from openai import OpenAI
from dataclasses import dataclass

@dataclass
class AgentSpec:
    name: str
    role: str
    tools: list[str]
    model: str = "gpt-4o"

def meta_agent(task: str, available_tool_registry: dict) -> str:
    """
    Analyzes the task and dynamically spawns the right specialist agents.
    Returns the final assembled result.
    """
    client = OpenAI()

    # Phase 1: meta-agent decides what agents are needed
    planning_response = client.chat.completions.create(
        model="gpt-4o",
        messages=[
            {"role": "system", "content": (
                "You are a team architect. Given a task, decide what specialist agents are needed. "
                f"Available tools: {list(available_tool_registry.keys())}. "
                "Return JSON: [{name, role, tools, reason}]"
            )},
            {"role": "user", "content": f"Task: {task}"},
        ],
        response_format={"type": "json_object"},
    )

    import json
    team_specs = json.loads(planning_response.choices[0].message.content)["agents"]

    # Phase 2: spawn and run each specialist
    results = {}
    for spec in team_specs:
        agent_tools = [available_tool_registry[t] for t in spec["tools"] if t in available_tool_registry]
        result = run_specialist(
            role=spec["role"],
            task=f"Your part of the task: {spec['reason']}. Full context: {task}",
            tools=agent_tools,
            model=spec.get("model", "gpt-4o"),
        )
        results[spec["name"]] = result

    # Phase 3: meta-agent assembles final output
    final = client.chat.completions.create(
        model="gpt-4o",
        messages=[
            {"role": "system", "content": "Synthesize the specialist outputs into a single coherent result."},
            {"role": "user", "content": f"Original task: {task}\n\nSpecialist results:\n{json.dumps(results, indent=2)}"},
        ],
    )
    return final.choices[0].message.content