the caller stores this in a state file, and
-    absolute paths must not end up there.
-    """
-    pairs = []
-    failures = []
-    for shard in discover_message_shards(db_path):
-        derived_key, derived_salt = derive_database_key(
-            shard, key_hex, salt_hex, passphrase)
-        # Try the derived key first; fall back to the configured pair so
-        # installs without a passphrase keep working exactly as before.
-        candidates = [(derived_key, derived_salt)]
-        if (derived_key, derived_salt) != (key_hex, salt_hex):
-            candidates.append((key_hex, salt_hex))
-        failure_reason = 'OPEN_FAILED'
-        for candidate_key, candidate_salt in candidates:
-            conn = None
-            try:
-                conn, _ = connect_nt_db(shard, candidate_key, candidate_salt)
-                # PRAGMA key alone never fails; only a read surfaces a bad key.
-                conn.execute('SELECT count(*) FROM sqlite_master').fetchone()
-            except Exception:
-                # Reached the database but could not decrypt it: the key is
-                # wrong, not the file. Close before retrying, or the handle
-                # leaks and on Windows the file stays locked.
-                if conn is not None:
-                    try:
-                        conn.close()
-                    except Exception:
-                        pass
-                failure_reason = 'KEY_REJECTED'
-                continue
-            pairs.append((shard, conn))
-            break
-        else:
-            failures.append({'name': os.path.basename(shard), 'reason': failure_reason})
-    return pairs, failures
-
-
-def connect_message_shards(db_path, key_hex, salt_hex, passphrase=''):
-    """Open the configured database plus every sibling shard.
-
-    Returns the connections that actually opened. A shard that fails is
-    skipped rather than fatal: a partially readable transcript beats a
-    command that refuses to run at all.
-
-    Behaviour-preserving delegate: callers that do not ask for the failure
-    detail get exactly what they got before.
-    """
-    pairs, _failures = connect_message_shards_detailed(
-        db_path, key_hex, salt_hex, passphrase)
-    return [conn for _path, conn in pairs]
-
-
-def count_talker_rows(conn, msg_table):
-    """Rows this conversation has in one shard, or None if it has no table there.
-
-    Distinguishes "this shard holds nothing for the conversation" from "this
-    shard could not be read" - without it both look like zero.
-    """
-    try:
-        cur = conn.execute(
-            "SELECT COUNT(*) FROM sqlite_master WHERE name=?", (msg_table,))
-        if cur.fetchone()[0] == 0:
-            return None
-        return conn.execute('SELECT COUNT(*) FROM "%s"' % msg_table).fetchone()[0]
-    except Exception:
-        return None
-
-
-def msg_tables(conn):
-    """Names of the per-conversation Msg_ tables in one shard."""
-    try:
-        rows = conn.execute(
-            "SELECT name FROM sqlite_master WHERE type='table' AND name LIKE 'Msg\\_%' ESCAPE '\\'"
-        ).fetchall()
-    except Exception:
-        return set()
-    return {row[0] for row in rows}
-
-
-def get_fav_schema(db_path, key_hex):
-    """Dump favorite.db schema + sample rows (key verification / exploration)."""
-    try:
-        with open(db_path, 'rb') as fh:
-            salt = fh.read(16).hex()
-        conn = require_sqlcipher().connect(db_path)
-        conn.execute(f'PRAGMA key = "x\'{key_hex}{salt}\'";')
-        tables = conn.execute(
-            "SELECT name FROM sqlite_master WHERE type='table' ORDER BY name"
-        ).fetchall()
-        result = {"success": True, "tables": [t[0] for t in tables]}
-        if not tables:
-            result["error"] = "数据库为空或密钥错误"
-            result["success"] = False
-        conn.close()
-        return result
-    except Exception as e:
-        return {"success": False, "error": str(e).split('\n')[0][:200]}
-
-
-FAV_TYPE_NAMES = {
-    1: 'text',        # 文字
-    2: 'image',       # 图片
-    4: 'video',       # 视频
-    5: 'article',     # 公众号文章/网页链接
-    14: 'chatrecord', # 聊天记录
-}
-
-
-def parse_fav_content(content):
-    """Extract display fields (title/link/desc/source/cover) from favitem XML."""
-    out = {}
-    if not content:
-        return out
-    import xml.etree.ElementTree as ET
-    try:
-        root = ET.fromstring(content)
-    except ET.ParseError:
-        return out
-
-    def text(path):
-        el = root.find(path)
-        if el is not None and el.text and el.text.strip():
-            return el.text.strip()
-        return None
-
-    title = (text('weburlitem/pagetitle')
-             or text('datalist/dataitem/datatitle')
-             or text('title')
-             or text('desc'))
-    if title:
-        out['title'] = title
-    link = (text('weburlitem/clean_url')
-            or text('source/link')
-            or text('datalist/dataitem/stream_weburl'))
-    if link:
-        out['link'] = link
-    desc = text('weburlitem/pagedesc') or text('datalist/dataitem/datadesc')
-    if desc and desc != out.get('title'):
-        out['desc'] = desc
-    src = text('weburlitem/appmsgshareitem/srcdisplayname')
-    if src:
-        out['source_name'] = src
-    cover = (text('weburlitem/pagethumb_url')
-             or text('datalist/dataitem/dataext')
-             or text('datalist/dataitem/cdn_thumburl'))
-    if cover:
-        out['cover'] = cover
-    fmt = text('datalist/dataitem/datafmt')
-    if fmt:
-        out['format'] = fmt
-    return out
-
-
-def get_favorites(db_path, key_hex, limit=100, offset=0, keyword=None, fav_type=None):
-    """List favorite items from favorite.db with parsed content."""
-    try:
-        with open(db_path, 'rb') as fh:
-            salt = fh.read(16).hex()
-        conn = require_sqlcipher().connect(db_path)
-        conn.execute(f'PRAGMA key = "x\'{key_hex}{salt}\'";')
-        c = conn.cursor()
-
-        tables = [t[0] for t in c.execute(
-            "SELECT name FROM sqlite_master WHERE type='table'").fetchall()]
-        if 'fav_db_item' not in tables:
-            conn.close()
-            return {"error": f"未找到 fav_db_item 表，现有表: {tables[:20]}"}
-
-        total = c.execute('SELECT count(*) FROM fav_db_item').fetchone()[0]
-
-        sql = ('SELECT local_id, server_id, type, update_time, fromusr, realchatname, content '
-               'FROM fav_db_item WHERE 1=1')
-        params = []
-        if fav_type is not None:
-            sql += ' AND type = ?'
-            params.append(fav_type)
-        if keyword:
-            sql += ' AND content LIKE ?'
-            params.append(f'%{keyword}%')
-        sql += ' ORDER BY update_time DESC LIMIT ? OFFSET ?'
-        params.extend([limit, offset])
-        rows = c.execute(sql, params).fetchall()
-        conn.close()
-
-        items = []
-        for local_id, server_id, ftype, update_time, fromusr, realchatname, content in rows:
-            ct = content if isinstance(content, str) else (
-                content.decode('utf-8', errors='replace') if content else '')
-            item = {
-                'local_id': local_id,
-                'server_id': server_id,
-                'type': ftype,
-                'type_name': FAV_TYPE_NAMES.get(ftype, 'type_%s' % ftype),
-                'update_time': update_time,
-                'from_user': fromusr,
-                'chat_name': realchatname or None,
-            }
-            item.update(parse_fav_content(ct))
-            items.append(item)
-        return {"favorites": items, "total": total, "count": len(items),
-                "limit": limit, "offset": offset}
-    except Exception as e:
-        return {"error": str(e).split('\n')[0][:200]}
-
-
-#: 非文本消息在**列表预览**里的显示形态。与 `sqlcipherCore.processMessage` 的口径一致。
-#: 列表预览只需要回答"这是什么"，不需要内容 —— 而 49 类的 `source` 往往是被压缩的
-#: appmsg，硬去剥 XML 标签只会得到 `(/`$` 这种碎片（2026-10-19 用户在 `sessions` 里看到）。
-_PREVIEW_LABELS = {
-    3: '[图片]', 34: '[语音]', 43: '[视频]', 47: '[表情]', 48: '[位置]',
-    42: '[名片]', 49: '[链接/文件]', 10000: '[系统消息]', 10002: '[系统消息]',
-}
-
-
-def _clean_preview(text, limit=50):
-    """预览用：清掉控制字符、压掉换行/连续空白、去掉行首的二进制残渣、截断。
-
-    行首清洗只针对 **ASCII 标点**：群消息的 `source` 有时前面带三四个字节的碎片
-    （实测一条广告消息长这样：`(/ }25984981651785937@openim:🔰周黑鸭…`），
-    不处理的话列表第一列就是这堆符号。中文标点与 emoji 开头**不动**——
-    `？好的`、`😂收到` 都是正常消息。
-    """
-    if not isinstance(text, str):
-        return ''
-    cleaned = ''.join(ch for ch in text if ch >= ' ' or ch == '\t')
-    cleaned = re.sub(r'\s+', ' ', cleaned).strip()
-    # 空格也要算在"垃圾前缀"里：实测前缀是 `(/ }`，中间那个空格会把
-    # 「连续 ASCII 标点」这条匹配打断，于是整段清理失效（2026-10-19 实测）。
-    cleaned = re.sub(r'^[\x20-\x2f\x3a-\x40\x5b-\x60\x7b-\x7e]{1,8}(?=[A-Za-z0-9\u4e00-\u9fff])', '', cleaned)
-    return cleaned[:limit]
-
-
-def _looks_broken(text):
-    """这段"文本"其实是被当成文本的二进制吗？
-
-    判据只看两样：UTF-8 解码时留下的替换字符、以及不可打印字符的占比。
-    实测样本：一条 openim 消息的 `message_content` 解出来是
-    `(/ }25984981651785937@openim:` 加一个替换字符 —— 直接显示就是垃圾。
-    """
-    if not text:
-        return False
-    # 出现**任何**替换字符就说明原始字节不是合法 UTF-8 —— 正常中文/emoji 不会产生它。
-    # 阈值不能是"两个以上"：实测那条 openim 消息只在结尾丢了一个字节，
-    # 前面全是可打印的压缩碎片，比例判据根本拦不住。
-    if '\ufffd' in text:
-        return True
-    printable = sum(1 for ch in text if ch.isprintable())
-    return printable / max(1, len(text)) < 0.9
-
-
-def _summarise(row):
-    """(last_time, summary) from the newest message row of a conversation."""
-    last_time = row[0] or 0
-    msg_type = row[3] or 0
-    source_text = row[1]
-    content_text = row[2]
-
-    if msg_type == 1:
-        # 文本：`message_content` 与 `source` 里**哪个更像人话就用哪个**。
-        # 只认 `message_content` 会踩到"这一列其实是二进制"的消息（见 `_looks_broken`）。
-        candidates = []
-        for value in (content_text, source_text):
-            if isinstance(value, bytes):
-                value = value.decode('utf-8', errors='ignore')
-            if isinstance(value, str) and value.strip():
-                candidates.append(value)
-        if not candidates:
-            return last_time, ""
-        best = min(candidates, key=lambda s: s.count('\ufffd') / max(1, len(s)))
-        if _looks_broken(best):
-            return last_time, '[内容无法显示]'
-        return last_time, _clean_preview(best)
-
-    # 非文本：先按类型给标签。**不再从 `source` 里剥标签抠文字** —— 那条路只有在
-    # XML 是明文 appmsg 时才成立，而 4.x 上它经常是压缩的，抠出来就是二进制碎片。
-    #
-    # `local_type` 在 4.x 里是**打包值**：高位放 subtype，低 32 位才是基础类型
-    # （实测 `25769803825` = 0x6_00000031，低 8 位 0x31 = 49 = 链接/文件）。
-    # 不 mask 的话，一张「链接卡片」会被报成 `[消息类型 25769803825]`。
-    raw_type = int(msg_type) if msg_type else 0
-    base_type = raw_type & 0xFFFFFFFF
-    label = _PREVIEW_LABELS.get(base_type)
-    if label:
-        return last_time, label
-
-    if isinstance(source_text, str) and source_text.strip():
-        clean = re.sub(r'<[^>]+>', '', source_text)
-        if clean.strip():
-            # 兜底也要过一遍"这其实是二进制吗"：某些消息两列都是压缩数据，
-            # 剥掉尖括号后看着像文字，实际是 `(/ }25984...` 这种碎片。
-            if _looks_broken(clean):
-                return last_time, '[内容无法显示]'
-            return last_time, _clean_preview(clean)
-    return last_time, ('[消息类型 %d]' % base_type) if base_type else ''
-
-
-def get_sessions(conns):
-    """Get chat sessions, newest message per conversation across every shard."""
-    sessions = {}
-
-    # NT format: each chat has its own Msg_<MD5> table
-    # The Name2Id table maps usernames to IDs (user_name, is_session)
-    for conn in conns:
-        c = conn.cursor()
-        try:
-            c.execute("SELECT user_name FROM Name2Id WHERE is_session = 1 LIMIT 500")
-            usernames = [row[0] for row in c.fetchall() if row[0]]
-        except Exception:
-            continue
-
-        tables = msg_tables(conn)
-        for username in usernames:
-            entry = sessions.setdefault(username, {"summary": "", "last_time": 0})
-            msg_table = f"Msg_{hashlib.md5(username.encode()).hexdigest()}"
-            if msg_table not in tables:
-                continue
-            try:
-                c.execute(f'SELECT create_time, source, message_content, local_type FROM "{msg_table}" ORDER BY create_time DESC LIMIT 1')
-                row = c.fetchone()
-                if row:
-                    last_time, summary = _summarise(row)
-                    # A conversation spans shards; only the newest speaks for it.
-                    if last_time >= entry["last_time"]:
-                        entry["last_time"] = last_time
-                        entry["summary"] = summary
-            except Exception:
-                continue
-
-    result = [{
-        "username": username,
-        "type": 1 if "@chatroom" in username else 0,
-        "unreadCount": 0,
-        "summary": entry["summary"],
-        "sortTimestamp": entry["last_time"],
-        "lastTimestamp": entry["last_time"],
-        "displayName": username,
-    } for username, entry in sessions.items()]
-
-    # Sort by timestamp descending
-    result.sort(key=lambda s: s.get("sortTimestamp", 0), reverse=True)
-    return {"sessions": result}
-
-
-def _strip_group_speaker(content, known_ids):
-    """Drop the `wxid_...:` prefix group rows carry in their content.
-
-    Only an id the shard's Name2Id actually knows is accepted, so a message
-    that merely starts with `note: ...` is left alone.
-    """
-    match = re.match(r'^([A-Za-z0-9_@.-]{5,64})\s*[:：]\s', str(content or ''))
-    if not match or match.group(1) not in known_ids:
-        return content
-    return content[match.end():]
-
-
-# The columns the reader actually consumes. The SELECT is built from these
-# only: a WeChat version that renames or drops any *other* column of the
-# message table cannot break a read of a column nobody looks at.
-MESSAGE_COLUMNS = ('local_id', 'server_id', 'local_type', 'real_sender_id',
-                   'create_time', 'message_content')
-
-# Any one of these is enough to give rows an identity and an order. With none
-# of them a row cannot become a message at all, and that is a schema mismatch
-# rather than an empty conversation.
-# 定义搬去了 `nt_common`：`ORDER BY` 那处（同一个文件下面）过去自己又写了一份
-# 字面量，导出器再写一份——顺序在这里就是行为（先按 create_time 排），
-# 三份各自为政等于谁改一处谁静默换一种排法。
-
-
-# ---- 非文本消息的显示形态 ---------------------------------------------------
-#
-# 这条路径此前**把非文本消息丢掉了**：`parsedContent` 只在 local_type == 1 时才有值，
-# 而 BLOB 内容不是 str，会在下面被置成空串。于是图片/表情/文件/引用/撤回/红包在下游
-# 一律表现为"空内容"——实测那 44 条里"空"的三条，分别是**一条撤回提示**和两个表情。
-#
-# 类型编码有结构：`local_type = apptype * 2**32 + 49`，49 表示 appmsg 家族。所以这里
-# 是**推导**而不是一张长表；只有少数几个固定类型需要单独列。
-APPMSG_SUBTYPE = 49
-NON_TEXT_LABELS = {
-    3: '图片', 34: '语音', 42: '名片', 43: '视频', 47: '表情', 48: '位置',
-}
-APPMSG_LABELS = {
-    4: '链接', 5: '链接', 6: '文件', 19: '聊天记录', 33: '小程序', 57: '引用', 63: '直播',
-    2000: '转账', 2001: '红包',
-}
-# 不认识的 apptype 回 `[应用消息]` 而不是猜成"链接"：标签是给下游读的，不能支撑不起也写。
-APPMSG_UNKNOWN_LABEL = '应用消息'
-
-# 引用消息（apptype 57）的正文与被引原文之间的分隔。用词而不是符号，因为下游会把它读成
-# 一句话；`|` 在别处会被当表格分隔。
-QUOTE_SEP = ' ｜ 引：'
-# 被引原文的截断长度。实测 37 条真实引用消息：中位 36 字、17 条超过 60 字、最长的
-# 12733 字（引用了一整篇公众号文章）。60 会切掉近一半，120 能留住大多数完整句子，
-# 又把那个极端值压住。
-QUOTE_CLIP = 120
-
-
-ZSTD_MAGIC = bytes([0x28, 0xB5, 0x2F, 0xFD])
-def _decode_content(raw):
-    """消息内容 → 文本。BLOB 先按 zstd 解压（公众号消息同一套机制）。解不出回空串。"""
-    if isinstance(raw, str):
-        return raw
-    if not raw:
-        return ''
-    data = bytes(raw)
-    if data[:4] == ZSTD_MAGIC:        # zstd frame header
-        try:
-            import zstandard
-            return zstandard.ZstdDecompressor().decompress(data).decode('utf-8', 'ignore')
-        except Exception:
-            return ''
-    return data.decode('utf-8', 'ignore')
-
-
-def _xml_block(xml, tag):
-    """取一个元素的起始标签到它自己的结束标签之间的整段（含标签）。
-
-    只在区块内找 `title`/`des`：整份 payload 里还有 `<emotionpageshared><title>` 这类
-    同名标签，在全文里搜会匹配到那个去（自验时真的把一段 XML 当文本吐出来过）。
-    """
-    text = xml or ''
-    match = re.search(r'<%s(?:\s[^>]*)?>.*?</%s>' % (tag, tag), text, re.S)
-    return match.group(0) if match else ''
-
-
-def _xml_text(xml, tag):
-    """取一个标签的文本（含 CDATA）。取不到回空串——`<title />` 这种自闭合就是取不到。
-
-    起始标签里**不许出现 `/`**：否则 `<title />` 会被当成开标签，一路吃到后面某个
-    `</title>`，把中间整段 XML 当成文本返回（自验时真的吐出来过）。
-
-    实体在这**解掉**：显示形态是给人（和模型）读的，`a&amp;b` 该显示成 `a&b`。
-    TS 侧的 `appMsgFormat.tagText` 同样解，两边一致——否则同一条消息 3.x 与 4.x 读起来
-    不一样。
-    """
-    match = re.search(
-        r'<%s(?:\s[^>/]*)?>(?:<!\[CDATA\[)?(.*?)(?:\]\]>)?</%s>' % (tag, tag), xml or '', re.S)
-    return html.unescape(match.group(1).strip()) if match else ''
-
-
-def _clip(text, limit):
-    """超长截断并**留下省略号**。
-
-    不标记的截断读起来像一句说完的话——`title[:60]` 此前就是这样，被引原文最长的
-    12733 字，切完看着像"这就是全部"。
-    """
-    if len(text) <= limit:
-        return text
-    return text[:limit] + '…'
-
-
-def non_text_display(local_type, raw):
-    """非文本消息 → 给下游看的文本。**绝不返回空串**——这是这次修复的全部要点。
-
-    宁可写 `[未识别的消息类型 81604378673]`，也不要让下游看到空字符串：空串在下游与
-    "这条消息不存在"无法区分，而它的真实后果是助手回答"我没解析出内容"。
-    """
-    lt = int(local_type or 0)
-    text = _decode_content(raw)
-
-    if lt & 0xFFFFFFFF == APPMSG_SUBTYPE:
-        apptype = lt >> 32
-        label = APPMSG_LABELS.get(apptype, APPMSG_UNKNOWN_LABEL)
-        block = _xml_block(text, 'appmsg')
-        title = _xml_text(block, 'title') or _xml_text(block, 'des')
-        # 实测 payload 里 `<emotionpageshared>` 这类**嵌套容器**也带 `<title>`，取到的是
-        # 它们自己的占位值（`null`）。这种不是标题，按"没有标题"处理。
-        if title.lower() in ('null', 'undefined', '0'):
-            title = ''
-        # 截断：解析万一走偏，也不许把一段 XML 当成标题交给下游。
-        title = _clip(title, 60)
-        # 引用消息（apptype 57）：`title` 是**回复正文**，被引用的原文在 `refermsg/content`。
-        # 实测 37 条真实引用消息，被引原文中位 36 字。此前只留回复，模型看到的是一句
-        # "是呀，够得意个"却不知道在回什么——引用不带原文等于没引用。
-        if apptype == 57:
-            quoted = _xml_text(_xml_block(text, 'refermsg'), 'content')
-            if quoted.lower() in ('null', 'undefined', '0'):
-                quoted = ''
-            quoted = _clip(quoted, QUOTE_CLIP)
-            if quoted:
-                body = (title + QUOTE_SEP + quoted) if title else quoted
-                return '[%s] %s' % (label, body)
-        return '[%s] %s' % (label, title) if title else '[%s]' % label
-
-    label = NON_TEXT_LABELS.get(lt)
-    if label:
-        return '[%s]' % label
-    if lt == 10000:                       # 系统消息：撤回、拍一拍、入群提示
-        return _xml_text(text, 'content') or '[系统消息]'
-    return '[未识别的消息类型 %d]' % lt
-
-
-def _message_dict(row, sender_id_map, name_map, own_wxid, is_group=False):
-    """One message row -> the CLI's message shape.
