"""
集成 AI 能力的 QQ 消息处理器。
"""

import asyncio
import json
from pathlib import Path
import re
from typing import Any, Dict, Optional

import httpx

from botpy.message import Message

from src.ai import AIClient
from src.ai.base import ModelConfig, ModelProvider
from src.ai.mcp import MCPManager
from src.ai.mcp.servers import FileSystemMCPServer
from src.ai.personality import PersonalityProfile, PersonalityTrait
from src.ai.skills import SkillsManager
from src.core.config import Config
from src.utils.logger import setup_logger

logger = setup_logger("MessageHandler")


class MessageHandler:
    """消息处理器（集成 AI 功能）。"""

    MODEL_KEY_PATTERN = re.compile(r"[^a-zA-Z0-9_]")

    MARKDOWN_PATTERNS = [
        (re.compile(r"```[\s\S]*?```"), lambda m: m.group(0).replace("```", "")),
        (re.compile(r"`([^`]+)`"), r"\1"),
        (re.compile(r"\*\*([^*]+)\*\*"), r"\1"),
        (re.compile(r"\*([^*]+)\*"), r"\1"),
        (re.compile(r"__([^_]+)__"), r"\1"),
        # Avoid stripping underscores inside identifiers like model keys.
        (re.compile(r"(?<!\w)_([^_\n]+)_(?!\w)"), r"\1"),
        (re.compile(r"^#{1,6}\s*", re.MULTILINE), ""),
        (re.compile(r"^>\s?", re.MULTILINE), ""),
        (re.compile(r"\[([^\]]+)\]\(([^)]+)\)"), r"\1: \2"),
        (re.compile(r"^[-*]\s+", re.MULTILINE), "- "),
        (re.compile(r"^\d+\.\s+", re.MULTILINE), "- "),
        (re.compile(r"\n{3,}"), "\n\n"),
    ]

    def __init__(self, bot):
        self.bot = bot
        self.ai_client = None
        self.skills_manager = None
        self.mcp_manager = None
        self.model_profiles_path = Path("config/models.json")
        self.model_profiles: Dict[str, Dict[str, Any]] = {}
        self.active_model_key = "default"
        self._ai_initialized = False
        self._init_lock = asyncio.Lock()

    @staticmethod
    def _get_user_id(message: Message) -> str:
        author = getattr(message, "author", None)
        if not author:
            return "unknown"

        return (
            getattr(author, "id", None)
            or getattr(author, "user_openid", None)
            or getattr(author, "member_openid", None)
            or getattr(author, "union_openid", None)
            or "unknown"
        )

    @staticmethod
    def _build_skills_usage(command_name: str = "/skills") -> str:
        return (
            "技能命令：\n"
            f"{command_name} 或 {command_name} list\n"
            f"{command_name} install <source> [skill_name]\n"
            f"  source 支持：本地技能名、URL、owner/repo、owner/repo#branch、GitHub 仓库 URL(.git)\n"
            f"{command_name} uninstall <skill_name>\n"
            f"{command_name} reload <skill_name>"
        )

    @staticmethod
    def _build_personality_usage() -> str:
        return (
            "人设命令：\n"
            "/personality\n"
            "/personality list\n"
            "/personality set <key>\n"
            "/personality add <name> <Introduction>\n"
            "  兼容旧格式: <json> 或 管道 name|description|speaking_style|TRAIT1,TRAIT2|custom_instructions\n"
            "/personality remove <key>"
        )

    @staticmethod
    def _build_models_usage(command_name: str = "/models") -> str:
        return (
            "模型命令：\n"
            f"{command_name} 或 {command_name} list\n"
            f"{command_name} current\n"
            f"{command_name} add <model_name>  （保持 provider/api_base/api_key 不变）\n"
            f"{command_name} add <key> <provider> <model_name> [api_base]\n"
            f"{command_name} add <key> <json>\n"
            "  json 字段：provider, model_name, api_base, api_key, temperature, max_tokens, top_p, timeout\n"
            f"{command_name} switch <key|index>\n"
            f"{command_name} remove <key|index>"
        )

    @staticmethod
    def _build_memory_usage(command_name: str = "/memory") -> str:
        return (
            "记忆命令：\n"
            f"{command_name} 或 {command_name} list [limit]\n"
            f"{command_name} get <id>\n"
            f"{command_name} add <content>\n"
            f"{command_name} add <json>\n"
            f"{command_name} update <id> <content or json>\n"
            f"{command_name} delete <id>\n"
            f"{command_name} search <query> [limit]\n"
            "/clear  (默认等价 /clear short)\n"
            "/clear short\n"
            "/clear long\n"
            "/clear all"
        )

    @staticmethod
    def _provider_map() -> Dict[str, ModelProvider]:
        return {
            "openai": ModelProvider.OPENAI,
            "anthropic": ModelProvider.ANTHROPIC,
            "deepseek": ModelProvider.DEEPSEEK,
            "qwen": ModelProvider.QWEN,
            "siliconflow": ModelProvider.OPENAI,
        }

    @classmethod
    def _normalize_model_key(cls, raw_key: str) -> str:
        key = raw_key.strip().lower().replace("-", "_").replace(" ", "_")
        key = cls.MODEL_KEY_PATTERN.sub("_", key)
        key = re.sub(r"_+", "_", key).strip("_")
        if not key:
            raise ValueError("模型 key 不能为空")
        if key[0].isdigit():
            key = f"model_{key}"
        return key

