Action Plan:
AI Agent Team Build

# agents/ceo/graph.py

from typing import TypedDict, Annotated, Literal
from langgraph.graph import StateGraph, END
from langchain_anthropic import ChatAnthropic
from core.db import get_pending_tasks, save_result

class TeamState(TypedDict):
    command: str              # Входная команда
    tasks: list              # Список задач для агентов
    results: dict            # Результаты выполнения
    current_agent: str      # Текущий активный агент
    plan: str               # План действий от CEO

llm = ChatAnthropic(model="claude-sonnet-4-20250514")

def ceo_plan(state: TeamState) -> TeamState:
    """CEO анализирует команду и создаёт план."""
    response = llm.invoke(
        f"""You are the CEO of a music promotion AI team.
        Command: {state['command']}

        Create a plan: which agents need to act, in what order.
        Available agents: social, analytics, content, distribution,
                         playlist, engagement, youtube

        Return JSON: {{"tasks": [{{"agent": "...", "action": "..."}}]}}"""
    )
    # Парсим план и создаём задачи
    state["plan"] = response.content
    return state

def route_to_agent(state: TeamState) -> str:
    """Определяет следующего агента."""
    pending = [t for t in state["tasks"] if t["status"] == "pending"]
    if not pending:
        return "summarize"
    return pending[0]["agent"]

def social_node(state): ...  # Вызывает Social Agent
def analytics_node(state): ...  # Вызывает Analytics Agent
def summarize_node(state): ...  # CEO подводит итоги

# Сборка графа
graph = StateGraph(TeamState)
graph.add_node("ceo_plan", ceo_plan)
graph.add_node("social", social_node)
graph.add_node("analytics", analytics_node)
graph.add_node("summarize", summarize_node)

graph.set_entry_point("ceo_plan")
graph.add_conditional_edges("ceo_plan", route_to_agent)
graph.add_edge("social", "ceo_plan")      # Возврат к CEO
graph.add_edge("analytics", "ceo_plan")
graph.add_edge("summarize", END)

app = graph.compile()