-
-    `row` is a column-name keyed dict, not a tuple. Positional access assumes
-    every column exists; when a WeChat version drops one, the whole shard read
-    fails and the conversation looks empty. Reading by name means a missing
-    column costs that one field instead.
-    """
-    local_type = row.get('local_type') or 0
-    create_time = row.get('create_time') or 0
-    real_sender_id = row.get('real_sender_id') or 0
-
-    # Resolve sender: real_sender_id -> Name2Id -> user_name
-    sender_username = sender_id_map.get(real_sender_id, "")
-
-    # Determine if message is from self
-    # own_wxid may have _xxxx suffix (from xwechat_files dir), try both
-    is_self = bool(own_wxid and (
-        sender_username == own_wxid or
-        (own_wxid.endswith('_') is False and sender_username.startswith(own_wxid))
-    ))
-    if not is_self and own_wxid:
-        # Strip _xxxx suffix and retry
-        parts = own_wxid.rsplit('_', 1)
-        if len(parts) == 2 and len(parts[1]) == 4 and parts[1].isalnum():
-            is_self = (sender_username == parts[0])
-
-    # Resolve sender display name from contact map
-    if is_self:
-        sender_display = ""  # Let the CLI show "我"
-    else:
-        sender_display = name_map.get(sender_username, sender_username) if sender_username else sender_username
-
-    # Parse message_content - TEXT column
-    raw_content = row.get('message_content')
-    content = raw_content if isinstance(raw_content, str) else ""
-
-    # `content`/`rawContent` stay exactly as stored; only `parsedContent` - the
-    # field every consumer reads first - gets the display-ready form.
-    #
-    # 非文本消息（图片/表情/文件/引用/撤回/红包…）在过去得到的是**空串**：它们的内容是
-    # JSON 里不是 str 的 BLOB，上面那行会把它置空。现在它们走 `non_text_display`，
-    # 至少拿到 `[图片]`，多数还能带上文件名、被引用的原文或"谁撤回了一条消息"。
-    if content:
-        display = _strip_group_speaker(content, set(sender_id_map.values())) if is_group else content
-    else:
-        display = non_text_display(local_type, raw_content)
-
-    return {
-        "localId": row.get('local_id') or 0,
-        "serverId": str(row.get('server_id') or ''),
-        "localType": local_type,
-        "createTime": create_time,
-        "isSend": 1 if is_self else 0,  # 1 = I sent this
-        "senderUsername": sender_username,
-        "senderDisplay": sender_display,
-        "content": content,
-        "rawContent": content,
-        "parsedContent": display[:200] if local_type == 1 else display[:200],
-    }
-
-
-def get_messages(conns, talker, limit=100, offset=0, name_map=None, own_wxid=None,
-                 shard_names=None, shard_report=None, from_time=None, to_time=None):
-    """Get messages for a specific talker, merged across every shard.
-
-    Args:
-        name_map: optional {wxid: display_name} dict for resolving sender names
-        own_wxid: account owner wxid for self-message detection
-        shard_names: optional basename per connection, parallel to `conns`
-        shard_report: optional list to append one outcome dict per shard to.
-            Purely additive: the returned messages are identical with or
-            without it, which is asserted by test/nt_decrypt_shards_test.py.
-        from_time: optional inclusive lower bound on `create_time` (unix
-            seconds), pushed into SQL so a bounded read does not have to fetch
-            and discard the whole conversation.
-        to_time: optional inclusive upper bound, same units.
-
-    A requested window is only ever honoured or refused, never approximated: a
-    shard whose table has no `create_time` is skipped and reported as
-    WINDOW_UNAVAILABLE rather than returning out-of-range rows that a caller
-    tracking coverage would count as covered. Every in-tree caller that passes
-    a window also passes `shard_report`, so that refusal is always visible.
-    """
-    if name_map is None:
-        name_map = {}
-
-    window_requested = from_time is not None or to_time is not None
-
-    msg_table = f"Msg_{hashlib.md5(talker.encode()).hexdigest()}"
-    is_group = '@chatroom' in talker
-
-    def record(index, opened, has_table, rows_for_talker, reason, missing=None):
-        if shard_report is None:
-            return
-        name = shard_names[index] if shard_names and index < len(shard_names) else ''
-        shard_report.append({
-            'name': name,
-            'opened': opened,
-            'hasTalkerTable': has_table,
-            'rowsForTalker': rows_for_talker,
-            'reason': reason,
-            # Columns this shard does not have. A non-empty list together with
-            # a null `reason` means the read succeeded but lost fields, which is
-            # a different thing from a failed read and has to stay visible.
-            'missingColumns': list(missing or ()),
-        })
-
-    # Each shard only needs to yield its newest window: once every shard's rows
-    # are merged and re-sorted, nothing older than that can reach this page.
-    window = 0 if limit <= 0 else limit + offset
-    collected = []
-    found = False
-
-    for index, conn in enumerate(conns):
-        c = conn.cursor()
-        try:
-            c.execute("SELECT COUNT(*) FROM sqlite_master WHERE name=?", (msg_table,))
-            if c.fetchone()[0] == 0:
-                # The shard is readable and simply holds nothing for this
-                # conversation - distinct from not being readable at all.
-                record(index, True, False, None, None)
-                continue
-            found = True
-            rows_for_talker = c.execute(
-                'SELECT COUNT(*) FROM "%s"' % msg_table).fetchone()[0]
-
-            available = table_columns(c, msg_table)
-            selected = [col for col in MESSAGE_COLUMNS if col in available]
-            missing = [col for col in MESSAGE_COLUMNS if col not in available]
-            if not [col for col in MESSAGE_ANCHOR_COLUMNS if col in available]:
-                # The row shape no longer matches anything the reader can name.
-                record(index, True, True, rows_for_talker, 'SCHEMA_MISMATCH',
-                       missing=missing)
-                continue
-            if window_requested and 'create_time' not in available:
-                record(index, True, True, rows_for_talker, 'WINDOW_UNAVAILABLE',
-                       missing=missing)
-                continue
-
-            order = [col for col in MESSAGE_ANCHOR_COLUMNS
-                     if col in available]
-            sql = 'SELECT %s FROM "%s"' % (
-                ', '.join('"%s"' % col for col in selected), msg_table)
-            params = []
-            clauses = []
-            if from_time is not None:
-                clauses.append('"create_time" >= ?')
-                params.append(from_time)
-            if to_time is not None:
-                clauses.append('"create_time" <= ?')
-                params.append(to_time)
-            if clauses:
-                sql += ' WHERE ' + ' AND '.join(clauses)
-            sql += ' ORDER BY ' + ', '.join('"%s" DESC' % col for col in order)
-            if window:
-                sql += ' LIMIT ?'
-                params.append(window)
-            c.execute(sql, params)
-            # Dicts rather than driver rows: the SELECT list varies per shard,
-            # so position no longer identifies a column. Built here rather than
-            # via row_factory, which every other reader sharing these
-            # connections would inherit.
-            rows = [dict(zip(selected, values)) for values in c.fetchall()]
-
-            # Sender ids are rowids, so the map has to come from the same
-            # shard. Absent, it costs sender names and nothing else, so it is
-            # probed rather than left to raise and lose the whole shard.
-            has_name_table = c.execute(
-                "SELECT COUNT(*) FROM sqlite_master WHERE name='Name2Id'").fetchone()[0] > 0
-            sender_id_map = {}
-            if has_name_table:
-                c.execute("SELECT rowid, user_name FROM Name2Id")
-                sender_id_map = {rowid: uname for rowid, uname in c.fetchall()}
-        except Exception:
-            # This shard had the conversation's table but could not be read.
-            # Previously this was indistinguishable from "no rows here".
-            record(index, True, None, None, 'READ_FAILED')
-            continue
-        record(index, True, True, rows_for_talker, None, missing=missing)
-
-        for row in rows:
-            collected.append(_message_dict(row, sender_id_map, name_map, own_wxid, is_group))
-
-    if not found:
-        return {"error": f"未找到会话: {talker}"}
-
-    collected.sort(key=lambda m: (m["createTime"], m["localId"]), reverse=True)
-    if limit > 0:
-        collected = collected[offset:offset + limit]
-    return {"messages": collected}
-
-
-def cursor_probe(db_path, key_hex, salt_hex, passphrase='',
-                 tables_per_shard=5, rows_per_table=500):
-    """Measure what a stable incremental cursor would need on this database.
-
-    D-027 keeps `stableCursor: false` until every message backend can honour
-    the same semantics. This probe answers the NT backend's **data-side** half:
-    whether every message table carries the anchor columns, and whether sampled
-    rows actually hold a nonzero `server_id` (the reader's identity falls back
-    to localId+createTime without one, and `local_id` is per-shard, so those
-    rows cannot be resumed across shards). Duplicate server ids are collected
-    across the sample too - an id that appears in two places cannot key a
-    cursor, whatever the column definition says.
-
-    Sampling is bounded and says so (`exhaustive: false`): `tables_per_shard`
-    tables per shard (sorted names, deterministic) read at both rowid ends,
-    `rows_per_table` rows each. A full scan of every shard would cost minutes
-    on a real account; the schema check IS exhaustive - `PRAGMA table_info` is
-    cheap, so every Msg_* table's anchor columns are checked, not sampled.
-
-    Read-only: no write, no network, nothing leaves stdout.
-    """
-    pairs, failures = connect_message_shards_detailed(
-        db_path, key_hex, salt_hex, passphrase)
-    shard_items = []
-    totals = {'tables': 0, 'tablesSampled': 0, 'rowsSampled': 0,
-              'serverIdMissing': 0}
-    anchor_missing_total = 0
-    # server_id(str) -> {(shard, table)} it was sampled in. A set per id keeps
-    # the two rowid ends of one small table from double-counting the same row
-    # as a "duplicate" - both ends see it, but it is one location.
-    id_locations = {}
-
-    for failure in failures:
-        shard_items.append({
-            'name': failure['name'], 'opened': False, 'reason': failure['reason'],
-            'tables': None,
-        })
-
-    for path, conn in pairs:
-        item = {'name': os.path.basename(path), 'opened': True, 'reason': None}
-        try:
-            tables = sorted(msg_tables(conn))
-            item['tables'] = len(tables)
-            totals['tables'] += len(tables)
-
-            # Schema half: exhaustive over every message table.
-            anchor_missing = []
-            for table in tables:
-                cols = table_columns(conn.cursor(), table)
-                missing = [c for c in MESSAGE_ANCHOR_COLUMNS if c not in cols]
-                if missing and missing not in anchor_missing:
-                    anchor_missing.append(missing)
-            anchor_missing_total += len(anchor_missing)
-            item['anchorColumnsMissing'] = anchor_missing
-
-            # Data half: bounded sample, both rowid ends per table. A table
-            # shorter than the bound hands the same rows to both ends - they
-            # are one sample, not two, so rowids are deduplicated per table
-            # (and the same id read at both ends must not look "duplicated").
-            sample_targets = tables[:tables_per_shard]
-            item['tablesSampled'] = len(sample_targets)
-            totals['tablesSampled'] += len(sample_targets)
-            sampled = 0
-            missing_ids = 0
-            for table in sample_targets:
-                c = conn.cursor()
-                seen_rowids = set()
-                table_ids = set()
-                for order in ('ASC', 'DESC'):
-                    c.execute(
-                        'SELECT rowid, server_id FROM "%s" ORDER BY rowid %s LIMIT ?'
-                        % (table, order), (rows_per_table,))
-                    for rowid, server_id in c.fetchall():
-                        if rowid in seen_rowids:
-                            continue
-                        seen_rowids.add(rowid)
-                        sampled += 1
-                        if server_id is None or server_id == 0:
-                            missing_ids += 1
-                        else:
-                            table_ids.add(str(int(server_id)))
-                for sid in table_ids:
-                    id_locations.setdefault(sid, set()).add((item['name'], table))
-            item['rowsSampled'] = sampled
-            item['serverIdMissing'] = missing_ids
-            totals['rowsSampled'] += sampled
-            totals['serverIdMissing'] += missing_ids
-        except Exception as exc:
-            item['reason'] = 'READ_FAILED: %s' % str(exc).split('\n')[0][:120]
-        finally:
-            shard_items.append(item)
-            conn.close()
-
-    duplicates = {sid: locs for sid, locs in id_locations.items() if len(locs) > 1}
-    return {
-        'shards': shard_items,
-        'schemaAnchorMissingTables': anchor_missing_total,
-        'sample': {
-            'tablesPerShard': tables_per_shard,
-            'rowsPerTable': rows_per_table,
-            'exhaustive': False,
-            'tablesTotal': totals['tables'],
-            'tablesSampled': totals['tablesSampled'],
-            'rowsSampled': totals['rowsSampled'],
-            'serverIdMissing': totals['serverIdMissing'],
-            'duplicateServerIds': len(duplicates),
-            'duplicateServerIdSamples': [
-                {'serverId': sid, 'locations': sorted('%s/%s' % loc for loc in locs)}
-                for sid, locs in sorted(duplicates.items())[:20]
-            ],
-        },
-    }
-
-
-def get_contacts(conns, limit=200, keyword=None):
-    """Get contacts from NT database, merged across every shard.
-
-    **给了 `keyword` 就不加 `LIMIT`**，由调用方在名字解析之后过滤、再按 `limit` 截断。
-    原来无条件 `LIMIT`，于是过滤发生在**截断之后**：关键字只能在前 `limit` 行里找，
-    后面的人搜不到（实测 500 人的通讯录、按备注名找 10 个只命中 3 个）。名字解析要用
-    contact.db 的备注/昵称，那是调用方才有的东西，所以截断必须挪到过滤之后。
-    """
-    seen = set()
-    contacts = []
-    for conn in conns:
-        try:
-            sql = "SELECT user_name FROM Name2Id" if keyword else "SELECT user_name FROM Name2Id LIMIT ?"
-            rows = conn.execute(sql, () if keyword else (limit,)).fetchall()
-        except Exception:
-            continue
-        for (username,) in rows:
-            # Name2Id carries a placeholder row with no user_name; rendering it
-            # produced a blank line at the top of every contact list.
-            if not username or username in seen:
-                continue
-            seen.add(username)
-            contacts.append({"username": username, "displayName": username})
-    return {"contacts": contacts}
-
-
-# ========== SNS (朋友圈) Queries ==========
-
-def parse_sns_content(content_str):
-    """Parse SNS content XML/Protobuf text to extract title, description, media etc."""
-    result = {
-        'content': '',
-        'create_time': 0,
-        'username': '',
-        'object_id': '',
-        'media_count': 0,
-    }
-    if not content_str:
-        return result
-
-    import re
-
-    # Extract createTime
-    m = re.search(r'<createTime>(\d+)</createTime>', content_str)
-    if m:
-        result['create_time'] = int(m.group(1))
-
-    # Extract username
-    m = re.search(r'<username>([^<]+)</username>', content_str)
-    if m:
-        result['username'] = m.group(1)
-
-    # Extract id
-    m = re.search(r'<id>(\d+)</id>', content_str)
-    if m:
-        result['object_id'] = m.group(1)
-
-    # Extract contentDesc (main text)
-    m = re.search(r'<contentDesc>([^<]*)</contentDesc>', content_str)
-    if m:
-        result['content'] = m.group(1)
-
-    # Extract contentDesc CDATA
-    m = re.search(r'<contentDesc>\s*<!\[CDATA\[(.*?)\]\]>\s*</contentDesc>', content_str, re.DOTALL)
-    if m:
-        result['content'] = m.group(1).strip()
-
-    # Extract title if present
-    m = re.search(r'<title>([^<]*)</title>', content_str)
-    if m:
-        title = m.group(1)
-        if title and not result['content']:
-            result['content'] = title
-        elif title:
-            result['content'] = title + '\n' + result['content']
-
-    # Count media (ContentObject tags)
-    result['media_count'] = len(re.findall(r'<ContentObject[ >]', content_str))
-
-    return result
-
-
-def get_sns_timeline(cursor, limit=20, offset=0, usernames=None, keyword=None,
-                     start_time=None, end_time=None):
-    """Query SNS timeline posts from SnsTimeLine table."""
-    # Check if SnsTimeLine exists, fall back to SnsTopItem_1
-    tables = [r[0] for r in cursor.execute(
-        "SELECT name FROM sqlite_master WHERE type='table';").fetchall()]
-
-    table = 'SnsTimeLine' if 'SnsTimeLine' in tables else None
-    if not table:
-        table = 'SnsTopItem_1' if 'SnsTopItem_1' in tables else None
-    if not table:
-        return {'success': False, 'error': 'No SNS table found'}
-
-    conditions = []
-    params = []
-
-    if table == 'SnsTimeLine':
-        if usernames:
-            placeholders = ','.join(['?' for _ in usernames])
-            conditions.append(f'user_name IN ({placeholders})')
-            params.extend(usernames)
-        if keyword:
-            conditions.append('content LIKE ?')
-            params.append(f'%{keyword}%')
-        if start_time:
-            conditions.append("CAST(substr(content, instr(content, '<createTime>') + 12, 10) AS INTEGER) >= ?")