    @staticmethod
    def _compact_model_key(raw_key: str) -> str:
        return re.sub(r"[^a-z0-9]", "", (raw_key or "").strip().lower())

    def _ordered_model_keys(self) -> list[str]:
        return sorted(self.model_profiles.keys())

    def _resolve_model_selector(self, selector: str) -> str:
        raw = (selector or "").strip()
        if not raw:
            raise ValueError("模型 key 不能为空")

        ordered_keys = self._ordered_model_keys()
        if raw.isdigit():
            index = int(raw)
            if index < 1 or index > len(ordered_keys):
                raise ValueError(
                    f"模型序号超出范围: {index}，可选 1-{len(ordered_keys)}"
                )
            return ordered_keys[index - 1]

        if raw in self.model_profiles:
            return raw

        normalized_selector: Optional[str]
        try:
            normalized_selector = self._normalize_model_key(raw)
        except ValueError:
            normalized_selector = None

        if normalized_selector and normalized_selector in self.model_profiles:
            return normalized_selector

        normalized_candidates: Dict[str, list[str]] = {}
        compact_candidates: Dict[str, list[str]] = {}
        for key in ordered_keys:
            try:
                normalized_key = self._normalize_model_key(key)
            except ValueError:
                normalized_key = key.strip().lower()
            normalized_candidates.setdefault(normalized_key, []).append(key)
            compact_candidates.setdefault(
                self._compact_model_key(normalized_key), []
            ).append(key)

        if normalized_selector and normalized_selector in normalized_candidates:
            matches = normalized_candidates[normalized_selector]
            if len(matches) == 1:
                return matches[0]
            raise ValueError(f"匹配到多个模型 key，请使用完整 key: {', '.join(matches)}")

        compact_selector = self._compact_model_key(normalized_selector or raw)
        if compact_selector in compact_candidates:
            matches = compact_candidates[compact_selector]
            if len(matches) == 1:
                return matches[0]
            raise ValueError(f"匹配到多个模型 key，请使用完整 key: {', '.join(matches)}")

        raise ValueError(f"模型配置不存在: {raw}")

    @classmethod
    def _parse_provider(cls, raw_provider: str) -> ModelProvider:
        provider = cls._provider_map().get(raw_provider.strip().lower())
        if not provider:
            raise ValueError(f"不支持的 provider: {raw_provider}")
        return provider

    @staticmethod
    def _coerce_float(value: Any, default: float) -> float:
        try:
            return float(value)
        except (TypeError, ValueError):
            return default

    @staticmethod
    def _coerce_int(value: Any, default: int) -> int:
        try:
            return int(value)
        except (TypeError, ValueError):
            return default

    @staticmethod
    def _coerce_bool(value: Any, default: bool) -> bool:
        if isinstance(value, bool):
            return value
        if isinstance(value, str):
            lowered = value.strip().lower()
            if lowered in {"1", "true", "yes", "on"}:
                return True
            if lowered in {"0", "false", "no", "off"}:
                return False
            return default
        if value is None:
            return default
        return bool(value)

    @staticmethod
    def _model_config_to_dict(
        config: ModelConfig, include_api_key: bool = True
    ) -> Dict[str, Any]:
        data: Dict[str, Any] = {
            "provider": config.provider.value,
            "model_name": config.model_name,
            "temperature": config.temperature,
            "max_tokens": config.max_tokens,
            "top_p": config.top_p,
            "frequency_penalty": config.frequency_penalty,
            "presence_penalty": config.presence_penalty,
            "timeout": config.timeout,
            "stream": config.stream,
        }
        if config.api_base:
            data["api_base"] = config.api_base
        if include_api_key and config.api_key:
            data["api_key"] = config.api_key
        return data

    def _model_config_from_dict(
        self, raw: Dict[str, Any], fallback: ModelConfig
    ) -> ModelConfig:
        provider = self._parse_provider(str(raw.get("provider") or fallback.provider.value))
        model_name = str(raw.get("model_name") or fallback.model_name).strip()
        if not model_name:
            raise ValueError("model_name 不能为空")

        api_key = raw.get("api_key")
        if api_key is None:
            api_key = fallback.api_key or Config.AI_API_KEY
        api_key = str(api_key) if api_key else ""

        api_base = raw.get("api_base", fallback.api_base)
        api_base = str(api_base) if api_base else None

        return ModelConfig(
            provider=provider,
            model_name=model_name,
            api_key=api_key,
            api_base=api_base,
            temperature=self._coerce_float(raw.get("temperature"), fallback.temperature),
            max_tokens=self._coerce_int(raw.get("max_tokens"), fallback.max_tokens),
            top_p=self._coerce_float(raw.get("top_p"), fallback.top_p),
            frequency_penalty=self._coerce_float(
                raw.get("frequency_penalty"), fallback.frequency_penalty
            ),
            presence_penalty=self._coerce_float(
                raw.get("presence_penalty"), fallback.presence_penalty
            ),
            timeout=self._coerce_int(raw.get("timeout"), fallback.timeout),
            stream=self._coerce_bool(raw.get("stream"), fallback.stream),
        )

