LQConstPlanWriterAgent/core/document_chat/component/skill_dispatcher.py

# -*- coding: utf-8 -*-
"""文档对话技能注册表与分发器。"""

from dataclasses import dataclass
from pathlib import Path
from typing import Callable, Dict, List, Type

import yaml

from core.document_chat.schemas import DocumentChatSkillInput, DocumentChatSkillOutput
from core.document_chat.skills.base import BaseDocumentChatSkill
from core.document_chat.skills.document_answer import DocumentAnswerSkill
from core.document_chat.skills.document_modify import DocumentModifySkill


@dataclass(frozen=True)
class SkillDefinition:
    name: str
    description: str
    intent: str
    function_name: str
    handler_class: Type[BaseDocumentChatSkill]
    response_type: str

    def to_registry_item(self) -> Dict[str, str]:
        return {
            "name": self.name,
            "description": self.description,
            "intent": self.intent,
            "function_name": self.function_name,
            "handler_class": self.handler_class.__name__,
            "response_type": self.response_type,
        }


class SkillDispatcher:
    """基于白名单的技能分发器。"""

    _HANDLER_CLASSES: Dict[str, Type[BaseDocumentChatSkill]] = {
        "DocumentModifySkill": DocumentModifySkill,
        "DocumentAnswerSkill": DocumentAnswerSkill,
    }

    def __init__(self):
        self._definitions: Dict[str, SkillDefinition] = self._load_definitions()
        self._instances: Dict[str, BaseDocumentChatSkill] = {}

    def registry_for_prompt(self) -> List[Dict[str, str]]:
        return [definition.to_registry_item() for definition in self._definitions.values()]

    def has_skill(self, skill_name: str) -> bool:
        return skill_name in self._definitions

    async def run_skill(
        self,
        skill_name: str,
        skill_input: DocumentChatSkillInput,
    ) -> DocumentChatSkillOutput:
        if skill_name not in self._definitions:
            raise ValueError(f"Unsupported document chat skill: {skill_name}")
        skill = self._get_instance(skill_name)
        return await skill.run(skill_input)

    async def run_skill_stream(
        self,
        skill_name: str,
        skill_input: DocumentChatSkillInput,
        on_chunk: Callable[[str], None],
    ) -> DocumentChatSkillOutput:
        if skill_name not in self._definitions:
            raise ValueError(f"Unsupported document chat skill: {skill_name}")
        skill = self._get_instance(skill_name)
        return await skill.run_stream(skill_input, on_chunk)

    def _get_instance(self, skill_name: str) -> BaseDocumentChatSkill:
        if skill_name not in self._instances:
            definition = self._definitions[skill_name]
            self._instances[skill_name] = definition.handler_class(
                name=definition.name,
                function_name=definition.function_name,
            )
        return self._instances[skill_name]

    def _load_definitions(self) -> Dict[str, SkillDefinition]:
        skills_root = Path(__file__).resolve().parents[1] / "skills"
        definitions: Dict[str, SkillDefinition] = {}
        for skill_yaml in sorted(skills_root.glob("*/skill.yaml")):
            with open(skill_yaml, "r", encoding="utf-8") as handle:
                data = yaml.safe_load(handle) or {}
            definition = self._definition_from_yaml(data, skill_yaml)
            definitions[definition.name] = definition
        return definitions

    def _definition_from_yaml(self, data: dict, source: Path) -> SkillDefinition:
        required_fields = ["name", "description", "intent", "function_name", "handler_class", "response_type"]
        missing = [field for field in required_fields if not data.get(field)]
        if missing:
            raise ValueError(f"Skill配置缺少字段 {missing}: {source}")

        handler_name = str(data["handler_class"])
        handler_class = self._HANDLER_CLASSES.get(handler_name)
        if handler_class is None:
            raise ValueError(f"Skill配置使用了未注册的 handler_class: {handler_name}, source={source}")

        return SkillDefinition(
            name=str(data["name"]),
            description=str(data["description"]),
            intent=str(data["intent"]),
            function_name=str(data["function_name"]),
            handler_class=handler_class,
            response_type=str(data["response_type"]),
        )