-            params.append(start_time)
-        if end_time:
-            conditions.append("CAST(substr(content, instr(content, '<createTime>') + 12, 10) AS INTEGER) <= ?")
-            params.append(end_time)
-    elif table == 'SnsTopItem_1':
-        uname_col = 'username' if 'username' in [r[1] for r in cursor.execute(f'PRAGMA table_info([{table}]);').fetchall()] else 'user_name'
-        if usernames:
-            placeholders = ','.join(['?' for _ in usernames])
-            conditions.append(f'{uname_col} IN ({placeholders})')
-            params.extend(usernames)
-        if keyword:
-            conditions.append('summary LIKE ?')
-            params.append(f'%{keyword}%')
-        if start_time:
-            conditions.append('create_time >= ?')
-            params.append(start_time)
-        if end_time:
-            conditions.append('create_time <= ?')
-            params.append(end_time)
-
-    where = ' AND '.join(conditions) if conditions else '1=1'
-    order = 'tid DESC' if table == 'SnsTimeLine' else 'create_time DESC'
-
-    try:
-        cols = [r[1] for r in cursor.execute(f'PRAGMA table_info([{table}]);').fetchall()]
-        # Only select known text columns to avoid binary decode errors
-        text_cols = [c for c in cols if c in ('tid', 'user_name', 'content', 'username', 'summary',
-                                               'create_time', 'last_read_time', 'is_read',
-                                               'from_username', 'from_nickname', 'to_username',
-                                               'to_nickname', 'comment_id', 'feed_id',
-                                               'createTime', 'userName')]
-        if not text_cols:
-            text_cols = ['*']
-        select_str = ', '.join(text_cols)
-        rows = cursor.execute(
-            f'SELECT {select_str} FROM [{table}] WHERE {where} ORDER BY {order} LIMIT ? OFFSET ?',
-            params + [limit, offset]
-        ).fetchall()
-
-        timeline = []
-
-        for row in rows:
-            item = dict(zip(text_cols, row))
-
-            if table == 'SnsTimeLine':
-                # Parse XML content
-                content_str = item.get('content', '') or ''
-                parsed = parse_sns_content(content_str)
-                create_time = parsed['create_time'] or 0
-                username = parsed['username'] or item.get('user_name', '')
-                text = parsed['content'] or ''
-                media_count = parsed['media_count']
-            else:
-                create_time = item.get('create_time', 0) or 0
-                username = item.get('username', '') or item.get('user_name', '')
-                text = item.get('summary', '') or ''
-                media_count = 0
-
-            timeline.append({
-                'create_time': create_time,
-                'username': username,
-                'content': text,
-                'media_count': media_count,
-                'table': table,
-            })
-
-        return {'success': True, 'timeline': timeline, 'table': table}
-    except Exception as e:
-        return {'success': False, 'error': str(e)}
-
-
-def get_sns_usernames(cursor):
-    """Get unique usernames from SnsTopItem_1."""
-    tables = [r[0] for r in cursor.execute(
-        "SELECT name FROM sqlite_master WHERE type='table';").fetchall()]
-
-    # Prefer SnsTopItem_1 for user listing (larger)
-    table = 'SnsTopItem_1' if 'SnsTopItem_1' in tables else ('SnsTimeLine' if 'SnsTimeLine' in tables else None)
-    if not table:
-        return {'success': False, 'error': 'No SNS table found'}
-
-    # Discover username column
-    cols = [r[1] for r in cursor.execute(f'PRAGMA table_info([{table}]);').fetchall()]
-    uname_col = 'username' if 'username' in cols else 'user_name'
-
-    try:
-        rows = cursor.execute(
-            f'SELECT [{uname_col}], COUNT(*) as cnt FROM [{table}] GROUP BY [{uname_col}] ORDER BY cnt DESC'
-        ).fetchall()
-        usernames = [r[0] for r in rows if r[0]]
-        counts = {r[0]: r[1] for r in rows if r[0]}
-        return {'success': True, 'usernames': usernames, 'counts': counts}
-    except Exception as e:
-        return {'success': False, 'error': str(e)}
-
-
-def get_sns_stats(cursor, my_wxid=None):
-    """Get SNS statistics from SnsTopItem_1."""
-    tables = [r[0] for r in cursor.execute(
-        "SELECT name FROM sqlite_master WHERE type='table';").fetchall()]
-
-    table = 'SnsTopItem_1' if 'SnsTopItem_1' in tables else ('SnsTimeLine' if 'SnsTimeLine' in tables else None)
-    if not table:
-        return {'success': False, 'error': 'No SNS table found'}
-
-    try:
-        total = cursor.execute(f'SELECT COUNT(*) FROM [{table}]').fetchone()[0]
-
-        cols = [r[1] for r in cursor.execute(f'PRAGMA table_info([{table}]);').fetchall()]
-        uname_col = 'username' if 'username' in cols else 'user_name'
-
-        total_friends = 0
-        if uname_col:
-            total_friends = cursor.execute(
-                f'SELECT COUNT(DISTINCT [{uname_col}]) FROM [{table}] WHERE [{uname_col}] IS NOT NULL'
-            ).fetchone()[0]
-
-        my_posts = None
-        if my_wxid and uname_col:
-            my_posts = cursor.execute(
-                f'SELECT COUNT(*) FROM [{table}] WHERE [{uname_col}] = ?',
-                (my_wxid,)).fetchone()[0]
-
-        return {
-            'success': True,
-            'data': {
-                'totalPosts': total,
-                'totalFriends': total_friends,
-                'myPosts': my_posts,
-            }
-        }
-    except Exception as e:
-        return {'success': False, 'error': str(e)}
-
-
-# ========== Main CLI ==========
-
-def main():
-    import argparse
-    parser = argparse.ArgumentParser(description='WeChat NT Database Tool')
-    sub = parser.add_subparsers(dest='command')
-
-    # scan command
-    scan_parser = sub.add_parser('scan', help='Scan memory for keys and match NT databases')
-    scan_parser.add_argument('--json', action='store_true', help='Output as JSON')
-    scan_parser.add_argument('--root', default=os.environ.get('WEFLOW_SCAN_ROOT'), help='xwechat_files root directory override')
-
-    # sessions command
-    sessions_parser = sub.add_parser('sessions', help='List chat sessions')
-    sessions_parser.add_argument('--db', default=os.environ.get('WEFLOW_DB_PATH'), required=not os.environ.get('WEFLOW_DB_PATH'), help='Path to NT database')
-    sessions_parser.add_argument('--key', default=os.environ.get('WEFLOW_NT_KEY'), required=not os.environ.get('WEFLOW_NT_KEY'), help='Key hex (64 chars)')
-    sessions_parser.add_argument('--salt', default=os.environ.get('WEFLOW_NT_SALT'), required=not os.environ.get('WEFLOW_NT_SALT'), help='Salt hex (32 chars)')
-    sessions_parser.add_argument('--passphrase', default=os.environ.get('WEFLOW_NT_PASSPHRASE'), help='Shared passphrase for deriving per-shard keys')
-    sessions_parser.add_argument('--keyword', default=os.environ.get('WEFLOW_QUERY_KEYWORD'), help='Filter by keyword')
-    sessions_parser.add_argument('--contact-db', default=os.environ.get('WEFLOW_CONTACT_DB'), help='Path to contact.db for display names')
-    sessions_parser.add_argument('--contact-key', default=os.environ.get('WEFLOW_CONTACT_KEY'), help='Contact DB key hex (64 chars)')
-    sessions_parser.add_argument('--contact-salt', default=os.environ.get('WEFLOW_CONTACT_SALT'), help='Contact DB salt hex (32 chars)')
-
-    # messages command
-    msg_parser = sub.add_parser('messages', help='Get messages')
-    msg_parser.add_argument('--db', default=os.environ.get('WEFLOW_DB_PATH'), required=not os.environ.get('WEFLOW_DB_PATH'), help='Path to NT database')
-    msg_parser.add_argument('--key', default=os.environ.get('WEFLOW_NT_KEY'), required=not os.environ.get('WEFLOW_NT_KEY'), help='Key hex (64 chars)')
-    msg_parser.add_argument('--salt', default=os.environ.get('WEFLOW_NT_SALT'), required=not os.environ.get('WEFLOW_NT_SALT'), help='Salt hex (32 chars)')
-    msg_parser.add_argument('--passphrase', default=os.environ.get('WEFLOW_NT_PASSPHRASE'), help='Shared passphrase for deriving per-shard keys')
-    msg_parser.add_argument('--talker', default=os.environ.get('WEFLOW_TALKER'), required=not os.environ.get('WEFLOW_TALKER'), help='Talker username')
-    msg_parser.add_argument('--limit', type=int, default=100)
-    msg_parser.add_argument('--offset', type=int, default=0)
-    # Unix seconds, inclusive at both ends. Pushed into SQL, so a bounded read
-    # no longer has to fetch the whole conversation to discard most of it.
-    msg_parser.add_argument('--from', dest='from_time', type=int,
-                            default=os.environ.get('WEFLOW_FROM_TIME'),
-                            help='Only messages at or after this unix time')
-    msg_parser.add_argument('--to', dest='to_time', type=int,
-                            default=os.environ.get('WEFLOW_TO_TIME'),
-                            help='Only messages at or before this unix time')
-    msg_parser.add_argument('--contact-db', default=os.environ.get('WEFLOW_CONTACT_DB'), help='Path to contact.db for sender names')
-    msg_parser.add_argument('--contact-key', default=os.environ.get('WEFLOW_CONTACT_KEY'), help='Contact DB key hex (64 chars)')
-    msg_parser.add_argument('--contact-salt', default=os.environ.get('WEFLOW_CONTACT_SALT'), help='Contact DB salt hex (32 chars)')
-    msg_parser.add_argument('--own-wxid', default=os.environ.get('WEFLOW_OWN_WXID'), help='Account owner wxid (for self-message detection)')
-    # Opt-in: without it the JSON is byte-identical to before this flag existed.
-    msg_parser.add_argument('--report-shards', action='store_true',
-                            help='Add a per-shard read report to the result (additive)')
-
-    # contacts command
-    contacts_parser = sub.add_parser('contacts', help='List contacts')
-    contacts_parser.add_argument('--db', default=os.environ.get('WEFLOW_DB_PATH'), required=not os.environ.get('WEFLOW_DB_PATH'), help='Path to NT database')
-    contacts_parser.add_argument('--key', default=os.environ.get('WEFLOW_NT_KEY'), required=not os.environ.get('WEFLOW_NT_KEY'), help='Key hex (64 chars)')
-    contacts_parser.add_argument('--salt', default=os.environ.get('WEFLOW_NT_SALT'), required=not os.environ.get('WEFLOW_NT_SALT'), help='Salt hex (32 chars)')
-    contacts_parser.add_argument('--passphrase', default=os.environ.get('WEFLOW_NT_PASSPHRASE'), help='Shared passphrase for deriving per-shard keys')
-    contacts_parser.add_argument('--keyword', default=os.environ.get('WEFLOW_QUERY_KEYWORD'), help='Filter by keyword')
-    contacts_parser.add_argument('--limit', type=int, default=200)
-    contacts_parser.add_argument('--contact-db', default=os.environ.get('WEFLOW_CONTACT_DB'), help='Path to contact.db for display names')
-    contacts_parser.add_argument('--contact-key', default=os.environ.get('WEFLOW_CONTACT_KEY'), help='Contact DB key hex (64 chars)')
-    contacts_parser.add_argument('--contact-salt', default=os.environ.get('WEFLOW_CONTACT_SALT'), help='Contact DB salt hex (32 chars)')
-
-    # sns-timeline command
-    sns_tl_parser = sub.add_parser('sns-timeline', help='Get SNS/Moments timeline')
-    sns_tl_parser.add_argument('--db', default=os.environ.get('WEFLOW_DB_PATH'), required=not os.environ.get('WEFLOW_DB_PATH'), help='Path to sns.db')
-    sns_tl_parser.add_argument('--key', default=os.environ.get('WEFLOW_NT_KEY'), required=not os.environ.get('WEFLOW_NT_KEY'), help='Key hex (64 chars)')
-    sns_tl_parser.add_argument('--salt', default=os.environ.get('WEFLOW_NT_SALT'), required=not os.environ.get('WEFLOW_NT_SALT'), help='Salt hex (32 chars)')
-    sns_tl_parser.add_argument('--limit', type=int, default=20)
-    sns_tl_parser.add_argument('--offset', type=int, default=0)
-    sns_tl_parser.add_argument('--usernames', default=os.environ.get('WEFLOW_QUERY_USERNAMES'), help='JSON array of usernames to filter')
-    sns_tl_parser.add_argument('--keyword', default=os.environ.get('WEFLOW_QUERY_KEYWORD'), help='Search keyword')
-    sns_tl_parser.add_argument('--start-time', type=int, help='Start timestamp')
-    sns_tl_parser.add_argument('--end-time', type=int, help='End timestamp')
-
-    # sns-usernames command
-    sns_un_parser = sub.add_parser('sns-usernames', help='List usernames with SNS posts')
-    sns_un_parser.add_argument('--db', default=os.environ.get('WEFLOW_DB_PATH'), required=not os.environ.get('WEFLOW_DB_PATH'), help='Path to sns.db')
-    sns_un_parser.add_argument('--key', default=os.environ.get('WEFLOW_NT_KEY'), required=not os.environ.get('WEFLOW_NT_KEY'), help='Key hex (64 chars)')
-    sns_un_parser.add_argument('--salt', default=os.environ.get('WEFLOW_NT_SALT'), required=not os.environ.get('WEFLOW_NT_SALT'), help='Salt hex (32 chars)')
-
-    # sns-stats command
-    sns_stats_parser = sub.add_parser('sns-stats', help='SNS statistics')
-    sns_stats_parser.add_argument('--db', default=os.environ.get('WEFLOW_DB_PATH'), required=not os.environ.get('WEFLOW_DB_PATH'), help='Path to sns.db')
-    sns_stats_parser.add_argument('--key', default=os.environ.get('WEFLOW_NT_KEY'), required=not os.environ.get('WEFLOW_NT_KEY'), help='Key hex (64 chars)')
-    sns_stats_parser.add_argument('--salt', default=os.environ.get('WEFLOW_NT_SALT'), required=not os.environ.get('WEFLOW_NT_SALT'), help='Salt hex (32 chars)')
-    sns_stats_parser.add_argument('--my-wxid', default=os.environ.get('WEFLOW_OWN_WXID'), help='Account owner wxid for my-posts count')
-
-    # fav-schema command
-    fav_schema_parser = sub.add_parser('fav-schema', help='Dump favorite.db schema (key verification)')
-    fav_schema_parser.add_argument('--db', default=os.environ.get('WEFLOW_DB_PATH'), required=not os.environ.get('WEFLOW_DB_PATH'), help='Path to favorite.db')
-    fav_schema_parser.add_argument('--key', default=os.environ.get('WEFLOW_NT_KEY'), required=not os.environ.get('WEFLOW_NT_KEY'), help='Key hex (64 chars)')
-
-    verify_parser = sub.add_parser('verify', help='Verify a key+salt can open a database')
-    verify_parser.add_argument('--db', default=os.environ.get('WEFLOW_DB_PATH'), required=not os.environ.get('WEFLOW_DB_PATH'), help='Path to NT database')
-    verify_parser.add_argument('--key', default=os.environ.get('WEFLOW_NT_KEY'), required=not os.environ.get('WEFLOW_NT_KEY'), help='Key hex (64 chars)')
-    verify_parser.add_argument('--salt', default=os.environ.get('WEFLOW_NT_SALT'), required=not os.environ.get('WEFLOW_NT_SALT'), help='Salt hex (32 chars)')
-
-    # Same question as `verify`, answered without the native library - so a
-    # missing or broken sqlcipher3 does not hide whether the key is correct.
-    native_parser = sub.add_parser('verify-native',
-                                   help='Verify a passphrase from the file header alone (no sqlcipher3)')
-    native_parser.add_argument('--db', default=os.environ.get('WEFLOW_DB_PATH'), required=not os.environ.get('WEFLOW_DB_PATH'), help='Path to any database in the set')
-    native_parser.add_argument('--passphrase', default=os.environ.get('WEFLOW_NT_PASSPHRASE'), required=not os.environ.get('WEFLOW_NT_PASSPHRASE'), help='Passphrase hex (64 chars)')
-    native_parser.add_argument('--internal-key', default=os.environ.get('WEFLOW_INTERNAL_DB_KEY', ''), help='Optional 64-char hex XOR key (some builds)')
-
-    # fav-list command
-    fav_list_parser = sub.add_parser('fav-list', help='List favorite items')
-    fav_list_parser.add_argument('--db', default=os.environ.get('WEFLOW_DB_PATH'), required=not os.environ.get('WEFLOW_DB_PATH'), help='Path to favorite.db')
-    fav_list_parser.add_argument('--key', default=os.environ.get('WEFLOW_NT_KEY'), required=not os.environ.get('WEFLOW_NT_KEY'), help='Key hex (64 chars)')
-    fav_list_parser.add_argument('--limit', type=int, default=100)
-    fav_list_parser.add_argument('--offset', type=int, default=0)
-    fav_list_parser.add_argument('--keyword', default=os.environ.get('WEFLOW_QUERY_KEYWORD'), help='Filter by keyword')
-    fav_list_parser.add_argument('--type', type=int, dest='fav_type',
-                                 help='Filter by type: 1=text 2=image 4=video 5=article 14=chatrecord')
-
-    # cursor-probe command: the data-side half of D-027's stable-cursor answer.
-    cursor_parser = sub.add_parser(
-        'cursor-probe',
-        help='Probe what a stable incremental cursor would need here (read-only, bounded sample)')
-    cursor_parser.add_argument('--db', default=os.environ.get('WEFLOW_DB_PATH'), required=not os.environ.get('WEFLOW_DB_PATH'), help='Path to NT database')
-    cursor_parser.add_argument('--key', default=os.environ.get('WEFLOW_NT_KEY'), required=not os.environ.get('WEFLOW_NT_KEY'), help='Key hex (64 chars)')
-    cursor_parser.add_argument('--salt', default=os.environ.get('WEFLOW_NT_SALT'), required=not os.environ.get('WEFLOW_NT_SALT'), help='Salt hex (32 chars)')
-    cursor_parser.add_argument('--passphrase', default=os.environ.get('WEFLOW_NT_PASSPHRASE'), help='Shared passphrase for deriving per-shard keys')
-    cursor_parser.add_argument('--tables-per-shard', type=int, default=5,
-                               help='Message tables sampled per shard for the identity half (schema half is exhaustive)')
-    cursor_parser.add_argument('--rows-per-table', type=int, default=500,
-                               help='Rows read at each rowid end per sampled table')
-
-    args = parser.parse_args()
-
-    if args.command == 'scan':
-        # Discover the database files before touching the process. Walking the
-        # filesystem does not need WeChat running, but key matching does - and
-        # the caller derives keys from the passphrase when the memory scan
-        # finds nothing, which only needs the file list. Returning early on
-        # "process not running" without the databases threw that list away and
-        # left the caller with nothing to derive from.
-        databases = find_nt_databases(getattr(args, 'root', None))
-
-        pid = find_weixin_pid()
-        if not pid:
-            if IS_WINDOWS:
-                print(json.dumps({"error": "Weixin.exe 未运行", "databases": databases}))
-            else:
-                print(json.dumps({"error": "未检测到 Linux 微信进程，请确认微信已启动并登录",
-                                  "databases": databases}))
-            return
-
-        if not args.json:
-            print(f"扫描进程 PID {pid}...")