    def _save_model_profiles(self):
        payload = {"active": self.active_model_key, "profiles": self.model_profiles}
        self.model_profiles_path.parent.mkdir(parents=True, exist_ok=True)
        with open(self.model_profiles_path, "w", encoding="utf-8") as f:
            json.dump(payload, f, ensure_ascii=False, indent=2)

    def _load_model_profiles(self, default_config: ModelConfig) -> ModelConfig:
        payload: Dict[str, Any] = {}
        if self.model_profiles_path.exists():
            try:
                with open(self.model_profiles_path, "r", encoding="utf-8") as f:
                    payload = json.load(f)
            except Exception as exc:
                logger.warning(f"load model profiles failed, reset to defaults: {exc}")
                payload = {}

        raw_profiles = payload.get("profiles")
        profiles: Dict[str, Dict[str, Any]] = {}
        if isinstance(raw_profiles, dict):
            for raw_key, raw_profile in raw_profiles.items():
                if not isinstance(raw_profile, dict):
                    continue

                key_text = str(raw_key or "").strip()
                if not key_text:
                    continue

                try:
                    normalized_key = self._normalize_model_key(key_text)
                except ValueError:
                    continue

                if normalized_key in profiles and profiles[normalized_key] != raw_profile:
                    logger.warning(
                        f"duplicate model key after normalization, keep first: {normalized_key}"
                    )
                    continue

                profiles[normalized_key] = raw_profile

        if not profiles:
            profiles = {
                "default": self._model_config_to_dict(
                    default_config, include_api_key=False
                )
            }

        active_raw = str(payload.get("active") or "").strip()
        active = ""
        if active_raw in profiles:
            active = active_raw
        elif active_raw:
            try:
                normalized_active = self._normalize_model_key(active_raw)
            except ValueError:
                normalized_active = ""
            if normalized_active in profiles:
                active = normalized_active

        if not active:
            active = "default" if "default" in profiles else sorted(profiles.keys())[0]

        self.model_profiles = profiles
        self.active_model_key = active

        try:
            active_config = self._model_config_from_dict(
                self.model_profiles[self.active_model_key], default_config
            )
        except Exception as exc:
            logger.warning(f"active model profile invalid, fallback to default: {exc}")
            self.model_profiles = {
                "default": self._model_config_to_dict(
                    default_config, include_api_key=False
                )
            }
            self.active_model_key = "default"
            active_config = default_config

        self._save_model_profiles()
        return active_config

    def _ensure_model_profiles_ready(self):
        if not self.ai_client:
            return

        if self.model_profiles and self.active_model_key in self.model_profiles:
            return

        active_config = self._load_model_profiles(self.ai_client.config)
        if active_config != self.ai_client.config:
            self.ai_client.switch_model(active_config)

    def _plain_text(self, text: str) -> str:
        if not text:
            return ""

        result = text
        for pattern, replacement in self.MARKDOWN_PATTERNS:
            result = pattern.sub(replacement, result)

        return result.strip()

    async def _reply_plain(self, message: Message, text: str):
        await message.reply(content=self._plain_text(text))

    def _register_skill_tools(self, skill_name: str) -> int:
        if not self.skills_manager or not self.ai_client:
            return 0

        skill = self.skills_manager.get_skill(skill_name)
        if not skill:
            return 0

        count = 0
        for tool_name, tool_func in skill.get_tools().items():
            full_tool_name = f"{skill_name}.{tool_name}"
            self.ai_client.register_tool(
                name=full_tool_name,
                description=f"技能工具: {full_tool_name}",
                parameters={"type": "object", "properties": {}},
                function=tool_func,
                source="skills",
            )
            count += 1

        return count

    async def _register_mcp_tools(self) -> int:
        if not self.mcp_manager or not self.ai_client:
            return 0

        tools = await self.mcp_manager.get_all_tools_for_ai()
        count = 0

        for item in tools:
            function_info = item.get("function") if isinstance(item, dict) else None
            if not isinstance(function_info, dict):
                continue

            full_tool_name = function_info.get("name")
            if not full_tool_name:
                continue

            parameters = function_info.get("parameters")
            if not isinstance(parameters, dict):
                parameters = {"type": "object", "properties": {}}

            async def _mcp_proxy(_full_tool_name=full_tool_name, **kwargs):
                if not self.mcp_manager:
                    raise RuntimeError("MCP manager not initialized")
                return await self.mcp_manager.execute_tool(_full_tool_name, kwargs)

            self.ai_client.register_tool(
                name=full_tool_name,
                description=f"MCP工具: {full_tool_name}",
                parameters=parameters,
                function=_mcp_proxy,
                source="mcp",
            )
            count += 1

        return count

    def _parse_traits(self, traits_raw) -> list[PersonalityTrait]:
        traits = []

        if isinstance(traits_raw, str):
            trait_names = [name.strip().upper() for name in traits_raw.split(",") if name.strip()]
        elif isinstance(traits_raw, list):
            trait_names = [str(name).strip().upper() for name in traits_raw if str(name).strip()]
        else:
            trait_names = []

        for name in trait_names:
            if name in PersonalityTrait.__members__:
                traits.append(PersonalityTrait[name])

        if not traits:
            traits = [PersonalityTrait.FRIENDLY]

