"""把组织包装成 cabinet 可调用的高阶 tool。

每个组织 → 一个 ``dispatch_to_<org>(task_description)`` 工具。
ConsciousnessNode/ControlNode 通过这个工具向部门派单，等待部门完成。
"""

from __future__ import annotations

from typing import Callable, Dict


def make_dispatch_tool(org_name: str, display_name: str, description: str) -> Callable:
    """生成对应组织的 dispatch tool。

    工具签名故意保持简单：只收一个自然语言任务描述，cabinet 不需要懂部门内部
    capability 划分；部门内部 ReAct 自己决定怎么干。
    """
    tool_name = f"dispatch_to_{org_name}"
    desc_text = description or f"把任务派给{display_name or org_name}部门，由部门内部多 agent 协作完成。"

    async def _impl(task_description: str) -> str:
        from kilostar.utils.ray_hook import ray_actor_hook

        actor_name = f"org_{org_name}"
        actor = ray_actor_hook(actor_name)
        target = getattr(actor, actor_name)
        result = await target.dispatch.remote(task_description, {})
        if result.get("status") == "completed":
            return str(result.get("result") or "")
        return f"[{org_name} 任务失败] {result.get('error') or 'unknown'}"

    _impl.__name__ = tool_name
    _impl.__doc__ = (
        f"{desc_text}\n\n"
        "Args:\n"
        "    task_description: 用自然语言描述要部门完成的任务。\n\n"
        "Returns:\n"
        "    部门交付的结果文本，失败时返回错误说明。\n"
    )
    return _impl


def collect_dispatch_tools(org_specs: Dict[str, Dict[str, str]]) -> Dict[str, Callable]:
    """根据 ``{org_name: {"display_name": ..., "description": ...}}`` 批量生成。"""
    return {
        f"dispatch_to_{name}": make_dispatch_tool(
            name,
            spec.get("display_name", ""),
            spec.get("description", ""),
        )
        for name, spec in org_specs.items()
    }