-
-        keys, scan_err = scan_memory_keys(pid)
-        if scan_err == 'permission':
-            print(json.dumps({"error": "PERMISSION_DENIED: 读取微信进程内存需要 root 或 CAP_SYS_PTRACE 权限。可使用 sudo 运行，或执行: sudo setcap cap_sys_ptrace=ep $(which python3)",
-                              "databases": databases}))
-            return
-        if scan_err == 'gone':
-            print(json.dumps({"error": "微信进程已退出，请重试", "databases": databases}))
-            return
-        if not args.json:
-            print(f"找到 {len(keys)} 个密钥")
-
-        if not args.json:
-            print(f"找到 {len(databases)} 个 NT 数据库")
-
-        matched = match_keys_to_databases(keys, databases)
-        if not args.json:
-            print(f"匹配 {len(matched)} 个数据库")
-            for db in matched:
-                print(f"  {db['name']} ({db['size']/1024/1024:.1f}MB) key={db['key'][:16]}... salt={db['salt'][:16]}...")
-        else:
-            print(json.dumps({"keys": keys, "databases": databases, "matched": matched}))
-        return
-
-    if args.command == 'verify':
-        # sqlcipher 打开+读 sqlite_master 才触发解密, 错误密钥在此报 "file is not a database"
-        try:
-            conn, _ = connect_nt_db(args.db, args.key, args.salt)
-            n = conn.execute(
-                "SELECT count(*) FROM sqlite_master WHERE type='table'"
-            ).fetchone()[0]
-            conn.close()
-            print(json.dumps({"success": n > 0, "tables": n}, ensure_ascii=True))
-        except Exception as e:
-            print(json.dumps({"success": False, "error": str(e).split('\n')[0][:200]}, ensure_ascii=True))
-        return
-
-    if args.command == 'verify-native':
-        verdict = verify_passphrase_native(args.passphrase, args.db,
-                                           getattr(args, 'internal_key', '') or '')
-        if verdict is None:
-            print(json.dumps({"success": False, "verified": None,
-                              "error": "无法判定（文件不可读、过短，或 passphrase 不是 64 位 hex）"},
-                             ensure_ascii=True))
-            return
-        print(json.dumps({"success": True, "verified": verdict}, ensure_ascii=True))
-        return
-
-    # Build contact name map once if contact db provided
-    contact_name_map = {}
-    contact_db = getattr(args, 'contact_db', None)
-    contact_key = getattr(args, 'contact_key', None)
-    contact_salt = getattr(args, 'contact_salt', None)
-    if contact_db and contact_key and contact_salt:
-        contact_name_map = load_contact_names(contact_db, contact_key, contact_salt)
-
-    if args.command == 'sessions':
-        conns = connect_message_shards(args.db, args.key, args.salt, getattr(args, 'passphrase', '') or '')
-        if not conns:
-            print(json.dumps({"error": "无法打开消息数据库，请检查密钥"}, ensure_ascii=True))
-            return
-        result = get_sessions(conns)
-        if 'sessions' in result:
-            result['sessions'] = apply_contact_names(result['sessions'], contact_name_map)
-            if args.keyword:
-                kw = args.keyword.lower()
-                result['sessions'] = [
-                    s for s in result['sessions']
-                    if kw in (s.get('username', '') + s.get('displayName', '') + s.get('summary', '')).lower()
-                ]
-        print(json.dumps(result, ensure_ascii=True))
-        for conn in conns:
-            conn.close()
-
-    elif args.command == 'messages':
-        passphrase = getattr(args, 'passphrase', '') or ''
-        own_wxid = getattr(args, 'own_wxid', None)
-
-        if getattr(args, 'report_shards', False):
-            pairs, failures = connect_message_shards_detailed(
-                args.db, args.key, args.salt, passphrase)
-            if not pairs:
-                print(json.dumps({"error": "无法打开消息数据库，请检查密钥"}, ensure_ascii=True))
-                return
-            conns = [conn for _path, conn in pairs]
-            names = [os.path.basename(path) for path, _conn in pairs]
-            shard_items = []
-            result = get_messages(conns, args.talker, args.limit, args.offset,
-                                  contact_name_map, own_wxid,
-                                  shard_names=names, shard_report=shard_items,
-                                  from_time=args.from_time, to_time=args.to_time)
-            # Failure entries carry no opened connection, so they are not in
-            # `conns` and could not be recorded by get_messages itself.
-            result['shards'] = {
-                'scanned': len(pairs) + len(failures),
-                'opened': len(pairs),
-                'failed': len(failures),
-                'items': failures + shard_items,
-            }
-        else:
-            conns = connect_message_shards(args.db, args.key, args.salt, passphrase)
-            if not conns:
-                print(json.dumps({"error": "无法打开消息数据库，请检查密钥"}, ensure_ascii=True))
-                return
-            result = get_messages(conns, args.talker, args.limit, args.offset,
-                                  contact_name_map, own_wxid,
-                                  from_time=args.from_time, to_time=args.to_time)
-
-        print(json.dumps(result, ensure_ascii=True))
-        for conn in conns:
-            conn.close()
-
-    elif args.command == 'contacts':
-        conns = connect_message_shards(args.db, args.key, args.salt, getattr(args, 'passphrase', '') or '')
-        if not conns:
-            print(json.dumps({"error": "无法打开消息数据库，请检查密钥"}, ensure_ascii=True))
-            return
-        result = get_contacts(conns, args.limit, keyword=args.keyword)
-        if 'contacts' in result:
-            result['contacts'] = apply_contact_names(result['contacts'], contact_name_map)
-            result['contacts'] = filter_contacts(result['contacts'], args.keyword, args.limit)
-        print(json.dumps(result, ensure_ascii=True))
-        for conn in conns:
-            conn.close()
-
-    elif args.command == 'sns-timeline':
-        conn, _ = connect_nt_db(args.db, args.key, args.salt)
-        usernames = None
-        if args.usernames:
-            try:
-                usernames = json.loads(args.usernames)
-            except: pass
-        result = get_sns_timeline(conn.cursor(), args.limit, args.offset,
-                                   usernames, args.keyword,
-                                   args.start_time, args.end_time)
-        print(json.dumps(result, ensure_ascii=True, default=str))
-        conn.close()
-
-    elif args.command == 'sns-usernames':
-        conn, _ = connect_nt_db(args.db, args.key, args.salt)
-        result = get_sns_usernames(conn.cursor())
-        print(json.dumps(result, ensure_ascii=True, default=str))
-        conn.close()
-
-    elif args.command == 'sns-stats':
-        conn, _ = connect_nt_db(args.db, args.key, args.salt)
-        result = get_sns_stats(conn.cursor(), args.my_wxid)
-        print(json.dumps(result, ensure_ascii=True, default=str))
-        conn.close()
-
-    elif args.command == 'fav-schema':
-        result = get_fav_schema(args.db, args.key)
-        print(json.dumps(result, ensure_ascii=True))
-
-    elif args.command == 'fav-list':
-        result = get_favorites(args.db, args.key, args.limit, args.offset,
-                               args.keyword, getattr(args, 'fav_type', None))
-        print(json.dumps(result, ensure_ascii=True, default=str))
-
-    elif args.command == 'cursor-probe':
-        result = cursor_probe(args.db, args.key, args.salt,
-                              getattr(args, 'passphrase', '') or '',
-                              tables_per_shard=args.tables_per_shard,
-                              rows_per_table=args.rows_per_table)
-        print(json.dumps(result, ensure_ascii=True))
-
-    else:
-        parser.print_help()
-
-
-if __name__ == '__main__':
-    main()
+#!/usr/bin/env python3
+"""
+WeChat NT (4.x) Database Access Tool
+Uses sqlcipher3 to decrypt and query NT-format databases.
+"""
+import sys
+import os
+import json
+import html
+import re
+import hmac
+import struct
+import hashlib
+import ctypes
+from ctypes import wintypes, c_void_p, c_size_t, create_string_buffer, byref, sizeof
+from pathlib import Path
+
+sqlcipher = None
+
+
+def require_sqlcipher():
+    """Load SQLCipher only for database operations, not path discovery."""
+    global sqlcipher
+    if sqlcipher is not None:
+        return sqlcipher
+    try:
+        from sqlcipher3 import dbapi2 as sqlcipher_module
+    except ImportError as error:
+        raise RuntimeError("需要 sqlcipher3: pip install sqlcipher3") from error
+    sqlcipher = sqlcipher_module
+    return sqlcipher
+
+# ========== Memory Scanner ==========
+PROCESS_VM_READ = 0x0010
+PROCESS_QUERY_INFORMATION = 0x0400
+MEM_COMMIT = 0x1000
+PAGE_NOACCESS = 0x01
+PAGE_GUARD = 0x100
+
+class MEMORY_BASIC_INFORMATION(ctypes.Structure):
+    _fields_ = [
+        ('BaseAddress', ctypes.c_void_p),
+        ('AllocationBase', ctypes.c_void_p),
+        ('AllocationProtect', wintypes.DWORD),
+        ('PartitionId', wintypes.WORD),
+        ('RegionSize', ctypes.c_size_t),
+        ('State', wintypes.DWORD),
+        ('Protect', wintypes.DWORD),
+        ('Type', wintypes.DWORD),
+    ]
+
+IS_WINDOWS = os.name == 'nt'
+
+if IS_WINDOWS:
+    kernel32 = ctypes.windll.kernel32
+    ReadProcessMemory = kernel32.ReadProcessMemory
+    ReadProcessMemory.argtypes = [wintypes.HANDLE, wintypes.LPCVOID, wintypes.LPVOID, ctypes.c_size_t, ctypes.POINTER(c_size_t)]
+    ReadProcessMemory.restype = wintypes.BOOL
+    VirtualQueryEx = kernel32.VirtualQueryEx
+    VirtualQueryEx.argtypes = [wintypes.HANDLE, wintypes.LPCVOID, ctypes.c_void_p, ctypes.c_size_t]
+    VirtualQueryEx.restype = ctypes.c_size_t
+else:
+    kernel32 = None
+
+
+def find_weixin_pid():
+    """Find WeChat process ID (Windows: pymem; Linux: /proc scan)."""
+    if not IS_WINDOWS:
+        return find_weixin_pid_linux()
+    try:
+        import pymem, pymem.process
+        for proc in pymem.process.list_processes():
+            try:
+                name = proc.szExeFile
+                if isinstance(name, bytes):
+                    name = name.decode('utf-8', errors='ignore')
+                if name.lower() == 'weixin.exe':
+                    return proc.th32ProcessID
+            except:
+                pass
+    except ImportError:
+        pass
+    return None
+
+
+_LINUX_WECHAT_COMMS = {'wechat', 'wechatappex', 'weixin'}
+_LINUX_EXE_PREFIX_DENY = ('python', 'bash', 'sh', 'zsh', 'node', 'perl', 'ruby', 'electron')
+
+
+def _is_linux_wechat_process(pid):
+    if pid == os.getpid():
+        return False
+    try:
+        with open(f'/proc/{pid}/comm', encoding='utf-8') as f:
+            comm = f.read().strip().lower()
+        if comm in _LINUX_WECHAT_COMMS:
+            return True
+        try:
+            exe = os.path.realpath(os.readlink(f'/proc/{pid}/exe'))
+        except OSError:
+            return False
+        name = os.path.basename(exe).lower()
+        if any(name.startswith(p) for p in _LINUX_EXE_PREFIX_DENY):
+            return False
+        return 'wechat' in name or 'weixin' in name
+    except (PermissionError, FileNotFoundError, ProcessLookupError):
+        return False
+
+
+def find_weixin_pid_linux():
+    """Find Linux WeChat main process (largest RSS among candidates)."""
+    best = None
+    best_rss = -1
+    try:
+        pids = os.listdir('/proc')
+    except OSError:
+        return None
+    for pid_str in pids:
+        if not pid_str.isdigit():
+            continue
+        pid = int(pid_str)
+        if not _is_linux_wechat_process(pid):
+            continue
+        try:
+            with open(f'/proc/{pid}/statm', encoding='utf-8') as f:
+                rss_kb = int(f.read().split()[1]) * 4
+        except (OSError, IndexError, ValueError):
+            rss_kb = 0
+        if rss_kb > best_rss:
+            best_rss = rss_kb
+            best = pid
+    return best
+
+
+def scan_memory_keys(pid):
+    """Scan process memory for x'<64hex_key><32hex_salt>' patterns.
+
+    Returns (keys, error): keys is a list of {"key","salt"} dicts,
+    error is None on success or 'permission' / 'gone' / 'not_windows'.
+    """
+    if not IS_WINDOWS:
+        return scan_memory_keys_linux(pid)
+
+    hProcess = kernel32.OpenProcess(PROCESS_VM_READ | PROCESS_QUERY_INFORMATION, False, pid)
+    if not hProcess:
+        return [], None
+
+    pattern = re.compile(rb"x'([0-9a-fA-F]{64})([0-9a-fA-F]{32})'")
+    keys_found = []
+    address = 0x10000
+
+    while address < 0x7FFFFFFFFFFF:
+        mbi = MEMORY_BASIC_INFORMATION()
+        result = VirtualQueryEx(hProcess, ctypes.c_void_p(address), ctypes.byref(mbi), sizeof(mbi))
+        if result == 0:
+            break
+
+        region_addr = mbi.BaseAddress or 0
+        region_size = mbi.RegionSize or 0
+
+        if (mbi.State == MEM_COMMIT and
+                region_size > 256 and region_size < 200 * 1024 * 1024 and
+                mbi.Protect not in (0, PAGE_NOACCESS, PAGE_GUARD)):
+
+            pos = region_addr
+            end = region_addr + region_size
+            while pos < end:
+                chunk_size = min(65536, end - pos)
+                buf = create_string_buffer(chunk_size)
+                bytesRead = c_size_t(0)
+                ok = ReadProcessMemory(hProcess, ctypes.c_void_p(pos), buf, chunk_size, byref(bytesRead))
+                if ok and bytesRead.value > 0:
+                    data = buf.raw[:bytesRead.value]
+                    for m in pattern.finditer(data):
+                        key_hex = m.group(1).decode()
+                        salt_hex = m.group(2).decode()
+                        keys_found.append((key_hex, salt_hex))
+                pos += chunk_size
+
+        address = region_addr + region_size
+
+    kernel32.CloseHandle(hProcess)
+
+    # Deduplicate
+    seen = set()
+    unique_keys = []
+    for k, s in keys_found:
+        pair = (k, s)
+        if pair not in seen:
+            seen.add(pair)
+            unique_keys.append({"key": k, "salt": s})
+
+    return unique_keys, None
+
+
+_LINUX_SKIP_MAPPINGS = {'[vdso]', '[vsyscall]', '[vvar]'}
+_LINUX_SKIP_PREFIXES = ('/usr/lib/', '/lib/', '/usr/share/')
+
+
+def scan_memory_keys_linux(pid):
+    """Scan /proc/<pid>/maps + /proc/<pid>/mem for the key pattern.
+
+    Requires root or CAP_SYS_PTRACE (or the target being a descendant
+    of this process when yama ptrace_scope=1).
+    """
+    regions = []
+    try:
+        with open(f'/proc/{pid}/maps', encoding='utf-8') as f:
+            for line in f:
+                parts = line.split()
+                if len(parts) < 2 or 'r' not in parts[1]:
+                    continue
+                if len(parts) >= 6:
+                    name = parts[5]
+                    if name in _LINUX_SKIP_MAPPINGS:
+                        continue
+                    name_lower = name.lower()
+                    if name.startswith(_LINUX_SKIP_PREFIXES) and \
+                            'wcdb' not in name_lower and 'wechat' not in name_lower and 'weixin' not in name_lower:
+                        continue
+                try:
+                    start_s, end_s = parts[0].split('-')
+                    start = int(start_s, 16)
+                    size = int(end_s, 16) - start
+                except ValueError:
+                    continue
+                if 0 < size < 500 * 1024 * 1024:
+                    regions.append((start, size))
+    except PermissionError:
+        return [], 'permission'
+    except (FileNotFoundError, ProcessLookupError):
+        return [], 'gone'
+
+    pattern = re.compile(rb"x'([0-9a-fA-F]{64})([0-9a-fA-F]{32})'")
+    keys_found = []
+    try:
+        with open(f'/proc/{pid}/mem', 'rb') as mem:
+            for base, size in regions:
+                try:
+                    mem.seek(base)
+                    data = mem.read(size)
+                except (OSError, ValueError):
+                    continue
+                for m in pattern.finditer(data):
+                    keys_found.append((m.group(1).decode(), m.group(2).decode()))
+    except PermissionError:
+        return [], 'permission'
+    except (FileNotFoundError, ProcessLookupError):
+        return [], 'gone'
+
+    seen = set()
+    unique_keys = []
+    for k, s in keys_found:
+        pair = (k, s)
+        if pair not in seen:
+            seen.add(pair)
+            unique_keys.append({"key": k, "salt": s})
+
+    return unique_keys, None
+
+
+# ========== NT Database Discovery ==========
+
+def _is_nt_account_dir(path):
+    return os.path.isdir(os.path.join(path, 'db_storage')) or \
+        os.path.isdir(os.path.join(path, 'Msg'))
+
+
+def _normalize_nt_root(root):
+    if not root:
+        return None
+    path = os.path.abspath(os.path.expandvars(os.path.expanduser(root)))
+    if os.path.isfile(path):
+        path = os.path.dirname(path)
+    if not os.path.isdir(path):
+        return None
+
+    try:
+        for entry in os.listdir(path):
+            candidate = os.path.join(path, entry)
+            if os.path.isdir(candidate) and _is_nt_account_dir(candidate):
+                return path
+    except OSError:
+        return None
+
+    for _ in range(6):
+        if _is_nt_account_dir(path):
+            return os.path.dirname(path)
+        parent = os.path.dirname(path)
+        if parent == path:
+            break
+        path = parent
+
+    return None
+
+
+def find_nt_databases(root=None):
+    """Find all NT-format databases under xwechat_files (message + contact)."""
+    if root:
+        normalized_root = _normalize_nt_root(root)
+        candidates = [normalized_root] if normalized_root else []
+    elif IS_WINDOWS:
+        candidates = [
+            os.path.expandvars(r'%USERPROFILE%\xwechat_files'),
+            os.path.expandvars(r'%USERPROFILE%\Documents\xwechat_files'),
+        ]
+    else:
+        home = os.path.expanduser('~')
+        candidates = [
+            os.path.join(home, '.local', 'share', 'xwechat_files'),
+            os.path.join(home, 'xwechat_files'),
+            os.path.join(home, 'Documents', 'xwechat_files'),
+            os.path.join(home, '文档', 'xwechat_files'),
+        ]
+    xwechat = None
+    for c in candidates:
+        if os.path.isdir(c):
+            xwechat = c
+            break
+    if not xwechat:
+        return []
+
+    databases = []
+    for wxid_dir in os.listdir(xwechat):
+        # Scan message databases
+        msg_storage = os.path.join(xwechat, wxid_dir, 'db_storage', 'message')
+        if os.path.isdir(msg_storage):
+            for f in os.listdir(msg_storage):
+                if f.endswith('.db') and not any(x in f for x in ['-shm', '-wal']):
+                    full_path = os.path.join(msg_storage, f)
+                    try:
+                        with open(full_path, 'rb') as fh:
+                            salt = fh.read(16)
+                        databases.append({
+                            "path": full_path,
+                            "name": f"message/{f}",
+                            "salt": salt.hex(),
+                            "size": os.path.getsize(full_path),
+                            "wxid": wxid_dir,
+                        })
+                    except:
+                        pass
+
+        # Scan contact database
+        contact_db = os.path.join(xwechat, wxid_dir, 'db_storage', 'contact', 'contact.db')
+        if os.path.isfile(contact_db):
+            try:
+                with open(contact_db, 'rb') as fh:
+                    salt = fh.read(16)
+                databases.append({
+                    "path": contact_db,
+                    "name": "contact/contact.db",
+                    "salt": salt.hex(),
+                    "size": os.path.getsize(contact_db),
+                    "wxid": wxid_dir,
+                })
+            except:
+                pass
+
+        # Scan SNS (朋友圈) database
+        sns_db = os.path.join(xwechat, wxid_dir, 'db_storage', 'sns', 'sns.db')
+        if os.path.isfile(sns_db):
+            try:
+                with open(sns_db, 'rb') as fh:
+                    salt = fh.read(16)
+                databases.append({
+                    "path": sns_db,
+                    "name": "sns/sns.db",
+                    "salt": salt.hex(),
+                    "size": os.path.getsize(sns_db),
+                    "wxid": wxid_dir,
+                })
+            except:
+                pass
+
+        # Scan favorites (收藏) database
+        fav_db = os.path.join(xwechat, wxid_dir, 'db_storage', 'favorite', 'favorite.db')
+        if os.path.isfile(fav_db):
+            try:
+                with open(fav_db, 'rb') as fh:
+                    salt = fh.read(16)
+                databases.append({
+                    "path": fav_db,
+                    "name": "favorite/favorite.db",
+                    "salt": salt.hex(),
+                    "size": os.path.getsize(fav_db),
+                    "wxid": wxid_dir,
+                })
+            except:
+                pass
+
+    return databases
+
+
+def find_contact_db_path(message_db_path):
+    """Derive contact.db path from message_0.db path.