        return traits

    def _parse_personality_payload(self, key: str, payload: str) -> PersonalityProfile:
        payload = payload.strip()

        if payload.startswith("{"):
            data = json.loads(payload)
            return PersonalityProfile(
                name=str(data.get("name") or key),
                description=str(data.get("description") or "自定义人设"),
                traits=self._parse_traits(data.get("traits", [])),
                speaking_style=str(data.get("speaking_style") or "自然口语"),
                example_responses=list(data.get("example_responses", [])),
                custom_instructions=str(data.get("custom_instructions") or ""),
            )

        if "|" not in payload:
            introduction = payload
            return PersonalityProfile(
                name=key,
                description=introduction,
                traits=[PersonalityTrait.FRIENDLY],
                speaking_style="自然口语",
                custom_instructions=introduction,
            )

        parts = [part.strip() for part in payload.split("|")]
        name = parts[0] if len(parts) >= 1 and parts[0] else key
        description = parts[1] if len(parts) >= 2 and parts[1] else "自定义人设"
        speaking_style = parts[2] if len(parts) >= 3 and parts[2] else "自然口语"
        traits = self._parse_traits(parts[3] if len(parts) >= 4 else "")
        custom_instructions = parts[4] if len(parts) >= 5 else ""

        return PersonalityProfile(
            name=name,
            description=description,
            traits=traits,
            speaking_style=speaking_style,
            custom_instructions=custom_instructions,
        )

    async def _init_ai(self):
        if self._ai_initialized:
            return

        try:
            provider = self._provider_map().get(
                Config.AI_PROVIDER.lower(), ModelProvider.OPENAI
            )

            config = ModelConfig(
                provider=provider,
                model_name=Config.AI_MODEL,
                api_key=Config.AI_API_KEY,
                api_base=Config.AI_API_BASE,
                temperature=0.7,
            )
            config = self._load_model_profiles(config)

            embed_config = None
            if Config.AI_EMBED_PROVIDER and Config.AI_EMBED_MODEL:
                embed_provider = self._provider_map().get(
                    Config.AI_EMBED_PROVIDER.lower(), ModelProvider.OPENAI
                )

                embed_config = ModelConfig(
                    provider=embed_provider,
                    model_name=Config.AI_EMBED_MODEL,
                    api_key=Config.AI_EMBED_API_KEY or Config.AI_API_KEY,
                    api_base=Config.AI_EMBED_API_BASE or Config.AI_API_BASE,
                )

            self.ai_client = AIClient(
                model_config=config,
                embed_config=embed_config,
                data_dir=Path("data/ai"),
                use_vector_db=Config.AI_USE_VECTOR_DB,
            )

            self.skills_manager = SkillsManager(Path("skills"))
            await self.skills_manager.load_all_skills()

            total_tools = 0
            for skill_name in self.skills_manager.list_skills():
                total_tools += self._register_skill_tools(skill_name)
            logger.info(
                f"技能系统初始化完成: {len(self.skills_manager.list_skills())} skills, {total_tools} tools"
            )

            try:
                self.mcp_manager = MCPManager(Path("config/mcp.json"))
                fs_server = FileSystemMCPServer(root_path=Path("data"))
                await self.mcp_manager.register_server(fs_server)
                mcp_tool_count = await self._register_mcp_tools()
                logger.info(f"MCP 工具注册完成: {mcp_tool_count} tools")
            except Exception as exc:
                logger.warning(f"MCP 初始化失败: {exc}")

            self._ai_initialized = True
            logger.info("AI 系统初始化完成")

        except Exception as exc:
            logger.error(f"AI 初始化失败: {exc}")
            import traceback

            logger.error(traceback.format_exc())

    async def handle_at_message(self, message: Message):
        try:
            if not self._ai_initialized:
                async with self._init_lock:
                    if not self._ai_initialized:
                        await self._init_ai()

            if not self.ai_client:
                await self._reply_plain(message, "AI 初始化失败，请检查配置")
                return

            content = (message.content or "").strip()
            user_id = self._get_user_id(message)

            if content.startswith(f"<@!{self.bot.robot.id}>"):
                content = content.replace(f"<@!{self.bot.robot.id}>", "").strip()

            if not content:
                return

            if content.startswith("/"):
                await self._handle_command(message, content)
                return

            response = await self.ai_client.chat(
                user_id=user_id,
                user_message=content,
                use_memory=True,
                use_tools=True,
            )
            await self._reply_plain(message, response)

        except Exception as exc:
            logger.error(f"处理消息失败: {exc}")
            import traceback

            logger.error(traceback.format_exc())
            if isinstance(exc, (httpx.ReadTimeout, TimeoutError, asyncio.TimeoutError)):
                await self._reply_plain(
                    message,
                    "模型响应超时，请稍后重试，或将当前模型配置的 timeout 调大（建议 120-180 秒）。",
                )
                return
            await self._reply_plain(message, "消息处理失败，请稍后重试")

    async def _handle_skills_command(self, message: Message, command: str):
        if not self.skills_manager or not self.ai_client:
            await self._reply_plain(message, "技能系统未初始化")
            return

        parts = command.split()
        command_name = parts[0].lower()
        action = parts[1].lower() if len(parts) > 1 else "list"

