Context Engineering MCP Platform

import logging from typing import List, Dict, Optional, Any from datetime import datetime import json import google.generativeai as genai from workflow_models import Workflow, Task, Agent, AgentType, TaskStatus logger = logging.getLogger(__name__) class AgentManager: def __init__(self, gemini_api_key: str): genai.configure(api_key=gemini_api_key) self.model = genai.GenerativeModel('gemini-2.0-flash-exp') async def assign_tasks(self, workflow: Workflow) -> Dict[str, str]: """ ワークフロー内のタスクを最適なエージェントにアサイン """ assignments = {} ready_tasks = workflow.get_ready_tasks() available_agents = workflow.get_available_agents() for task in ready_tasks: best_agent = await self._find_best_agent(task, available_agents, workflow) if best_agent: self._assign_task_to_agent(task, best_agent) assignments[task.id] = best_agent.id # エージェントが満杯になったら除外 if not best_agent.is_available: available_agents.remove(best_agent) return assignments async def _find_best_agent(self, task: Task, agents: List[Agent], workflow: Workflow) -> Optional[Agent]: """ タスクに最適なエージェントを選択 """ if not agents: return None try: # Gemini AIを使って最適なエージェントを選択 prompt = self._create_agent_selection_prompt(task, agents, workflow) response = self.model.generate_content(prompt) selection_data = json.loads(response.text) selected_agent_id = selection_data.get('selected_agent_id') reasoning = selection_data.get('reasoning', '') # 選択されたエージェントを見つける for agent in agents: if agent.id == selected_agent_id: logger.info(f"Assigned task '{task.title}' to agent '{agent.name}': {reasoning}") return agent # フォールバック: 最初の利用可能なエージェント return agents[0] if agents else None except Exception as e: logger.error(f"Agent selection failed: {str(e)}") # フォールバック: 最初の利用可能なエージェント return agents[0] if agents else None def _create_agent_selection_prompt(self, task: Task, agents: List[Agent], workflow: Workflow) -> str: """エージェント選択用のプロンプトを作成""" agents_info = [] for agent in agents: agents_info.append({ 'id': agent.id, 'name': agent.name, 'type': agent.type.value, 'capabilities': agent.capabilities, 'current_load': f"{len(agent.current_tasks)}/{agent.max_concurrent_tasks}", 'load_percentage': agent.load_percentage }) return f""" Task to assign: Title: {task.title} Description: {task.description} Priority: {task.priority} Estimated Duration: {task.estimated_duration} minutes Available Agents: {json.dumps(agents_info, indent=2)} Workflow Context: Title: {workflow.title} Total Tasks: {len(workflow.tasks)} Progress: {workflow.progress_percentage:.1f}% Select the best agent for this task considering: 1. Agent capabilities matching task requirements 2. Current workload and availability 3. Agent type suitability for the task 4. Overall workflow efficiency Return JSON in this format: {{ "selected_agent_id": "agent_id_here", "reasoning": "Explanation of why this agent was selected", "confidence": 0.85 }} Return only valid JSON. """ def _assign_task_to_agent(self, task: Task, agent: Agent): """タスクをエージェントにアサイン""" task.assigned_agent_id = agent.id task.status = TaskStatus.IN_PROGRESS task.started_at = datetime.now() agent.current_tasks.append(task.id) async def reassign_task(self, task: Task, workflow: Workflow, reason: str = "") -> Optional[Agent]: """ タスクを別のエージェントに再アサイン """ # 現在のエージェントから解除 if task.assigned_agent_id: current_agent = next((a for a in workflow.agents if a.id == task.assigned_agent_id), None) if current_agent and task.id in current_agent.current_tasks: current_agent.current_tasks.remove(task.id) # 新しいエージェントを見つけて再アサイン available_agents = workflow.get_available_agents() new_agent = await self._find_best_agent(task, available_agents, workflow) if new_agent: self._assign_task_to_agent(task, new_agent) logger.info(f"Reassigned task '{task.title}' to agent '{new_agent.name}'. Reason: {reason}") return new_agent return None def get_agent_performance_metrics(self, agent: Agent, workflow: Workflow) -> Dict[str, Any]: """エージェントのパフォーマンス指標を取得""" agent_tasks = [task for task in workflow.tasks if task.assigned_agent_id == agent.id] completed_tasks = [task for task in agent_tasks if task.status == TaskStatus.COMPLETED] failed_tasks = [task for task in agent_tasks if task.status == TaskStatus.FAILED] total_estimated_time = sum(task.estimated_duration or 0 for task in completed_tasks) total_actual_time = sum(task.duration_minutes or 0 for task in completed_tasks) return { 'agent_id': agent.id, 'agent_name': agent.name, 'total_tasks': len(agent_tasks), 'completed_tasks': len(completed_tasks), 'failed_tasks': len(failed_tasks), 'success_rate': len(completed_tasks) / len(agent_tasks) if agent_tasks else 0, 'current_load': len(agent.current_tasks), 'load_percentage': agent.load_percentage, 'average_task_duration': sum(task.duration_minutes or 0 for task in completed_tasks) / len(completed_tasks) if completed_tasks else 0, 'efficiency_ratio': total_estimated_time / total_actual_time if total_actual_time > 0 else 1.0 } class WorkflowExecutor: def __init__(self, gemini_api_key: str): self.agent_manager = AgentManager(gemini_api_key) async def execute_workflow(self, workflow: Workflow) -> None: """ ワークフローを実行 """ workflow.status = WorkflowStatus.EXECUTING workflow.started_at = datetime.now() while workflow.progress_percentage < 100: # 準備完了のタスクをアサイン assignments = await self.agent_manager.assign_tasks(workflow) if not assignments: # アサインできるタスクがない場合は、完了を待つかブロックされたタスクを処理 blocked_tasks = [task for task in workflow.tasks if task.status == TaskStatus.BLOCKED] if blocked_tasks: await self._handle_blocked_tasks(blocked_tasks, workflow) break # ここで実際のタスク実行ロジックを呼び出す # （実際の実装では、エージェントが非同期でタスクを実行） if workflow.progress_percentage == 100: workflow.status = WorkflowStatus.COMPLETED workflow.completed_at = datetime.now() async def _handle_blocked_tasks(self, blocked_tasks: List[Task], workflow: Workflow) -> None: """ブロックされたタスクを処理""" for task in blocked_tasks: # 依存関係を再チェック completed_task_ids = {t.id for t in workflow.tasks if t.status == TaskStatus.COMPLETED} if all(dep_id in completed_task_ids for dep_id in task.dependencies): task.status = TaskStatus.PENDING logger.info(f"Unblocked task '{task.title}'") def get_workflow_status(self, workflow: Workflow) -> Dict[str, Any]: """ワークフローの現在の状態を取得""" return { 'workflow_id': workflow.id, 'title': workflow.title, 'status': workflow.status.value, 'progress_percentage': workflow.progress_percentage, 'task_distribution': {status.value: count for status, count in workflow.task_distribution.items()}, 'total_tasks': len(workflow.tasks), 'total_agents': len(workflow.agents), 'created_at': workflow.created_at.isoformat(), 'started_at': workflow.started_at.isoformat() if workflow.started_at else None, 'completed_at': workflow.completed_at.isoformat() if workflow.completed_at else None, 'estimated_completion': self._estimate_completion_time(workflow) } def _estimate_completion_time(self, workflow: Workflow) -> Optional[str]: """完了予定時刻を推定""" if workflow.status == WorkflowStatus.COMPLETED: return None remaining_tasks = [task for task in workflow.tasks if task.status in [TaskStatus.PENDING, TaskStatus.IN_PROGRESS]] if not remaining_tasks: return None total_remaining_time = sum(task.estimated_duration or 30 for task in remaining_tasks) available_agents = len(workflow.get_available_agents()) if available_agents > 0: estimated_minutes = total_remaining_time / available_agents estimated_completion = datetime.now() + timedelta(minutes=estimated_minutes) return estimated_completion.isoformat() return None