+
+    message_0.db:  <xwechat_files>/<wxid>/db_storage/message/message_0.db
+    contact.db:    <xwechat_files>/<wxid>/db_storage/contact/contact.db
+    """
+    msg_dir = os.path.dirname(message_db_path)
+    wxid_dir = os.path.dirname(msg_dir)     # .../db_storage
+    xwechat_dir = os.path.dirname(wxid_dir) # .../<wxid>
+    contact_db = os.path.join(xwechat_dir, 'db_storage', 'contact', 'contact.db')
+    if os.path.isfile(contact_db):
+        return contact_db
+    return None
+
+
+def load_contact_names(contact_db_path, contact_key_hex, contact_salt_hex):
+    """Load wxid -> {remark, nick_name} map from contact.db.
+
+    Returns dict: {wxid: display_name}
+    display_name priority: remark > nick_name > alias > wxid
+
+    取不到就返回空表（名字是装饰性的，调用方会退回 wxid）。**连接必须在 finally
+    里关**：原先 `conn.close()` 写在 try 末尾，异常路径上泄漏——而在 Windows 上
+    一个没关的连接会把文件锁住（同一个坑早先在这个文件的分片读取里踩过一次）。
+    读了一半也保留已经拿到的名字，而不是整份丢掉。
+    """
+    if not contact_db_path or not contact_key_hex or not contact_salt_hex:
+        return {}
+    if not os.path.isfile(contact_db_path):
+        return {}
+
+    name_map = {}
+    conn = None
+    try:
+        raw_key = f"x'{contact_key_hex}{contact_salt_hex}'"
+        conn = require_sqlcipher().connect(contact_db_path)
+        c = conn.cursor()
+        c.execute(f'PRAGMA key = "{raw_key}";')
+
+        # contact.db schema: username, alias, remark, nick_name, ...
+        c.execute("SELECT username, COALESCE(NULLIF(remark,''), NULLIF(nick_name,''), NULLIF(alias,''), username) FROM contact")
+        for username, display in c.fetchall():
+            if username:
+                name_map[username] = display
+    except Exception:
+        pass
+    finally:
+        if conn is not None:
+            try:
+                conn.close()
+            except Exception:
+                pass
+    return name_map
+
+
+def filter_contacts(contacts, keyword, limit=0):
+    """按关键字过滤联系人，**然后**才截断到 `limit`。
+
+    顺序是这里的全部要点：先截断会让第 `limit` 个之后的人搜不到（实测 500 人的通讯录里，
+    按备注名找 10 个只命中 3 个）。关键字比对 `username + displayName + remark + nickname`，
+    大小写不敏感。
+    """
+    if not keyword:
+        return contacts[:limit] if limit else contacts
+    needle = keyword.lower()
+    hits = [
+        c for c in contacts
+        if needle in (c.get('username', '') + c.get('displayName', '')
+                      + c.get('remark', '') + c.get('nickname', '')).lower()
+    ]
+    return hits[:limit] if limit else hits
+
+
+def apply_contact_names(sessions, name_map):
+    """Apply contact names to session list, replacing bare wxid displayNames."""
+    if not name_map:
+        return sessions
+    for s in sessions:
+        username = s.get('username', '')
+        if username in name_map:
+            s['displayName'] = name_map[username]
+    return sessions
+
+
+def match_keys_to_databases(keys, databases, use_cache=True):
+    """Match memory keys to databases by comparing salts.
+
+    Args:
+        keys: list of {"key": str, "salt": str} from memory scan
+        databases: list of {"path": str, "salt": str, ...}
+        use_cache: if True, try cache first and store new keys
+
+    Returns:
+        list of matched databases with "key" field added
+    """
+    cache = get_key_cache() if use_cache else None
+
+    # Build salt -> key map from memory scan
+    salt_to_key = {}
+    for k in keys:
+        salt_to_key[k["salt"]] = k["key"]
+
+    # Store newly scanned keys to cache
+    if cache and keys:
+        cache.put_batch(keys)
+
+    matched = []
+    unmatched = []
+
+    for db in databases:
+        db_salt = db["salt"]
+
+        # Try cache first
+        if cache:
+            cached_key = cache.get(db_salt)
+            if cached_key:
+                db["key"] = cached_key
+                db["key_source"] = "cache"
+                matched.append(db)
+                continue
+
+        # Try memory scan result
+        if db_salt in salt_to_key:
+            db["key"] = salt_to_key[db_salt]
+            db["key_source"] = "memory"
+            matched.append(db)
+        else:
+            unmatched.append(db)
+
+    return matched
+
+
+def match_databases_from_cache_only(databases):
+    """Match databases using only cached keys (no memory scan).
+
+    用于微信未运行或不想扫内存的场景。
+
+    Args:
+        databases: list of {"path": str, "salt": str, ...}
+
+    Returns:
+        tuple: (matched, unmatched)
+            matched: list of databases with "key" field
+            unmatched: list of databases without key
+    """
+    cache = get_key_cache()
+    matched = []
+    unmatched = []
+
+    for db in databases:
+        db_salt = db["salt"]
+        cached_key = cache.get(db_salt)
+
+        if cached_key:
+            db["key"] = cached_key
+            db["key_source"] = "cache"
+            matched.append(db)
+        else:
+            unmatched.append(db)
+
+    return matched, unmatched
+
+
+# ========== Database Operations ==========
+
+def connect_nt_db(db_path, key_hex, salt_hex):
+    """Connect to an NT database using sqlcipher3."""
+    raw_key = f"x'{key_hex}{salt_hex}'"
+    conn = require_sqlcipher().connect(db_path)
+    c = conn.cursor()
+    c.execute(f'PRAGMA key = "{raw_key}";')
+    return conn, c
+
+
+# WeChat rolls a conversation into a new shard over time, and every shard is
+# encrypted with its own key. Reading only the configured database therefore
+# shows a transcript that stops wherever the first shard's last write left off
+# - which is silent, because the query still succeeds. Only the HTML exporter
+# used to merge shards, so the same chat exported two different histories
+# depending on the format asked for.
+
+# 分片发现、密钥派生、列探测都在 nt_common 里——**只此一份**。
+# 原先 `nt_decrypt` 与 `export_chat_html` 各有一份，而契约并不相同。
+#
+# 这一行不能省：本文件被 `spec_from_file_location` 加载时（测试就是这么加的），
+# 解释器**不会**把脚本目录放进 sys.path——只有直接运行才会。少了它，单独跑
+# `nt_decrypt_shards_test.py` 会以 ModuleNotFoundError 失败，而全套一起跑却能过
+# （别的测试文件先插了路径）。那种"碰巧能过"正是要避免的。
+sys.path.insert(0, os.path.dirname(os.path.abspath(__file__)))
+from nt_common import (discover_message_shards, derive_database_key,
+                       table_columns, MESSAGE_ANCHOR_COLUMNS)
+from key_cache import get_key_cache
+
+
+def verify_passphrase_native(passphrase_hex, db_path, internal_key_hex=''):
+    """Check a passphrase against a database without sqlcipher3.
+
+    Verifies the SQLCipher page-1 HMAC directly with stdlib primitives:
+
+        mac_salt = header_salt XOR 0x3a
+        key      = PBKDF2-HMAC-SHA512(passphrase, header_salt, 256000, 32)
+        mac_key  = PBKDF2-HMAC-SHA512(key, mac_salt, 2, 32)
+        compare HMAC-SHA512(mac_key, page1_body || page_number=1) with the
+        digest stored at the end of page 1
+
+    Why this exists: the normal way to check a key is to open the database,
+    which needs the native library. When sqlcipher3 is missing or misbehaving,
+    that route says nothing about whether the key itself is right - and the
+    two failures look identical. This separates them.
+
+    Returns True (correct), False (wrong), or None (cannot tell - bad input,
+    file too small, or a passphrase that is not hex).
+    """
+    try:
+        raw_passphrase = bytes.fromhex(passphrase_hex)
+    except (TypeError, ValueError):
+        return None
+    if internal_key_hex:
+        # Some WeChat builds XOR the passphrase with a key embedded in
+        # Weixin.dll before deriving. Optional: the installs seen so far do
+        # not need it, so an empty value must stay the default.
+        try:
+            internal = bytes.fromhex(internal_key_hex)
+        except (TypeError, ValueError):
+            return None
+        if len(internal) == len(raw_passphrase):
+            raw_passphrase = bytes(a ^ b for a, b in zip(raw_passphrase, internal))
+
+    try:
+        with open(db_path, 'rb') as handle:
+            page = handle.read(4096)
+    except OSError:
+        return None
+    if len(page) < 4096 or page[:16] == b'\x00' * 16:
+        return None
+
+    salt = page[:16]
+    mac_salt = bytes(byte ^ 0x3a for byte in salt)
+    key = hashlib.pbkdf2_hmac('sha512', raw_passphrase, salt, 256000, 32)
+    mac_key = hashlib.pbkdf2_hmac('sha512', key, mac_salt, 2, 32)
+
+    reserve = 16 + 64                      # IV + HMAC-SHA512
+    reserve = ((reserve + 15) // 16) * 16  # rounded up to the AES block size
+    body_end = 4096 - reserve + 16
+    mac = hmac.new(mac_key, page[16:body_end], hashlib.sha512)
+    mac.update(struct.pack('<I', 1))       # page number
+    return hmac.compare_digest(mac.digest(), page[body_end:body_end + 64])
+
+
+def connect_message_shards_detailed(db_path, key_hex, salt_hex, passphrase=''):
+    """Open every shard and report what happened to each one.
+
+    Returns (pairs, failures):
+      pairs    [(shard_path, conn)] for the shards that opened
+      failures [{"name": basename, "reason": "KEY_REJECTED" | "OPEN_FAILED"}]
+
+    A shard that fails is still skipped rather than fatal - a partially
+    readable transcript beats a command that refuses to run at all - but the
+    failure stops being invisible. `shardsFailed` is the difference between
+    "read everything" and "read what happened to be reachable", which is the
+    whole point of the coverage report.
+
+    Only the basename is reported: the caller stores this in a state file, and
+    absolute paths must not end up there.
+    """
+    pairs = []
+    failures = []
+    for shard in discover_message_shards(db_path):
+        derived_key, derived_salt = derive_database_key(
+            shard, key_hex, salt_hex, passphrase)
+        # Try the derived key first; fall back to the configured pair so
+        # installs without a passphrase keep working exactly as before.
+        candidates = [(derived_key, derived_salt)]
+        if (derived_key, derived_salt) != (key_hex, salt_hex):
+            candidates.append((key_hex, salt_hex))
+        failure_reason = 'OPEN_FAILED'
+        for candidate_key, candidate_salt in candidates:
+            conn = None
+            try:
+                conn, _ = connect_nt_db(shard, candidate_key, candidate_salt)
+                # PRAGMA key alone never fails; only a read surfaces a bad key.
+                conn.execute('SELECT count(*) FROM sqlite_master').fetchone()
+            except Exception:
+                # Reached the database but could not decrypt it: the key is
+                # wrong, not the file. Close before retrying, or the handle
+                # leaks and on Windows the file stays locked.
+                if conn is not None:
+                    try:
+                        conn.close()
+                    except Exception:
+                        pass
+                failure_reason = 'KEY_REJECTED'
+                continue
+            pairs.append((shard, conn))
+            break
+        else:
+            failures.append({'name': os.path.basename(shard), 'reason': failure_reason})
+    return pairs, failures
+
+
+def connect_message_shards(db_path, key_hex, salt_hex, passphrase=''):
+    """Open the configured database plus every sibling shard.
+
+    Returns the connections that actually opened. A shard that fails is
+    skipped rather than fatal: a partially readable transcript beats a
+    command that refuses to run at all.
+
+    Behaviour-preserving delegate: callers that do not ask for the failure
+    detail get exactly what they got before.
+    """
+    pairs, _failures = connect_message_shards_detailed(
+        db_path, key_hex, salt_hex, passphrase)
+    return [conn for _path, conn in pairs]
+
+
+def count_talker_rows(conn, msg_table):
+    """Rows this conversation has in one shard, or None if it has no table there.
+
+    Distinguishes "this shard holds nothing for the conversation" from "this
+    shard could not be read" - without it both look like zero.
+    """
+    try:
+        cur = conn.execute(
+            "SELECT COUNT(*) FROM sqlite_master WHERE name=?", (msg_table,))
+        if cur.fetchone()[0] == 0:
+            return None
+        return conn.execute('SELECT COUNT(*) FROM "%s"' % msg_table).fetchone()[0]
+    except Exception:
+        return None
+
+
+def msg_tables(conn):
+    """Names of the per-conversation Msg_ tables in one shard."""
+    try:
+        rows = conn.execute(
+            "SELECT name FROM sqlite_master WHERE type='table' AND name LIKE 'Msg\\_%' ESCAPE '\\'"
+        ).fetchall()
+    except Exception:
+        return set()
+    return {row[0] for row in rows}
+
+
+def get_fav_schema(db_path, key_hex):
+    """Dump favorite.db schema + sample rows (key verification / exploration)."""
+    try:
+        with open(db_path, 'rb') as fh:
+            salt = fh.read(16).hex()
+        conn = require_sqlcipher().connect(db_path)
+        conn.execute(f'PRAGMA key = "x\'{key_hex}{salt}\'";')
+        tables = conn.execute(
+            "SELECT name FROM sqlite_master WHERE type='table' ORDER BY name"
+        ).fetchall()
+        result = {"success": True, "tables": [t[0] for t in tables]}
+        if not tables:
+            result["error"] = "数据库为空或密钥错误"
+            result["success"] = False
+        conn.close()
+        return result
+    except Exception as e:
+        return {"success": False, "error": str(e).split('\n')[0][:200]}
+
+
+FAV_TYPE_NAMES = {
+    1: 'text',        # 文字
+    2: 'image',       # 图片
+    4: 'video',       # 视频
+    5: 'article',     # 公众号文章/网页链接
+    14: 'chatrecord', # 聊天记录
+}
+
+
+def parse_fav_content(content):
+    """Extract display fields (title/link/desc/source/cover) from favitem XML."""
+    out = {}
+    if not content:
+        return out
+    import xml.etree.ElementTree as ET
+    try:
+        root = ET.fromstring(content)
+    except ET.ParseError:
+        return out
+
+    def text(path):
+        el = root.find(path)
+        if el is not None and el.text and el.text.strip():
+            return el.text.strip()
+        return None
+
+    title = (text('weburlitem/pagetitle')
+             or text('datalist/dataitem/datatitle')
+             or text('title')
+             or text('desc'))
+    if title:
+        out['title'] = title
+    link = (text('weburlitem/clean_url')
+            or text('source/link')
+            or text('datalist/dataitem/stream_weburl'))
+    if link:
+        out['link'] = link
+    desc = text('weburlitem/pagedesc') or text('datalist/dataitem/datadesc')
+    if desc and desc != out.get('title'):
+        out['desc'] = desc
+    src = text('weburlitem/appmsgshareitem/srcdisplayname')
+    if src:
+        out['source_name'] = src
+    cover = (text('weburlitem/pagethumb_url')
+             or text('datalist/dataitem/dataext')
+             or text('datalist/dataitem/cdn_thumburl'))
+    if cover:
+        out['cover'] = cover
+    fmt = text('datalist/dataitem/datafmt')
+    if fmt:
+        out['format'] = fmt
+    return out
+
+
+def get_favorites(db_path, key_hex, limit=100, offset=0, keyword=None, fav_type=None):
+    """List favorite items from favorite.db with parsed content."""
+    try:
+        with open(db_path, 'rb') as fh:
+            salt = fh.read(16).hex()
+        conn = require_sqlcipher().connect(db_path)
+        conn.execute(f'PRAGMA key = "x\'{key_hex}{salt}\'";')
+        c = conn.cursor()
+
+        tables = [t[0] for t in c.execute(
+            "SELECT name FROM sqlite_master WHERE type='table'").fetchall()]
+        if 'fav_db_item' not in tables:
+            conn.close()
+            return {"error": f"未找到 fav_db_item 表，现有表: {tables[:20]}"}
+
+        total = c.execute('SELECT count(*) FROM fav_db_item').fetchone()[0]
+
+        sql = ('SELECT local_id, server_id, type, update_time, fromusr, realchatname, content '
+               'FROM fav_db_item WHERE 1=1')
+        params = []
+        if fav_type is not None:
+            sql += ' AND type = ?'
+            params.append(fav_type)
+        if keyword:
+            sql += ' AND content LIKE ?'
+            params.append(f'%{keyword}%')
+        sql += ' ORDER BY update_time DESC LIMIT ? OFFSET ?'