        if action in {"list", "ls"} and len(parts) <= 2:
            loaded = self.skills_manager.list_skills()
            available = self.skills_manager.list_available_skills()
            unloaded = [name for name in available if name not in loaded]

            lines = ["技能状态："]
            lines.append("已加载: " + (", ".join(loaded) if loaded else "无"))
            lines.append("可安装: " + (", ".join(unloaded) if unloaded else "无"))
            lines.append(self._build_skills_usage(command_name))
            await self._reply_plain(message, "\n".join(lines))
            return

        if action not in {"install", "uninstall", "remove", "reload"}:
            await self._reply_plain(message, self._build_skills_usage(command_name))
            return

        if len(parts) < 3:
            await self._reply_plain(message, self._build_skills_usage(command_name))
            return

        if action == "install":
            source = parts[2]
            desired_name = parts[3] if len(parts) > 3 else None

            try:
                source_key = self.skills_manager.normalize_skill_key(source)
            except Exception:
                source_key = None

            if source_key and source_key in self.skills_manager.list_available_skills():
                success = await self.skills_manager.load_skill(source_key)
                if not success:
                    await self._reply_plain(message, f"加载技能失败: {source_key}")
                    return

                tool_count = self._register_skill_tools(source_key)
                await self._reply_plain(
                    message,
                    f"已加载本地技能: {source_key}\n注册工具: {tool_count}",
                )
                return

            ok, result = self.skills_manager.install_skill_from_source(
                source=source,
                skill_name=desired_name,
                overwrite=False,
            )
            if not ok:
                await self._reply_plain(message, f"安装失败: {result}")
                return

            installed_key = result
            success = await self.skills_manager.load_skill(installed_key)
            if not success:
                await self._reply_plain(message, f"安装成功但加载失败: {installed_key}")
                return

            tool_count = self._register_skill_tools(installed_key)
            await self._reply_plain(
                message,
                f"已从来源安装并加载技能: {installed_key}\n注册工具: {tool_count}",
            )
            return

        skill_name = parts[2]
        try:
            skill_key = self.skills_manager.normalize_skill_key(skill_name)
        except Exception:
            await self._reply_plain(message, f"非法技能名: {skill_name}")
            return

        if action in {"uninstall", "remove"}:
            removed_tools = self.ai_client.unregister_tools_by_prefix(f"{skill_key}.")
            removed = await self.skills_manager.uninstall_skill(skill_key, delete_files=True)
            if not removed:
                await self._reply_plain(message, f"卸载失败或技能不存在: {skill_key}")
                return

            await self._reply_plain(
                message,
                f"已卸载技能: {skill_key}\n注销工具: {removed_tools}",
            )
            return

        self.ai_client.unregister_tools_by_prefix(f"{skill_key}.")
        success = await self.skills_manager.reload_skill(skill_key)
        if not success:
            await self._reply_plain(message, f"重载失败: {skill_key}")
            return

        tool_count = self._register_skill_tools(skill_key)
        await self._reply_plain(message, f"已重载技能: {skill_key}\n注册工具: {tool_count}")

    async def _handle_personality_command(self, message: Message, command: str):
        parts = command.split(maxsplit=3)

        if len(parts) == 1:
            current = self.ai_client.personality.current_personality
            names = self.ai_client.list_personalities()
            if current:
                await self._reply_plain(
                    message,
                    f"当前人设: {current.name}\n简介: {current.description}\n可用: {', '.join(names)}",
                )
            else:
                await self._reply_plain(message, "当前没有激活的人设")
            return

        action = parts[1].lower()

        if action == "list":
            names = self.ai_client.list_personalities()
            await self._reply_plain(message, "可用人设: " + ", ".join(names))
            return

        if action in {"set", "use"}:
            if len(parts) < 3:
                await self._reply_plain(message, self._build_personality_usage())
                return

            key = parts[2]
            if self.ai_client.set_personality(key):
                await self._reply_plain(message, f"已切换人设: {key}")
            else:
                await self._reply_plain(message, f"人设不存在: {key}")
            return

        if action in {"add", "create"}:
            if len(parts) < 4:
                await self._reply_plain(message, self._build_personality_usage())
                return

            key = parts[2]
            payload = parts[3]

            try:
                profile = self._parse_personality_payload(key, payload)
                ok = self.ai_client.personality.add_personality(key, profile)
                if not ok:
                    await self._reply_plain(message, f"新增人设失败: {key}")
                    return

                self.ai_client.set_personality(key)
                await self._reply_plain(message, f"已新增并切换人设: {key}")
            except Exception as exc:
                await self._reply_plain(message, f"新增人设失败: {exc}")
            return

        if action in {"remove", "delete"}:
            if len(parts) < 3:
                await self._reply_plain(message, self._build_personality_usage())
                return

            key = parts[2]
            removed = self.ai_client.personality.remove_personality(key)
            if removed:
                await self._reply_plain(message, f"已删除人设: {key}")
            else:
                await self._reply_plain(message, f"删除失败（可能是默认人设或不存在）: {key}")
            return

        # 兼容旧命令: /personality <name>
        if self.ai_client.set_personality(parts[1]):
            await self._reply_plain(message, f"已切换人设: {parts[1]}")
            return

        await self._reply_plain(message, self._build_personality_usage())

    async def _handle_memory_command(self, message: Message, command: str):
        if not self.ai_client:
            await self._reply_plain(message, "AI 客户端未初始化")
            return

        user_id = self._get_user_id(message)
        parts = command.split(maxsplit=3)
        action = parts[1].lower() if len(parts) > 1 else "list"

        if action in {"list", "ls"}:
            limit = 10
            if len(parts) >= 3:
                try:
                    limit = max(1, min(100, int(parts[2])))
                except ValueError:
                    await self._reply_plain(message, self._build_memory_usage())
                    return

            memories = await self.ai_client.list_long_term_memories(user_id, limit=limit)
            if not memories:
                await self._reply_plain(message, "暂无长期记忆")
                return

            lines = ["长期记忆列表："]
            for memory in memories:
                content = self._plain_text(memory.content).replace("\n", " ").strip()
                if len(content) > 60:
                    content = content[:57] + "..."
                lines.append(
                    f"- {memory.id} | 重要性={memory.importance:.2f} | {memory.timestamp.isoformat()} | {content}"
                )
            await self._reply_plain(message, "\n".join(lines))
            return

        if action == "get":
            if len(parts) < 3:
                await self._reply_plain(message, self._build_memory_usage())
                return

            memory_id = parts[2].strip()
            memory = await self.ai_client.get_long_term_memory(user_id, memory_id)
            if not memory:
                await self._reply_plain(message, f"记忆不存在: {memory_id}")
                return

            meta_text = json.dumps(memory.metadata or {}, ensure_ascii=False)
            await self._reply_plain(
                message,
                "记忆详情：\n"
                f"id: {memory.id}\n"
                f"重要性: {memory.importance:.2f}\n"
                f"时间: {memory.timestamp.isoformat()}\n"
                f"访问次数: {memory.access_count}\n"
                f"内容: {memory.content}\n"
                f"元数据: {meta_text}",
            )
            return

        if action == "add":
            if len(parts) < 3:
                await self._reply_plain(message, self._build_memory_usage())
                return

            payload = " ".join(parts[2:]).strip()
            content = payload
            importance = 0.8
            metadata = None

            if payload.startswith("{"):
                try:
                    data = json.loads(payload)
                    if not isinstance(data, dict):
                        raise ValueError("payload 必须是对象")
                    content = str(data.get("content") or "").strip()
                    importance = float(data.get("importance", 0.8))
                    raw_meta = data.get("metadata")
                    metadata = raw_meta if isinstance(raw_meta, dict) else None
                except Exception as exc:
                    await self._reply_plain(message, f"JSON 解析失败: {exc}")
                    return

            if not content:
                await self._reply_plain(message, "内容不能为空")
                return

            memory = await self.ai_client.add_long_term_memory(
                user_id=user_id,
                content=content,
                importance=importance,
                metadata=metadata,
            )
            if not memory:
                await self._reply_plain(message, "新增长期记忆失败")
                return

            await self._reply_plain(
                message, f"已新增长期记忆: {memory.id} (重要性={memory.importance:.2f})"
            )
            return

        if action in {"update", "set"}:
            if len(parts) < 4:
                await self._reply_plain(message, self._build_memory_usage())
                return

            memory_id = parts[2].strip()
            payload = parts[3].strip()
            content = payload
            importance = None
            metadata = None

            if payload.startswith("{"):
                try:
                    data = json.loads(payload)
                    if not isinstance(data, dict):
                        raise ValueError("payload 必须是对象")
                    if "content" in data:
                        content = str(data.get("content") or "")
                    else:
                        content = None
                    if "importance" in data:
                        importance = float(data.get("importance"))
                    raw_meta = data.get("metadata")
                    if raw_meta is not None:
                        if not isinstance(raw_meta, dict):
                            raise ValueError("metadata 必须是对象")
                        metadata = raw_meta
                except Exception as exc:
                    await self._reply_plain(message, f"JSON 解析失败: {exc}")
                    return

            updated = await self.ai_client.update_long_term_memory(
                user_id=user_id,
                memory_id=memory_id,
                content=content,
                importance=importance,
                metadata=metadata,
            )
            if not updated:
                await self._reply_plain(message, f"更新失败或记忆不存在: {memory_id}")
                return

            await self._reply_plain(
                message,
                f"已更新长期记忆: {memory_id} (重要性={updated.importance:.2f})",
            )
            return

        if action in {"delete", "remove", "rm"}:
            if len(parts) < 3:
                await self._reply_plain(message, self._build_memory_usage())
                return

            memory_id = parts[2].strip()
            deleted = await self.ai_client.delete_long_term_memory(user_id, memory_id)
            if not deleted:
                await self._reply_plain(message, f"删除失败或记忆不存在: {memory_id}")
                return

            await self._reply_plain(message, f"已删除长期记忆: {memory_id}")
            return

        if action in {"search", "find"}:
            if len(parts) < 3:
                await self._reply_plain(message, self._build_memory_usage())
                return

            query_payload = " ".join(parts[2:]).strip()
            limit = 10
            query = query_payload
            if " " in query_payload:
                possible_query, possible_limit = query_payload.rsplit(" ", 1)
                if possible_limit.isdigit():
                    query = possible_query
                    limit = max(1, min(100, int(possible_limit)))