+        params.extend([limit, offset])
+        rows = c.execute(sql, params).fetchall()
+        conn.close()
+
+        items = []
+        for local_id, server_id, ftype, update_time, fromusr, realchatname, content in rows:
+            ct = content if isinstance(content, str) else (
+                content.decode('utf-8', errors='replace') if content else '')
+            item = {
+                'local_id': local_id,
+                'server_id': server_id,
+                'type': ftype,
+                'type_name': FAV_TYPE_NAMES.get(ftype, 'type_%s' % ftype),
+                'update_time': update_time,
+                'from_user': fromusr,
+                'chat_name': realchatname or None,
+            }
+            item.update(parse_fav_content(ct))
+            items.append(item)
+        return {"favorites": items, "total": total, "count": len(items),
+                "limit": limit, "offset": offset}
+    except Exception as e:
+        return {"error": str(e).split('\n')[0][:200]}
+
+
+#: 非文本消息在**列表预览**里的显示形态。与 `sqlcipherCore.processMessage` 的口径一致。
+#: 列表预览只需要回答"这是什么"，不需要内容 —— 而 49 类的 `source` 往往是被压缩的
+#: appmsg，硬去剥 XML 标签只会得到 `(/`$` 这种碎片（2026-10-19 用户在 `sessions` 里看到）。
+_PREVIEW_LABELS = {
+    3: '[图片]', 34: '[语音]', 43: '[视频]', 47: '[表情]', 48: '[位置]',
+    42: '[名片]', 49: '[链接/文件]', 10000: '[系统消息]', 10002: '[系统消息]',
+}
+
+
+def _clean_preview(text, limit=50):
+    """预览用：清掉控制字符、压掉换行/连续空白、去掉行首的二进制残渣、截断。
+
+    行首清洗只针对 **ASCII 标点**：群消息的 `source` 有时前面带三四个字节的碎片
+    （实测一条广告消息长这样：`(/ }25984981651785937@openim:🔰周黑鸭…`），
+    不处理的话列表第一列就是这堆符号。中文标点与 emoji 开头**不动**——
+    `？好的`、`😂收到` 都是正常消息。
+    """
+    if not isinstance(text, str):
+        return ''
+    cleaned = ''.join(ch for ch in text if ch >= ' ' or ch == '\t')
+    cleaned = re.sub(r'\s+', ' ', cleaned).strip()
+    # 空格也要算在"垃圾前缀"里：实测前缀是 `(/ }`，中间那个空格会把
+    # 「连续 ASCII 标点」这条匹配打断，于是整段清理失效（2026-10-19 实测）。
+    cleaned = re.sub(r'^[\x20-\x2f\x3a-\x40\x5b-\x60\x7b-\x7e]{1,8}(?=[A-Za-z0-9\u4e00-\u9fff])', '', cleaned)
+    return cleaned[:limit]
+
+
+def _looks_broken(text):
+    """这段"文本"其实是被当成文本的二进制吗？
+
+    判据只看两样：UTF-8 解码时留下的替换字符、以及不可打印字符的占比。
+    实测样本：一条 openim 消息的 `message_content` 解出来是
+    `(/ }25984981651785937@openim:` 加一个替换字符 —— 直接显示就是垃圾。
+    """
+    if not text:
+        return False
+    # 出现**任何**替换字符就说明原始字节不是合法 UTF-8 —— 正常中文/emoji 不会产生它。
+    # 阈值不能是"两个以上"：实测那条 openim 消息只在结尾丢了一个字节，
+    # 前面全是可打印的压缩碎片，比例判据根本拦不住。
+    if '\ufffd' in text:
+        return True
+    printable = sum(1 for ch in text if ch.isprintable())
+    return printable / max(1, len(text)) < 0.9
+
+
+def _summarise(row):
+    """(last_time, summary) from the newest message row of a conversation."""
+    last_time = row[0] or 0
+    msg_type = row[3] or 0
+    source_text = row[1]
+    content_text = row[2]
+
+    if msg_type == 1:
+        # 文本：`message_content` 与 `source` 里**哪个更像人话就用哪个**。
+        # 只认 `message_content` 会踩到"这一列其实是二进制"的消息（见 `_looks_broken`）。
+        candidates = []
+        for value in (content_text, source_text):
+            if isinstance(value, bytes):
+                value = value.decode('utf-8', errors='ignore')
+            if isinstance(value, str) and value.strip():
+                candidates.append(value)
+        if not candidates:
+            return last_time, ""
+        best = min(candidates, key=lambda s: s.count('\ufffd') / max(1, len(s)))
+        if _looks_broken(best):
+            return last_time, '[内容无法显示]'
+        return last_time, _clean_preview(best)
+
+    # 非文本：先按类型给标签。**不再从 `source` 里剥标签抠文字** —— 那条路只有在
+    # XML 是明文 appmsg 时才成立，而 4.x 上它经常是压缩的，抠出来就是二进制碎片。
+    #
+    # `local_type` 在 4.x 里是**打包值**：高位放 subtype，低 32 位才是基础类型
+    # （实测 `25769803825` = 0x6_00000031，低 8 位 0x31 = 49 = 链接/文件）。
+    # 不 mask 的话，一张「链接卡片」会被报成 `[消息类型 25769803825]`。
+    raw_type = int(msg_type) if msg_type else 0
+    base_type = raw_type & 0xFFFFFFFF
+    label = _PREVIEW_LABELS.get(base_type)
+    if label:
+        return last_time, label
+
+    if isinstance(source_text, str) and source_text.strip():
+        clean = re.sub(r'<[^>]+>', '', source_text)
+        if clean.strip():
+            # 兜底也要过一遍"这其实是二进制吗"：某些消息两列都是压缩数据，
+            # 剥掉尖括号后看着像文字，实际是 `(/ }25984...` 这种碎片。
+            if _looks_broken(clean):
+                return last_time, '[内容无法显示]'
+            return last_time, _clean_preview(clean)
+    return last_time, ('[消息类型 %d]' % base_type) if base_type else ''
+
+
+def get_sessions(conns):
+    """Get chat sessions, newest message per conversation across every shard."""
+    sessions = {}
+
+    # NT format: each chat has its own Msg_<MD5> table
+    # The Name2Id table maps usernames to IDs (user_name, is_session)
+    for conn in conns:
+        c = conn.cursor()
+        try:
+            c.execute("SELECT user_name FROM Name2Id WHERE is_session = 1 LIMIT 500")
+            usernames = [row[0] for row in c.fetchall() if row[0]]
+        except Exception:
+            continue
+
+        tables = msg_tables(conn)
+        for username in usernames:
+            entry = sessions.setdefault(username, {"summary": "", "last_time": 0})
+            msg_table = f"Msg_{hashlib.md5(username.encode()).hexdigest()}"
+            if msg_table not in tables:
+                continue
+            try:
+                c.execute(f'SELECT create_time, source, message_content, local_type FROM "{msg_table}" ORDER BY create_time DESC LIMIT 1')
+                row = c.fetchone()
+                if row:
+                    last_time, summary = _summarise(row)
+                    # A conversation spans shards; only the newest speaks for it.
+                    if last_time >= entry["last_time"]:
+                        entry["last_time"] = last_time
+                        entry["summary"] = summary
+            except Exception:
+                continue
+
+    result = [{
+        "username": username,
+        "type": 1 if "@chatroom" in username else 0,
+        "unreadCount": 0,
+        "summary": entry["summary"],
+        "sortTimestamp": entry["last_time"],
+        "lastTimestamp": entry["last_time"],
+        "displayName": username,
+    } for username, entry in sessions.items()]
+
+    # Sort by timestamp descending
+    result.sort(key=lambda s: s.get("sortTimestamp", 0), reverse=True)
+    return {"sessions": result}
+
+
+def _strip_group_speaker(content, known_ids):
+    """Drop the `wxid_...:` prefix group rows carry in their content.
+
+    Only an id the shard's Name2Id actually knows is accepted, so a message
+    that merely starts with `note: ...` is left alone.
+    """
+    match = re.match(r'^([A-Za-z0-9_@.-]{5,64})\s*[:：]\s', str(content or ''))
+    if not match or match.group(1) not in known_ids:
+        return content
+    return content[match.end():]
+
+
+# The columns the reader actually consumes. The SELECT is built from these
+# only: a WeChat version that renames or drops any *other* column of the
+# message table cannot break a read of a column nobody looks at.
+MESSAGE_COLUMNS = ('local_id', 'server_id', 'local_type', 'real_sender_id',
+                   'create_time', 'message_content')
+
+# Any one of these is enough to give rows an identity and an order. With none
+# of them a row cannot become a message at all, and that is a schema mismatch
+# rather than an empty conversation.
+# 定义搬去了 `nt_common`：`ORDER BY` 那处（同一个文件下面）过去自己又写了一份
+# 字面量，导出器再写一份——顺序在这里就是行为（先按 create_time 排），
+# 三份各自为政等于谁改一处谁静默换一种排法。
+
+
+# ---- 非文本消息的显示形态 ---------------------------------------------------
+#
+# 这条路径此前**把非文本消息丢掉了**：`parsedContent` 只在 local_type == 1 时才有值，
+# 而 BLOB 内容不是 str，会在下面被置成空串。于是图片/表情/文件/引用/撤回/红包在下游
+# 一律表现为"空内容"——实测那 44 条里"空"的三条，分别是**一条撤回提示**和两个表情。
+#
+# 类型编码有结构：`local_type = apptype * 2**32 + 49`，49 表示 appmsg 家族。所以这里
+# 是**推导**而不是一张长表；只有少数几个固定类型需要单独列。
+APPMSG_SUBTYPE = 49
+NON_TEXT_LABELS = {
+    3: '图片', 34: '语音', 42: '名片', 43: '视频', 47: '表情', 48: '位置',
+}
+APPMSG_LABELS = {
+    4: '链接', 5: '链接', 6: '文件', 19: '聊天记录', 33: '小程序', 57: '引用', 63: '直播',
+    2000: '转账', 2001: '红包',
+}
+# 不认识的 apptype 回 `[应用消息]` 而不是猜成"链接"：标签是给下游读的，不能支撑不起也写。
+APPMSG_UNKNOWN_LABEL = '应用消息'
+
+# 引用消息（apptype 57）的正文与被引原文之间的分隔。用词而不是符号，因为下游会把它读成
+# 一句话；`|` 在别处会被当表格分隔。
+QUOTE_SEP = ' ｜ 引：'
+# 被引原文的截断长度。实测 37 条真实引用消息：中位 36 字、17 条超过 60 字、最长的
+# 12733 字（引用了一整篇公众号文章）。60 会切掉近一半，120 能留住大多数完整句子，
+# 又把那个极端值压住。
+QUOTE_CLIP = 120
+
+
+ZSTD_MAGIC = bytes([0x28, 0xB5, 0x2F, 0xFD])
+def _decode_content(raw):
+    """消息内容 → 文本。实现在 `nt_common.decode_content`（各脚本共用一份）。
+
+    保留这个名字是为了不动 `non_text_display` 等既有调用点；行为与公共函数一致：
+    str 原样、zstd frame 解压、其它 bytes 按 UTF-8 解码，解不出回空串。
+    """
+    from nt_common import decode_content
+    return decode_content(raw)
+
+
+def _xml_block(xml, tag):
+    """取一个元素的起始标签到它自己的结束标签之间的整段（含标签）。
+
+    只在区块内找 `title`/`des`：整份 payload 里还有 `<emotionpageshared><title>` 这类
+    同名标签，在全文里搜会匹配到那个去（自验时真的把一段 XML 当文本吐出来过）。
+    """
+    text = xml or ''
+    match = re.search(r'<%s(?:\s[^>]*)?>.*?</%s>' % (tag, tag), text, re.S)
+    return match.group(0) if match else ''
+
+
+def _xml_text(xml, tag):
+    """取一个标签的文本（含 CDATA）。取不到回空串——`<title />` 这种自闭合就是取不到。
+
+    起始标签里**不许出现 `/`**：否则 `<title />` 会被当成开标签，一路吃到后面某个
+    `</title>`，把中间整段 XML 当成文本返回（自验时真的吐出来过）。
+
+    实体在这**解掉**：显示形态是给人（和模型）读的，`a&amp;b` 该显示成 `a&b`。
+    TS 侧的 `appMsgFormat.tagText` 同样解，两边一致——否则同一条消息 3.x 与 4.x 读起来
+    不一样。
+    """
+    match = re.search(
+        r'<%s(?:\s[^>/]*)?>(?:<!\[CDATA\[)?(.*?)(?:\]\]>)?</%s>' % (tag, tag), xml or '', re.S)
+    return html.unescape(match.group(1).strip()) if match else ''
+
+
+def _clip(text, limit):
+    """超长截断并**留下省略号**。
+
+    不标记的截断读起来像一句说完的话——`title[:60]` 此前就是这样，被引原文最长的
+    12733 字，切完看着像"这就是全部"。
+    """
+    if len(text) <= limit:
+        return text
+    return text[:limit] + '…'
+
+
+def non_text_display(local_type, raw):
+    """非文本消息 → 给下游看的文本。**绝不返回空串**——这是这次修复的全部要点。
+
+    宁可写 `[未识别的消息类型 81604378673]`，也不要让下游看到空字符串：空串在下游与
+    "这条消息不存在"无法区分，而它的真实后果是助手回答"我没解析出内容"。
+    """
+    lt = int(local_type or 0)
+    text = _decode_content(raw)
+
+    if lt & 0xFFFFFFFF == APPMSG_SUBTYPE:
+        apptype = lt >> 32
+        label = APPMSG_LABELS.get(apptype, APPMSG_UNKNOWN_LABEL)
+        block = _xml_block(text, 'appmsg')
+        title = _xml_text(block, 'title') or _xml_text(block, 'des')
+        # 实测 payload 里 `<emotionpageshared>` 这类**嵌套容器**也带 `<title>`，取到的是
+        # 它们自己的占位值（`null`）。这种不是标题，按"没有标题"处理。
+        if title.lower() in ('null', 'undefined', '0'):
+            title = ''
+        # 截断：解析万一走偏，也不许把一段 XML 当成标题交给下游。
+        title = _clip(title, 60)
+        # 引用消息（apptype 57）：`title` 是**回复正文**，被引用的原文在 `refermsg/content`。
+        # 实测 37 条真实引用消息，被引原文中位 36 字。此前只留回复，模型看到的是一句
+        # "是呀，够得意个"却不知道在回什么——引用不带原文等于没引用。
+        if apptype == 57:
+            quoted = _xml_text(_xml_block(text, 'refermsg'), 'content')
+            if quoted.lower() in ('null', 'undefined', '0'):
+                quoted = ''
+            quoted = _clip(quoted, QUOTE_CLIP)
+            if quoted:
+                body = (title + QUOTE_SEP + quoted) if title else quoted
+                return '[%s] %s' % (label, body)
+        return '[%s] %s' % (label, title) if title else '[%s]' % label
+
+    label = NON_TEXT_LABELS.get(lt)
+    if label:
+        return '[%s]' % label
+    if lt == 10000:                       # 系统消息：撤回、拍一拍、入群提示
+        return _xml_text(text, 'content') or '[系统消息]'
+    return '[未识别的消息类型 %d]' % lt
+
+
+def _message_dict(row, sender_id_map, name_map, own_wxid, is_group=False):
+    """One message row -> the CLI's message shape.
+
+    `row` is a column-name keyed dict, not a tuple. Positional access assumes
+    every column exists; when a WeChat version drops one, the whole shard read
+    fails and the conversation looks empty. Reading by name means a missing
+    column costs that one field instead.
+    """
+    local_type = row.get('local_type') or 0
+    create_time = row.get('create_time') or 0
+    real_sender_id = row.get('real_sender_id') or 0
+
+    # Resolve sender: real_sender_id -> Name2Id -> user_name
+    sender_username = sender_id_map.get(real_sender_id, "")
+
+    # Determine if message is from self
+    # own_wxid may have _xxxx suffix (from xwechat_files dir), try both
+    is_self = bool(own_wxid and (
+        sender_username == own_wxid or
+        (own_wxid.endswith('_') is False and sender_username.startswith(own_wxid))
+    ))
+    if not is_self and own_wxid:
+        # Strip _xxxx suffix and retry
+        parts = own_wxid.rsplit('_', 1)
+        if len(parts) == 2 and len(parts[1]) == 4 and parts[1].isalnum():
+            is_self = (sender_username == parts[0])
+
+    # Resolve sender display name from contact map
+    if is_self:
+        sender_display = ""  # Let the CLI show "我"
+    else:
+        sender_display = name_map.get(sender_username, sender_username) if sender_username else sender_username
+
+    # Parse message_content - TEXT column
+    raw_content = row.get('message_content')
+    # **普通文本也可能是 zstd 压缩的 BLOB**（issue #12）。这里原来只认 `str`，
+    # 于是 `local_type == 1` 的压缩正文被置空、再交给 `non_text_display()`，
+    # 最终输出 `[未识别的消息类型 1]` —— 消息还在，正文没了，下游按
+    # `parsedContent` 做的关键词 / @所有人 筛选就漏掉这些通知。
+    # `_decode_content()` 已经处理三种形态（str / zstd frame / 普通 bytes），复用它即可。
+    if local_type == 1:
+        content = _decode_content(raw_content)
+    else:
+        content = raw_content if isinstance(raw_content, str) else ""
+
+    # `content`/`rawContent` stay exactly as stored; only `parsedContent` - the
+    # field every consumer reads first - gets the display-ready form.
+    #
+    # 非文本消息（图片/表情/文件/引用/撤回/红包…）在过去得到的是**空串**：它们的内容是
+    # JSON 里不是 str 的 BLOB，上面那行会把它置空。现在它们走 `non_text_display`，
+    # 至少拿到 `[图片]`，多数还能带上文件名、被引用的原文或"谁撤回了一条消息"。
+    if content:
+        display = _strip_group_speaker(content, set(sender_id_map.values())) if is_group else content
+    else:
+        display = non_text_display(local_type, raw_content)
+
+    return {
+        "localId": row.get('local_id') or 0,
+        "serverId": str(row.get('server_id') or ''),
+        "localType": local_type,
+        "createTime": create_time,
+        "isSend": 1 if is_self else 0,  # 1 = I sent this
+        "senderUsername": sender_username,
+        "senderDisplay": sender_display,
+        "content": content,
+        "rawContent": content,
+        # 文本不截断：末尾的 `@所有人` 一类关键词正是下游筛选要用的，截掉等于静默漏消息
+        # （issue #12）。非文本仍保留 200 字的展示上限——它们是标签和标题，不是正文。
+        "parsedContent": display if local_type == 1 else display[:200],
+    }
+
+
+def get_messages(conns, talker, limit=100, offset=0, name_map=None, own_wxid=None,
+                 shard_names=None, shard_report=None, from_time=None, to_time=None):
+    """Get messages for a specific talker, merged across every shard.
+
+    Args:
+        name_map: optional {wxid: display_name} dict for resolving sender names
+        own_wxid: account owner wxid for self-message detection
+        shard_names: optional basename per connection, parallel to `conns`
+        shard_report: optional list to append one outcome dict per shard to.
+            Purely additive: the returned messages are identical with or
+            without it, which is asserted by test/nt_decrypt_shards_test.py.
+        from_time: optional inclusive lower bound on `create_time` (unix
+            seconds), pushed into SQL so a bounded read does not have to fetch
+            and discard the whole conversation.
+        to_time: optional inclusive upper bound, same units.
+
+    A requested window is only ever honoured or refused, never approximated: a
+    shard whose table has no `create_time` is skipped and reported as
+    WINDOW_UNAVAILABLE rather than returning out-of-range rows that a caller
+    tracking coverage would count as covered. Every in-tree caller that passes
+    a window also passes `shard_report`, so that refusal is always visible.
+    """
+    if name_map is None:
+        name_map = {}
+
+    window_requested = from_time is not None or to_time is not None
+
+    msg_table = f"Msg_{hashlib.md5(talker.encode()).hexdigest()}"
+    is_group = '@chatroom' in talker
+
+    def record(index, opened, has_table, rows_for_talker, reason, missing=None):
+        if shard_report is None:
+            return
+        name = shard_names[index] if shard_names and index < len(shard_names) else ''
+        shard_report.append({
+            'name': name,
+            'opened': opened,
+            'hasTalkerTable': has_table,
+            'rowsForTalker': rows_for_talker,
+            'reason': reason,
+            # Columns this shard does not have. A non-empty list together with
+            # a null `reason` means the read succeeded but lost fields, which is
+            # a different thing from a failed read and has to stay visible.
+            'missingColumns': list(missing or ()),
+        })
+
+    # Each shard only needs to yield its newest window: once every shard's rows
+    # are merged and re-sorted, nothing older than that can reach this page.
+    window = 0 if limit <= 0 else limit + offset
+    collected = []
+    found = False
+
+    for index, conn in enumerate(conns):
+        c = conn.cursor()
+        try:
+            c.execute("SELECT COUNT(*) FROM sqlite_master WHERE name=?", (msg_table,))
+            if c.fetchone()[0] == 0:
+                # The shard is readable and simply holds nothing for this
+                # conversation - distinct from not being readable at all.