            if not query:
                await self._reply_plain(message, "搜索词不能为空")
                return

            memories = await self.ai_client.search_long_term_memories(
                user_id=user_id,
                query=query,
                limit=limit,
            )
            if not memories:
                await self._reply_plain(message, f"没有匹配到相关记忆: {query}")
                return

            lines = [f"搜索结果（{len(memories)} 条）："]
            for memory in memories:
                content = self._plain_text(memory.content).replace("\n", " ").strip()
                if len(content) > 60:
                    content = content[:57] + "..."
                lines.append(f"- {memory.id} | 重要性={memory.importance:.2f} | {content}")
            await self._reply_plain(message, "\n".join(lines))
            return

        await self._reply_plain(message, self._build_memory_usage())

    async def _handle_models_command(self, message: Message, command: str):
        if not self.ai_client:
            await self._reply_plain(message, "AI 客户端未初始化")
            return

        self._ensure_model_profiles_ready()

        parts = command.split(maxsplit=3)
        action = parts[1].lower() if len(parts) > 1 else "list"

        if action in {"list", "ls"} and len(parts) <= 2:
            lines = [f"当前模型配置: {self.active_model_key}"]
            ordered_keys = self._ordered_model_keys()
            for idx, key in enumerate(ordered_keys, start=1):
                profile = self.model_profiles.get(key, {})
                marker = "*" if key == self.active_model_key else "-"
                provider = str(profile.get("provider") or "?")
                model_name = str(profile.get("model_name") or "?")
                lines.append(f"{marker} {idx}. {key}: {provider}/{model_name}")

            if ordered_keys:
                lines.append(f"提示: 可用 /models switch <序号>，例如 /models switch 2")

            lines.append(self._build_models_usage("/models"))
            await self._reply_plain(message, "\n".join(lines))
            return

        if action in {"current", "show"}:
            config = self.ai_client.config
            await self._reply_plain(
                message,
                "当前模型：\n"
                f"配置名: {self.active_model_key}\n"
                f"供应商: {config.provider.value}\n"
                f"模型: {config.model_name}\n"
                f"API 地址: {config.api_base or '-'}",
            )
            return

        if action in {"switch", "set", "use"}:
            if len(parts) < 3:
                await self._reply_plain(message, self._build_models_usage())
                return

            try:
                key = self._resolve_model_selector(parts[2])
            except ValueError as exc:
                await self._reply_plain(message, str(exc))
                return

            try:
                config = self._model_config_from_dict(
                    self.model_profiles[key], self.ai_client.config
                )
                self.ai_client.switch_model(config)
            except Exception as exc:
                await self._reply_plain(message, f"切换模型失败: {exc}")
                return

            self.active_model_key = key
            self._save_model_profiles()
            await self._reply_plain(
                message, f"已切换模型: {key} ({config.provider.value}/{config.model_name})"
            )
            return

        if action in {"add", "create"}:
            # 快捷方式: /models add <model_name>
            # 仅替换 model_name，保留 provider/api_base/api_key。
            if len(parts) == 3:
                model_name = parts[2].strip()
                if not model_name:
                    await self._reply_plain(message, self._build_models_usage())
                    return

                key = self._normalize_model_key(model_name)
                config = ModelConfig(
                    provider=self.ai_client.config.provider,
                    model_name=model_name,
                    api_key=self.ai_client.config.api_key,
                    api_base=self.ai_client.config.api_base,
                    temperature=self.ai_client.config.temperature,
                    max_tokens=self.ai_client.config.max_tokens,
                    top_p=self.ai_client.config.top_p,
                    frequency_penalty=self.ai_client.config.frequency_penalty,
                    presence_penalty=self.ai_client.config.presence_penalty,
                    timeout=self.ai_client.config.timeout,
                    stream=self.ai_client.config.stream,
                )

                self.model_profiles[key] = self._model_config_to_dict(
                    config, include_api_key=False
                )
                self.active_model_key = key
                self._save_model_profiles()
                self.ai_client.switch_model(config)

                await self._reply_plain(
                    message,
                    f"已保存并切换模型: {key} ({config.provider.value}/{config.model_name})",
                )
                return

            if len(parts) < 4:
                await self._reply_plain(message, self._build_models_usage())
                return

            try:
                key = self._normalize_model_key(parts[2])
            except ValueError as exc:
                await self._reply_plain(message, str(exc))
                return

            payload = parts[3].strip()
            include_api_key = False

            try:
                if payload.startswith("{"):
                    raw_profile = json.loads(payload)
                    if not isinstance(raw_profile, dict):
                        raise ValueError("模型参数必须是 JSON 对象")
                else:
                    payload_parts = payload.split()
                    if len(payload_parts) < 2:
                        raise ValueError(
                            "用法: /models add <key> <provider> <model_name> [api_base]"
                        )
                    raw_profile = {
                        "provider": payload_parts[0],
                        "model_name": payload_parts[1],
                    }
                    if len(payload_parts) >= 3:
                        raw_profile["api_base"] = payload_parts[2]

                include_api_key = bool(raw_profile.get("api_key"))
                config = self._model_config_from_dict(raw_profile, self.ai_client.config)
            except Exception as exc:
                await self._reply_plain(message, f"新增模型失败: {exc}")
                return

            self.model_profiles[key] = self._model_config_to_dict(
                config, include_api_key=include_api_key
            )
            self.active_model_key = key
            self._save_model_profiles()
            self.ai_client.switch_model(config)

            await self._reply_plain(
                message,
                f"已保存并切换模型: {key} ({config.provider.value}/{config.model_name})",
            )
            return

        if action in {"remove", "delete", "rm"}:
            if len(parts) < 3:
                await self._reply_plain(message, self._build_models_usage())
                return

            try:
                key = self._resolve_model_selector(parts[2])
            except ValueError as exc:
                await self._reply_plain(message, str(exc))
                return

            if key == "default":
                await self._reply_plain(message, "默认模型配置不能删除")
                return

            del self.model_profiles[key]
            switched_to = None

            if self.active_model_key == key:
                fallback_key = (
                    "default"
                    if "default" in self.model_profiles
                    else (sorted(self.model_profiles.keys())[0] if self.model_profiles else None)
                )

                if not fallback_key:
                    self.model_profiles["default"] = self._model_config_to_dict(
                        self.ai_client.config, include_api_key=False
                    )
                    fallback_key = "default"

                fallback_config = self._model_config_from_dict(
                    self.model_profiles[fallback_key], self.ai_client.config
                )
                self.ai_client.switch_model(fallback_config)
                self.active_model_key = fallback_key
                switched_to = fallback_key

            self._save_model_profiles()

            if switched_to:
                await self._reply_plain(
                    message, f"已删除模型: {key}，已切换到: {switched_to}"
                )
            else:
                await self._reply_plain(message, f"已删除模型: {key}")
            return

        await self._reply_plain(message, self._build_models_usage())

    async def _handle_command(self, message: Message, command: str):
        user_id = self._get_user_id(message)

        if command == "/help":
            await self._reply_plain(
                message,
                "命令帮助\n"
                "====================\n"
                "基础命令\n"
                "/help\n"
                "/clear  (默认等价 /clear short)\n"
                "/clear short\n"
                "/clear long\n"
                "/clear all\n"
                "\n"
                "人设命令\n"
                "/personality\n"
                "/personality list\n"
                "/personality set <key>\n"
                "/personality add <name> <Introduction>\n"
                "/personality remove <key>\n"
                "\n"
                "技能命令\n"
                "/skills\n"
                "/skills install <source> [skill_name]\n"
                "/skills uninstall <skill_name>\n"
                "/skills reload <skill_name>\n"
                "\n"
                "模型命令\n"
                "/models\n"
                "/models current\n"
                "/models add <model_name>\n"
                "/models add <key> <provider> <model_name> [api_base]\n"
                "/models switch <key|index>\n"
                "/models remove <key|index>\n"
                "\n"
                "记忆命令\n"
                "/memory\n"
                "/memory get <id>\n"
                "/memory add <content|json>\n"
                "/memory update <id> <content|json>\n"
                "/memory delete <id>\n"
                "/memory search <query> [limit]\n"
                "\n"
                "任务命令\n"
                "/task <task_id>",
            )
            return

        if command.startswith("/clear"):
            clear_parts = command.split(maxsplit=1)
            clear_scope = clear_parts[1].strip().lower() if len(clear_parts) > 1 else "short"

            if clear_scope in {"all", ""}:
                cleared_all = await self.ai_client.clear_all_memory(user_id)
                if cleared_all:
                    await self._reply_plain(message, "已清除短期记忆和长期记忆")
                else:
                    await self._reply_plain(
                        message,
                        "短期记忆已清除，但长期记忆清除失败",
                    )
                return

            if clear_scope in {"short", "short_term"}:
                self.ai_client.clear_memory(user_id)
                await self._reply_plain(message, "已清除短期记忆")
                return

            if clear_scope in {"long", "long_term"}:
                cleared_long = await self.ai_client.clear_long_term_memory(user_id)
                if cleared_long:
                    await self._reply_plain(message, "已清除长期记忆")
                else:
                    await self._reply_plain(message, "清除长期记忆失败")
                return

            await self._reply_plain(
                message,
                "用法:\n/clear\n/clear short\n/clear long\n/clear all",
            )
            return

        if command.startswith("/personality"):
            await self._handle_personality_command(message, command)
            return

        if command.startswith("/skills") or command.startswith("/skill"):
            await self._handle_skills_command(message, command)
            return

        if command.startswith("/models") or command.startswith("/model"):
            await self._handle_models_command(message, command)
            return

        if command.startswith("/memory") or command.startswith("/mem"):
            await self._handle_memory_command(message, command)
            return

        if command.startswith("/task"):
            parts = command.split(maxsplit=1)
            if len(parts) == 2:
                task_id = parts[1]
                status = self.ai_client.get_task_status(task_id)
                if status:
                    await self._reply_plain(
                        message,
                        "任务状态：\n"
                        f"标题: {status['title']}\n"
                        f"状态: {status['status']}\n"
                        f"进度: {status['progress'] * 100:.1f}%\n"
                        f"步骤: {status['completed_steps']}/{status['total_steps']}",
                    )
                else:
                    await self._reply_plain(message, "任务不存在")
            else:
                await self._reply_plain(message, "用法: /task <task_id>")
            return

        await self._reply_plain(message, "未知命令，请输入 /help 查看帮助")