+                record(index, True, False, None, None)
+                continue
+            found = True
+            rows_for_talker = c.execute(
+                'SELECT COUNT(*) FROM "%s"' % msg_table).fetchone()[0]
+
+            available = table_columns(c, msg_table)
+            selected = [col for col in MESSAGE_COLUMNS if col in available]
+            missing = [col for col in MESSAGE_COLUMNS if col not in available]
+            if not [col for col in MESSAGE_ANCHOR_COLUMNS if col in available]:
+                # The row shape no longer matches anything the reader can name.
+                record(index, True, True, rows_for_talker, 'SCHEMA_MISMATCH',
+                       missing=missing)
+                continue
+            if window_requested and 'create_time' not in available:
+                record(index, True, True, rows_for_talker, 'WINDOW_UNAVAILABLE',
+                       missing=missing)
+                continue
+
+            order = [col for col in MESSAGE_ANCHOR_COLUMNS
+                     if col in available]
+            sql = 'SELECT %s FROM "%s"' % (
+                ', '.join('"%s"' % col for col in selected), msg_table)
+            params = []
+            clauses = []
+            if from_time is not None:
+                clauses.append('"create_time" >= ?')
+                params.append(from_time)
+            if to_time is not None:
+                clauses.append('"create_time" <= ?')
+                params.append(to_time)
+            if clauses:
+                sql += ' WHERE ' + ' AND '.join(clauses)
+            sql += ' ORDER BY ' + ', '.join('"%s" DESC' % col for col in order)
+            if window:
+                sql += ' LIMIT ?'
+                params.append(window)
+            c.execute(sql, params)
+            # Dicts rather than driver rows: the SELECT list varies per shard,
+            # so position no longer identifies a column. Built here rather than
+            # via row_factory, which every other reader sharing these
+            # connections would inherit.
+            rows = [dict(zip(selected, values)) for values in c.fetchall()]
+
+            # Sender ids are rowids, so the map has to come from the same
+            # shard. Absent, it costs sender names and nothing else, so it is
+            # probed rather than left to raise and lose the whole shard.
+            has_name_table = c.execute(
+                "SELECT COUNT(*) FROM sqlite_master WHERE name='Name2Id'").fetchone()[0] > 0
+            sender_id_map = {}
+            if has_name_table:
+                c.execute("SELECT rowid, user_name FROM Name2Id")
+                sender_id_map = {rowid: uname for rowid, uname in c.fetchall()}
+        except Exception:
+            # This shard had the conversation's table but could not be read.
+            # Previously this was indistinguishable from "no rows here".
+            record(index, True, None, None, 'READ_FAILED')
+            continue
+        record(index, True, True, rows_for_talker, None, missing=missing)
+
+        for row in rows:
+            collected.append(_message_dict(row, sender_id_map, name_map, own_wxid, is_group))
+
+    if not found:
+        return {"error": f"未找到会话: {talker}"}
+
+    collected.sort(key=lambda m: (m["createTime"], m["localId"]), reverse=True)
+    if limit > 0:
+        collected = collected[offset:offset + limit]
+    return {"messages": collected}
+
+
+def cursor_probe(db_path, key_hex, salt_hex, passphrase='',
+                 tables_per_shard=5, rows_per_table=500):
+    """Measure what a stable incremental cursor would need on this database.
+
+    D-027 keeps `stableCursor: false` until every message backend can honour
+    the same semantics. This probe answers the NT backend's **data-side** half:
+    whether every message table carries the anchor columns, and whether sampled
+    rows actually hold a nonzero `server_id` (the reader's identity falls back
+    to localId+createTime without one, and `local_id` is per-shard, so those
+    rows cannot be resumed across shards). Duplicate server ids are collected
+    across the sample too - an id that appears in two places cannot key a
+    cursor, whatever the column definition says.
+
+    Sampling is bounded and says so (`exhaustive: false`): `tables_per_shard`
+    tables per shard (sorted names, deterministic) read at both rowid ends,
+    `rows_per_table` rows each. A full scan of every shard would cost minutes
+    on a real account; the schema check IS exhaustive - `PRAGMA table_info` is
+    cheap, so every Msg_* table's anchor columns are checked, not sampled.
+
+    Read-only: no write, no network, nothing leaves stdout.
+    """
+    pairs, failures = connect_message_shards_detailed(
+        db_path, key_hex, salt_hex, passphrase)
+    shard_items = []
+    totals = {'tables': 0, 'tablesSampled': 0, 'rowsSampled': 0,
+              'serverIdMissing': 0}
+    anchor_missing_total = 0
+    # server_id(str) -> {(shard, table)} it was sampled in. A set per id keeps
+    # the two rowid ends of one small table from double-counting the same row
+    # as a "duplicate" - both ends see it, but it is one location.
+    id_locations = {}
+
+    for failure in failures:
+        shard_items.append({
+            'name': failure['name'], 'opened': False, 'reason': failure['reason'],
+            'tables': None,
+        })
+
+    for path, conn in pairs:
+        item = {'name': os.path.basename(path), 'opened': True, 'reason': None}
+        try:
+            tables = sorted(msg_tables(conn))
+            item['tables'] = len(tables)
+            totals['tables'] += len(tables)
+
+            # Schema half: exhaustive over every message table.
+            anchor_missing = []
+            for table in tables:
+                cols = table_columns(conn.cursor(), table)
+                missing = [c for c in MESSAGE_ANCHOR_COLUMNS if c not in cols]
+                if missing and missing not in anchor_missing:
+                    anchor_missing.append(missing)
+            anchor_missing_total += len(anchor_missing)
+            item['anchorColumnsMissing'] = anchor_missing
+
+            # Data half: bounded sample, both rowid ends per table. A table
+            # shorter than the bound hands the same rows to both ends - they
+            # are one sample, not two, so rowids are deduplicated per table
+            # (and the same id read at both ends must not look "duplicated").
+            sample_targets = tables[:tables_per_shard]
+            item['tablesSampled'] = len(sample_targets)
+            totals['tablesSampled'] += len(sample_targets)
+            sampled = 0
+            missing_ids = 0
+            for table in sample_targets:
+                c = conn.cursor()
+                seen_rowids = set()
+                table_ids = set()
+                for order in ('ASC', 'DESC'):
+                    c.execute(
+                        'SELECT rowid, server_id FROM "%s" ORDER BY rowid %s LIMIT ?'
+                        % (table, order), (rows_per_table,))
+                    for rowid, server_id in c.fetchall():
+                        if rowid in seen_rowids:
+                            continue
+                        seen_rowids.add(rowid)
+                        sampled += 1
+                        if server_id is None or server_id == 0:
+                            missing_ids += 1
+                        else:
+                            table_ids.add(str(int(server_id)))
+                for sid in table_ids:
+                    id_locations.setdefault(sid, set()).add((item['name'], table))
+            item['rowsSampled'] = sampled
+            item['serverIdMissing'] = missing_ids
+            totals['rowsSampled'] += sampled
+            totals['serverIdMissing'] += missing_ids
+        except Exception as exc:
+            item['reason'] = 'READ_FAILED: %s' % str(exc).split('\n')[0][:120]
+        finally:
+            shard_items.append(item)
+            conn.close()
+
+    duplicates = {sid: locs for sid, locs in id_locations.items() if len(locs) > 1}
+    return {
+        'shards': shard_items,
+        'schemaAnchorMissingTables': anchor_missing_total,
+        'sample': {
+            'tablesPerShard': tables_per_shard,
+            'rowsPerTable': rows_per_table,
+            'exhaustive': False,
+            'tablesTotal': totals['tables'],
+            'tablesSampled': totals['tablesSampled'],
+            'rowsSampled': totals['rowsSampled'],
+            'serverIdMissing': totals['serverIdMissing'],
+            'duplicateServerIds': len(duplicates),
+            'duplicateServerIdSamples': [
+                {'serverId': sid, 'locations': sorted('%s/%s' % loc for loc in locs)}
+                for sid, locs in sorted(duplicates.items())[:20]
+            ],
+        },
+    }
+
+
+def get_contacts(conns, limit=200, keyword=None):
+    """Get contacts from NT database, merged across every shard.
+
+    **给了 `keyword` 就不加 `LIMIT`**，由调用方在名字解析之后过滤、再按 `limit` 截断。
+    原来无条件 `LIMIT`，于是过滤发生在**截断之后**：关键字只能在前 `limit` 行里找，
+    后面的人搜不到（实测 500 人的通讯录、按备注名找 10 个只命中 3 个）。名字解析要用
+    contact.db 的备注/昵称，那是调用方才有的东西，所以截断必须挪到过滤之后。
+    """
+    seen = set()
+    contacts = []
+    for conn in conns:
+        try:
+            sql = "SELECT user_name FROM Name2Id" if keyword else "SELECT user_name FROM Name2Id LIMIT ?"
+            rows = conn.execute(sql, () if keyword else (limit,)).fetchall()
+        except Exception:
+            continue
+        for (username,) in rows:
+            # Name2Id carries a placeholder row with no user_name; rendering it
+            # produced a blank line at the top of every contact list.
+            if not username or username in seen:
+                continue
+            seen.add(username)
+            contacts.append({"username": username, "displayName": username})
+    return {"contacts": contacts}
+
+
+# ========== SNS (朋友圈) Queries ==========
+
+def parse_sns_content(content_str):
+    """Parse SNS content XML/Protobuf text to extract title, description, media etc."""
+    result = {
+        'content': '',
+        'create_time': 0,
+        'username': '',
+        'object_id': '',
+        'media_count': 0,
+    }
+    if not content_str:
+        return result
+
+    import re
+
+    # Extract createTime
+    m = re.search(r'<createTime>(\d+)</createTime>', content_str)
+    if m:
+        result['create_time'] = int(m.group(1))
+
+    # Extract username
+    m = re.search(r'<username>([^<]+)</username>', content_str)
+    if m:
+        result['username'] = m.group(1)
+
+    # Extract id
+    m = re.search(r'<id>(\d+)</id>', content_str)
+    if m:
+        result['object_id'] = m.group(1)
+
+    # Extract contentDesc (main text)
+    m = re.search(r'<contentDesc>([^<]*)</contentDesc>', content_str)
+    if m:
+        result['content'] = m.group(1)
+
+    # Extract contentDesc CDATA
+    m = re.search(r'<contentDesc>\s*<!\[CDATA\[(.*?)\]\]>\s*</contentDesc>', content_str, re.DOTALL)
+    if m:
+        result['content'] = m.group(1).strip()
+
+    # Extract title if present
+    m = re.search(r'<title>([^<]*)</title>', content_str)
+    if m:
+        title = m.group(1)
+        if title and not result['content']:
+            result['content'] = title
+        elif title:
+            result['content'] = title + '\n' + result['content']
+
+    # Count media (ContentObject tags)
+    result['media_count'] = len(re.findall(r'<ContentObject[ >]', content_str))
+
+    return result
+
+
+def get_sns_timeline(cursor, limit=20, offset=0, usernames=None, keyword=None,
+                     start_time=None, end_time=None):
+    """Query SNS timeline posts from SnsTimeLine table."""
+    # Check if SnsTimeLine exists, fall back to SnsTopItem_1
+    tables = [r[0] for r in cursor.execute(
+        "SELECT name FROM sqlite_master WHERE type='table';").fetchall()]
+
+    table = 'SnsTimeLine' if 'SnsTimeLine' in tables else None
+    if not table:
+        table = 'SnsTopItem_1' if 'SnsTopItem_1' in tables else None
+    if not table:
+        return {'success': False, 'error': 'No SNS table found'}
+
+    conditions = []
+    params = []
+
+    if table == 'SnsTimeLine':
+        if usernames:
+            placeholders = ','.join(['?' for _ in usernames])
+            conditions.append(f'user_name IN ({placeholders})')
+            params.extend(usernames)
+        if keyword:
+            conditions.append('content LIKE ?')
+            params.append(f'%{keyword}%')
+        if start_time:
+            conditions.append("CAST(substr(content, instr(content, '<createTime>') + 12, 10) AS INTEGER) >= ?")
+            params.append(start_time)
+        if end_time:
+            conditions.append("CAST(substr(content, instr(content, '<createTime>') + 12, 10) AS INTEGER) <= ?")
+            params.append(end_time)
+    elif table == 'SnsTopItem_1':
+        uname_col = 'username' if 'username' in [r[1] for r in cursor.execute(f'PRAGMA table_info([{table}]);').fetchall()] else 'user_name'
+        if usernames:
+            placeholders = ','.join(['?' for _ in usernames])
+            conditions.append(f'{uname_col} IN ({placeholders})')
+            params.extend(usernames)
+        if keyword:
+            conditions.append('summary LIKE ?')
+            params.append(f'%{keyword}%')
+        if start_time:
+            conditions.append('create_time >= ?')
+            params.append(start_time)
+        if end_time:
+            conditions.append('create_time <= ?')
+            params.append(end_time)
+
+    where = ' AND '.join(conditions) if conditions else '1=1'
+    order = 'tid DESC' if table == 'SnsTimeLine' else 'create_time DESC'
+
+    try:
+        cols = [r[1] for r in cursor.execute(f'PRAGMA table_info([{table}]);').fetchall()]
+        # Only select known text columns to avoid binary decode errors
+        text_cols = [c for c in cols if c in ('tid', 'user_name', 'content', 'username', 'summary',
+                                               'create_time', 'last_read_time', 'is_read',
+                                               'from_username', 'from_nickname', 'to_username',
+                                               'to_nickname', 'comment_id', 'feed_id',
+                                               'createTime', 'userName')]
+        if not text_cols:
+            text_cols = ['*']
+        select_str = ', '.join(text_cols)
+        rows = cursor.execute(
+            f'SELECT {select_str} FROM [{table}] WHERE {where} ORDER BY {order} LIMIT ? OFFSET ?',
+            params + [limit, offset]
+        ).fetchall()
+
+        timeline = []
+
+        for row in rows:
+            item = dict(zip(text_cols, row))
+
+            if table == 'SnsTimeLine':
+                # Parse XML content
+                content_str = item.get('content', '') or ''
+                parsed = parse_sns_content(content_str)
+                create_time = parsed['create_time'] or 0
+                username = parsed['username'] or item.get('user_name', '')
+                text = parsed['content'] or ''
+                media_count = parsed['media_count']
+            else:
+                create_time = item.get('create_time', 0) or 0
+                username = item.get('username', '') or item.get('user_name', '')
+                text = item.get('summary', '') or ''
+                media_count = 0
+
+            timeline.append({
+                'create_time': create_time,
+                'username': username,
+                'content': text,
+                'media_count': media_count,
+                'table': table,
+            })
+
+        return {'success': True, 'timeline': timeline, 'table': table}
+    except Exception as e:
+        return {'success': False, 'error': str(e)}
+
+
+def get_sns_usernames(cursor):
+    """Get unique usernames from SnsTopItem_1."""
+    tables = [r[0] for r in cursor.execute(
+        "SELECT name FROM sqlite_master WHERE type='table';").fetchall()]
+
+    # Prefer SnsTopItem_1 for user listing (larger)
+    table = 'SnsTopItem_1' if 'SnsTopItem_1' in tables else ('SnsTimeLine' if 'SnsTimeLine' in tables else None)
+    if not table:
+        return {'success': False, 'error': 'No SNS table found'}
+
+    # Discover username column
+    cols = [r[1] for r in cursor.execute(f'PRAGMA table_info([{table}]);').fetchall()]
+    uname_col = 'username' if 'username' in cols else 'user_name'
+
+    try:
+        rows = cursor.execute(
+            f'SELECT [{uname_col}], COUNT(*) as cnt FROM [{table}] GROUP BY [{uname_col}] ORDER BY cnt DESC'
+        ).fetchall()
+        usernames = [r[0] for r in rows if r[0]]
+        counts = {r[0]: r[1] for r in rows if r[0]}
+        return {'success': True, 'usernames': usernames, 'counts': counts}
+    except Exception as e:
+        return {'success': False, 'error': str(e)}
+
+
+def get_sns_stats(cursor, my_wxid=None):
+    """Get SNS statistics from SnsTopItem_1."""
+    tables = [r[0] for r in cursor.execute(
+        "SELECT name FROM sqlite_master WHERE type='table';").fetchall()]
+
+    table = 'SnsTopItem_1' if 'SnsTopItem_1' in tables else ('SnsTimeLine' if 'SnsTimeLine' in tables else None)
+    if not table:
+        return {'success': False, 'error': 'No SNS table found'}
+
+    try:
+        total = cursor.execute(f'SELECT COUNT(*) FROM [{table}]').fetchone()[0]
+
+        cols = [r[1] for r in cursor.execute(f'PRAGMA table_info([{table}]);').fetchall()]
+        uname_col = 'username' if 'username' in cols else 'user_name'
+
+        total_friends = 0
+        if uname_col:
+            total_friends = cursor.execute(
+                f'SELECT COUNT(DISTINCT [{uname_col}]) FROM [{table}] WHERE [{uname_col}] IS NOT NULL'
+            ).fetchone()[0]
+
+        my_posts = None
+        if my_wxid and uname_col:
+            my_posts = cursor.execute(
+                f'SELECT COUNT(*) FROM [{table}] WHERE [{uname_col}] = ?',
+                (my_wxid,)).fetchone()[0]
+
+        return {
+            'success': True,
+            'data': {
+                'totalPosts': total,
+                'totalFriends': total_friends,
+                'myPosts': my_posts,
+            }
+        }
+    except Exception as e:
+        return {'success': False, 'error': str(e)}
+
+
+# ========== Main CLI ==========
+
+def main():
+    import argparse
+    parser = argparse.ArgumentParser(description='WeChat NT Database Tool')
+    sub = parser.add_subparsers(dest='command')
+
+    # scan command
+    scan_parser = sub.add_parser('scan', help='Scan memory for keys and match NT databases')
+    scan_parser.add_argument('--json', action='store_true', help='Output as JSON')
+    scan_parser.add_argument('--root', default=os.environ.get('WEFLOW_SCAN_ROOT'), help='xwechat_files root directory override')
+
+    # sessions command
+    sessions_parser = sub.add_parser('sessions', help='List chat sessions')
+    sessions_parser.add_argument('--db', default=os.environ.get('WEFLOW_DB_PATH'), required=not os.environ.get('WEFLOW_DB_PATH'), help='Path to NT database')
+    sessions_parser.add_argument('--key', default=os.environ.get('WEFLOW_NT_KEY'), required=not os.environ.get('WEFLOW_NT_KEY'), help='Key hex (64 chars)')
+    sessions_parser.add_argument('--salt', default=os.environ.get('WEFLOW_NT_SALT'), required=not os.environ.get('WEFLOW_NT_SALT'), help='Salt hex (32 chars)')
+    sessions_parser.add_argument('--passphrase', default=os.environ.get('WEFLOW_NT_PASSPHRASE'), help='Shared passphrase for deriving per-shard keys')
+    sessions_parser.add_argument('--keyword', default=os.environ.get('WEFLOW_QUERY_KEYWORD'), help='Filter by keyword')
+    sessions_parser.add_argument('--contact-db', default=os.environ.get('WEFLOW_CONTACT_DB'), help='Path to contact.db for display names')
+    sessions_parser.add_argument('--contact-key', default=os.environ.get('WEFLOW_CONTACT_KEY'), help='Contact DB key hex (64 chars)')
+    sessions_parser.add_argument('--contact-salt', default=os.environ.get('WEFLOW_CONTACT_SALT'), help='Contact DB salt hex (32 chars)')
+
+    # messages command
+    msg_parser = sub.add_parser('messages', help='Get messages')
+    msg_parser.add_argument('--db', default=os.environ.get('WEFLOW_DB_PATH'), required=not os.environ.get('WEFLOW_DB_PATH'), help='Path to NT database')
+    msg_parser.add_argument('--key', default=os.environ.get('WEFLOW_NT_KEY'), required=not os.environ.get('WEFLOW_NT_KEY'), help='Key hex (64 chars)')
+    msg_parser.add_argument('--salt', default=os.environ.get('WEFLOW_NT_SALT'), required=not os.environ.get('WEFLOW_NT_SALT'), help='Salt hex (32 chars)')
+    msg_parser.add_argument('--passphrase', default=os.environ.get('WEFLOW_NT_PASSPHRASE'), help='Shared passphrase for deriving per-shard keys')
+    msg_parser.add_argument('--talker', default=os.environ.get('WEFLOW_TALKER'), required=not os.environ.get('WEFLOW_TALKER'), help='Talker username')
+    msg_parser.add_argument('--limit', type=int, default=100)
+    msg_parser.add_argument('--offset', type=int, default=0)
+    # Unix seconds, inclusive at both ends. Pushed into SQL, so a bounded read
+    # no longer has to fetch the whole conversation to discard most of it.
+    msg_parser.add_argument('--from', dest='from_time', type=int,
+                            default=os.environ.get('WEFLOW_FROM_TIME'),
+                            help='Only messages at or after this unix time')
+    msg_parser.add_argument('--to', dest='to_time', type=int,
+                            default=os.environ.get('WEFLOW_TO_TIME'),
+                            help='Only messages at or before this unix time')
+    msg_parser.add_argument('--contact-db', default=os.environ.get('WEFLOW_CONTACT_DB'), help='Path to contact.db for sender names')
+    msg_parser.add_argument('--contact-key', default=os.environ.get('WEFLOW_CONTACT_KEY'), help='Contact DB key hex (64 chars)')
+    msg_parser.add_argument('--contact-salt', default=os.environ.get('WEFLOW_CONTACT_SALT'), help='Contact DB salt hex (32 chars)')
+    msg_parser.add_argument('--own-wxid', default=os.environ.get('WEFLOW_OWN_WXID'), help='Account owner wxid (for self-message detection)')
+    # Opt-in: without it the JSON is byte-identical to before this flag existed.
+    msg_parser.add_argument('--report-shards', action='store_true',
+                            help='Add a per-shard read report to the result (additive)')
+
+    # contacts command
+    contacts_parser = sub.add_parser('contacts', help='List contacts')
+    contacts_parser.add_argument('--db', default=os.environ.get('WEFLOW_DB_PATH'), required=not os.environ.get('WEFLOW_DB_PATH'), help='Path to NT database')
+    contacts_parser.add_argument('--key', default=os.environ.get('WEFLOW_NT_KEY'), required=not os.environ.get('WEFLOW_NT_KEY'), help='Key hex (64 chars)')
+    contacts_parser.add_argument('--salt', default=os.environ.get('WEFLOW_NT_SALT'), required=not os.environ.get('WEFLOW_NT_SALT'), help='Salt hex (32 chars)')
+    contacts_parser.add_argument('--passphrase', default=os.environ.get('WEFLOW_NT_PASSPHRASE'), help='Shared passphrase for deriving per-shard keys')
+    contacts_parser.add_argument('--keyword', default=os.environ.get('WEFLOW_QUERY_KEYWORD'), help='Filter by keyword')
+    contacts_parser.add_argument('--limit', type=int, default=200)
+    contacts_parser.add_argument('--contact-db', default=os.environ.get('WEFLOW_CONTACT_DB'), help='Path to contact.db for display names')
+    contacts_parser.add_argument('--contact-key', default=os.environ.get('WEFLOW_CONTACT_KEY'), help='Contact DB key hex (64 chars)')
+    contacts_parser.add_argument('--contact-salt', default=os.environ.get('WEFLOW_CONTACT_SALT'), help='Contact DB salt hex (32 chars)')
+
+    # sns-timeline command
+    sns_tl_parser = sub.add_parser('sns-timeline', help='Get SNS/Moments timeline')
+    sns_tl_parser.add_argument('--db', default=os.environ.get('WEFLOW_DB_PATH'), required=not os.environ.get('WEFLOW_DB_PATH'), help='Path to sns.db')
+    sns_tl_parser.add_argument('--key', default=os.environ.get('WEFLOW_NT_KEY'), required=not os.environ.get('WEFLOW_NT_KEY'), help='Key hex (64 chars)')
+    sns_tl_parser.add_argument('--salt', default=os.environ.get('WEFLOW_NT_SALT'), required=not os.environ.get('WEFLOW_NT_SALT'), help='Salt hex (32 chars)')
+    sns_tl_parser.add_argument('--limit', type=int, default=20)
+    sns_tl_parser.add_argument('--offset', type=int, default=0)
+    sns_tl_parser.add_argument('--usernames', default=os.environ.get('WEFLOW_QUERY_USERNAMES'), help='JSON array of usernames to filter')
+    sns_tl_parser.add_argument('--keyword', default=os.environ.get('WEFLOW_QUERY_KEYWORD'), help='Search keyword')
+    sns_tl_parser.add_argument('--start-time', type=int, help='Start timestamp')
+    sns_tl_parser.add_argument('--end-time', type=int, help='End timestamp')
+
+    # sns-usernames command
+    sns_un_parser = sub.add_parser('sns-usernames', help='List usernames with SNS posts')
+    sns_un_parser.add_argument('--db', default=os.environ.get('WEFLOW_DB_PATH'), required=not os.environ.get('WEFLOW_DB_PATH'), help='Path to sns.db')
+    sns_un_parser.add_argument('--key', default=os.environ.get('WEFLOW_NT_KEY'), required=not os.environ.get('WEFLOW_NT_KEY'), help='Key hex (64 chars)')
+    sns_un_parser.add_argument('--salt', default=os.environ.get('WEFLOW_NT_SALT'), required=not os.environ.get('WEFLOW_NT_SALT'), help='Salt hex (32 chars)')
+
+    # sns-stats command
+    sns_stats_parser = sub.add_parser('sns-stats', help='SNS statistics')
+    sns_stats_parser.add_argument('--db', default=os.environ.get('WEFLOW_DB_PATH'), required=not os.environ.get('WEFLOW_DB_PATH'), help='Path to sns.db')
+    sns_stats_parser.add_argument('--key', default=os.environ.get('WEFLOW_NT_KEY'), required=not os.environ.get('WEFLOW_NT_KEY'), help='Key hex (64 chars)')
+    sns_stats_parser.add_argument('--salt', default=os.environ.get('WEFLOW_NT_SALT'), required=not os.environ.get('WEFLOW_NT_SALT'), help='Salt hex (32 chars)')
+    sns_stats_parser.add_argument('--my-wxid', default=os.environ.get('WEFLOW_OWN_WXID'), help='Account owner wxid for my-posts count')
+
+    # fav-schema command
+    fav_schema_parser = sub.add_parser('fav-schema', help='Dump favorite.db schema (key verification)')
+    fav_schema_parser.add_argument('--db', default=os.environ.get('WEFLOW_DB_PATH'), required=not os.environ.get('WEFLOW_DB_PATH'), help='Path to favorite.db')
+    fav_schema_parser.add_argument('--key', default=os.environ.get('WEFLOW_NT_KEY'), required=not os.environ.get('WEFLOW_NT_KEY'), help='Key hex (64 chars)')
+
+    verify_parser = sub.add_parser('verify', help='Verify a key+salt can open a database')
+    verify_parser.add_argument('--db', default=os.environ.get('WEFLOW_DB_PATH'), required=not os.environ.get('WEFLOW_DB_PATH'), help='Path to NT database')
+    verify_parser.add_argument('--key', default=os.environ.get('WEFLOW_NT_KEY'), required=not os.environ.get('WEFLOW_NT_KEY'), help='Key hex (64 chars)')
+    verify_parser.add_argument('--salt', default=os.environ.get('WEFLOW_NT_SALT'), required=not os.environ.get('WEFLOW_NT_SALT'), help='Salt hex (32 chars)')
+
+    # Same question as `verify`, answered without the native library - so a
+    # missing or broken sqlcipher3 does not hide whether the key is correct.
+    native_parser = sub.add_parser('verify-native',
+                                   help='Verify a passphrase from the file header alone (no sqlcipher3)')
+    native_parser.add_argument('--db', default=os.environ.get('WEFLOW_DB_PATH'), required=not os.environ.get('WEFLOW_DB_PATH'), help='Path to any database in the set')
+    native_parser.add_argument('--passphrase', default=os.environ.get('WEFLOW_NT_PASSPHRASE'), required=not os.environ.get('WEFLOW_NT_PASSPHRASE'), help='Passphrase hex (64 chars)')
+    native_parser.add_argument('--internal-key', default=os.environ.get('WEFLOW_INTERNAL_DB_KEY', ''), help='Optional 64-char hex XOR key (some builds)')
+
+    # fav-list command
+    fav_list_parser = sub.add_parser('fav-list', help='List favorite items')
+    fav_list_parser.add_argument('--db', default=os.environ.get('WEFLOW_DB_PATH'), required=not os.environ.get('WEFLOW_DB_PATH'), help='Path to favorite.db')
+    fav_list_parser.add_argument('--key', default=os.environ.get('WEFLOW_NT_KEY'), required=not os.environ.get('WEFLOW_NT_KEY'), help='Key hex (64 chars)')
+    fav_list_parser.add_argument('--limit', type=int, default=100)
+    fav_list_parser.add_argument('--offset', type=int, default=0)
+    fav_list_parser.add_argument('--keyword', default=os.environ.get('WEFLOW_QUERY_KEYWORD'), help='Filter by keyword')
+    fav_list_parser.add_argument('--type', type=int, dest='fav_type',
+                                 help='Filter by type: 1=text 2=image 4=video 5=article 14=chatrecord')
+
+    # cursor-probe command: the data-side half of D-027's stable-cursor answer.
+    cursor_parser = sub.add_parser(
+        'cursor-probe',
+        help='Probe what a stable incremental cursor would need here (read-only, bounded sample)')
+    cursor_parser.add_argument('--db', default=os.environ.get('WEFLOW_DB_PATH'), required=not os.environ.get('WEFLOW_DB_PATH'), help='Path to NT database')
+    cursor_parser.add_argument('--key', default=os.environ.get('WEFLOW_NT_KEY'), required=not os.environ.get('WEFLOW_NT_KEY'), help='Key hex (64 chars)')
+    cursor_parser.add_argument('--salt', default=os.environ.get('WEFLOW_NT_SALT'), required=not os.environ.get('WEFLOW_NT_SALT'), help='Salt hex (32 chars)')
+    cursor_parser.add_argument('--passphrase', default=os.environ.get('WEFLOW_NT_PASSPHRASE'), help='Shared passphrase for deriving per-shard keys')
+    cursor_parser.add_argument('--tables-per-shard', type=int, default=5,
+                               help='Message tables sampled per shard for the identity half (schema half is exhaustive)')
+    cursor_parser.add_argument('--rows-per-table', type=int, default=500,
+                               help='Rows read at each rowid end per sampled table')
+
+    args = parser.parse_args()
+
+    if args.command == 'scan':
+        # Discover the database files before touching the process. Walking the
+        # filesystem does not need WeChat running, but key matching does - and
+        # the caller derives keys from the passphrase when the memory scan
+        # finds nothing, which only needs the file list. Returning early on
+        # "process not running" without the databases threw that list away and
+        # left the caller with nothing to derive from.
+        databases = find_nt_databases(getattr(args, 'root', None))
+
+        pid = find_weixin_pid()
+        if not pid:
+            if IS_WINDOWS:
+                print(json.dumps({"error": "Weixin.exe 未运行", "databases": databases}))
+            else:
+                print(json.dumps({"error": "未检测到 Linux 微信进程，请确认微信已启动并登录",
+                                  "databases": databases}))
+            return
+
+        if not args.json:
+            print(f"扫描进程 PID {pid}...")
+
+        keys, scan_err = scan_memory_keys(pid)
+        if scan_err == 'permission':
+            print(json.dumps({"error": "PERMISSION_DENIED: 读取微信进程内存需要 root 或 CAP_SYS_PTRACE 权限。可使用 sudo 运行，或执行: sudo setcap cap_sys_ptrace=ep $(which python3)",
+                              "databases": databases}))
+            return
+        if scan_err == 'gone':
+            print(json.dumps({"error": "微信进程已退出，请重试", "databases": databases}))
+            return
+        if not args.json:
+            print(f"找到 {len(keys)} 个密钥")
+
+        if not args.json:
+            print(f"找到 {len(databases)} 个 NT 数据库")
+
+        matched = match_keys_to_databases(keys, databases)
+        if not args.json:
+            print(f"匹配 {len(matched)} 个数据库")
+            for db in matched:
+                print(f"  {db['name']} ({db['size']/1024/1024:.1f}MB) key={db['key'][:16]}... salt={db['salt'][:16]}...")
+        else:
+            print(json.dumps({"keys": keys, "databases": databases, "matched": matched}))
+        return
+
+    if args.command == 'verify':
+        # sqlcipher 打开+读 sqlite_master 才触发解密, 错误密钥在此报 "file is not a database"
+        try:
+            conn, _ = connect_nt_db(args.db, args.key, args.salt)
+            n = conn.execute(
+                "SELECT count(*) FROM sqlite_master WHERE type='table'"
+            ).fetchone()[0]
+            conn.close()
+            print(json.dumps({"success": n > 0, "tables": n}, ensure_ascii=True))
+        except Exception as e:
+            print(json.dumps({"success": False, "error": str(e).split('\n')[0][:200]}, ensure_ascii=True))
+        return
+
+    if args.command == 'verify-native':
+        verdict = verify_passphrase_native(args.passphrase, args.db,
+                                           getattr(args, 'internal_key', '') or '')
+        if verdict is None:
+            print(json.dumps({"success": False, "verified": None,
+                              "error": "无法判定（文件不可读、过短，或 passphrase 不是 64 位 hex）"},
+                             ensure_ascii=True))
+            return
+        print(json.dumps({"success": True, "verified": verdict}, ensure_ascii=True))
+        return
+
+    # Build contact name map once if contact db provided
+    contact_name_map = {}
+    contact_db = getattr(args, 'contact_db', None)
+    contact_key = getattr(args, 'contact_key', None)
+    contact_salt = getattr(args, 'contact_salt', None)
+    if contact_db and contact_key and contact_salt:
+        contact_name_map = load_contact_names(contact_db, contact_key, contact_salt)
+
+    if args.command == 'sessions':
+        conns = connect_message_shards(args.db, args.key, args.salt, getattr(args, 'passphrase', '') or '')
+        if not conns:
+            print(json.dumps({"error": "无法打开消息数据库，请检查密钥"}, ensure_ascii=True))
+            return
+        result = get_sessions(conns)
+        if 'sessions' in result:
+            result['sessions'] = apply_contact_names(result['sessions'], contact_name_map)
+            if args.keyword:
+                kw = args.keyword.lower()
+                result['sessions'] = [
+                    s for s in result['sessions']
+                    if kw in (s.get('username', '') + s.get('displayName', '') + s.get('summary', '')).lower()
+                ]
+        print(json.dumps(result, ensure_ascii=True))
+        for conn in conns:
+            conn.close()
+
+    elif args.command == 'messages':
+        passphrase = getattr(args, 'passphrase', '') or ''
+        own_wxid = getattr(args, 'own_wxid', None)
+
+        if getattr(args, 'report_shards', False):
+            pairs, failures = connect_message_shards_detailed(
+                args.db, args.key, args.salt, passphrase)
+            if not pairs:
+                print(json.dumps({"error": "无法打开消息数据库，请检查密钥"}, ensure_ascii=True))
+                return
+            conns = [conn for _path, conn in pairs]
+            names = [os.path.basename(path) for path, _conn in pairs]
+            shard_items = []
+            result = get_messages(conns, args.talker, args.limit, args.offset,
+                                  contact_name_map, own_wxid,
+                                  shard_names=names, shard_report=shard_items,
+                                  from_time=args.from_time, to_time=args.to_time)
+            # Failure entries carry no opened connection, so they are not in
+            # `conns` and could not be recorded by get_messages itself.
+            result['shards'] = {
+                'scanned': len(pairs) + len(failures),
+                'opened': len(pairs),
+                'failed': len(failures),
+                'items': failures + shard_items,
+            }
+        else:
+            conns = connect_message_shards(args.db, args.key, args.salt, passphrase)
+            if not conns:
+                print(json.dumps({"error": "无法打开消息数据库，请检查密钥"}, ensure_ascii=True))
+                return
+            result = get_messages(conns, args.talker, args.limit, args.offset,
+                                  contact_name_map, own_wxid,
+                                  from_time=args.from_time, to_time=args.to_time)
+
+        print(json.dumps(result, ensure_ascii=True))
+        for conn in conns:
+            conn.close()
+
+    elif args.command == 'contacts':
+        conns = connect_message_shards(args.db, args.key, args.salt, getattr(args, 'passphrase', '') or '')
+        if not conns:
+            print(json.dumps({"error": "无法打开消息数据库，请检查密钥"}, ensure_ascii=True))
+            return
+        result = get_contacts(conns, args.limit, keyword=args.keyword)
+        if 'contacts' in result:
+            result['contacts'] = apply_contact_names(result['contacts'], contact_name_map)
+            result['contacts'] = filter_contacts(result['contacts'], args.keyword, args.limit)
+        print(json.dumps(result, ensure_ascii=True))
+        for conn in conns:
+            conn.close()
+
+    elif args.command == 'sns-timeline':
+        conn, _ = connect_nt_db(args.db, args.key, args.salt)
+        usernames = None
+        if args.usernames:
+            try:
+                usernames = json.loads(args.usernames)
+            except: pass
+        result = get_sns_timeline(conn.cursor(), args.limit, args.offset,
+                                   usernames, args.keyword,
+                                   args.start_time, args.end_time)
+        print(json.dumps(result, ensure_ascii=True, default=str))
+        conn.close()
+
+    elif args.command == 'sns-usernames':
+        conn, _ = connect_nt_db(args.db, args.key, args.salt)
+        result = get_sns_usernames(conn.cursor())
+        print(json.dumps(result, ensure_ascii=True, default=str))
+        conn.close()
+
+    elif args.command == 'sns-stats':
+        conn, _ = connect_nt_db(args.db, args.key, args.salt)
+        result = get_sns_stats(conn.cursor(), args.my_wxid)
+        print(json.dumps(result, ensure_ascii=True, default=str))
+        conn.close()
+
+    elif args.command == 'fav-schema':
+        result = get_fav_schema(args.db, args.key)
+        print(json.dumps(result, ensure_ascii=True))
+
+    elif args.command == 'fav-list':
+        result = get_favorites(args.db, args.key, args.limit, args.offset,
+                               args.keyword, getattr(args, 'fav_type', None))
+        print(json.dumps(result, ensure_ascii=True, default=str))
+
+    elif args.command == 'cursor-probe':
+        result = cursor_probe(args.db, args.key, args.salt,
+                              getattr(args, 'passphrase', '') or '',
+                              tables_per_shard=args.tables_per_shard,
+                              rows_per_table=args.rows_per_table)
+        print(json.dumps(result, ensure_ascii=True))
+
+    else:
+        parser.print_help()
+
+
+if __name__ == '__main__':
+    main()