ai-agent-orchestrator

dengineproblem's avatarfrom dengineproblem

Эксперт по оркестрации AI агентов. Используй для multi-agent systems, agent coordination, task delegation и agent workflows.

⭐0stars🔀0forks📁View on GitHub🕐Updated Jan 4, 2026

When & Why to Use This Skill

This Claude skill serves as a comprehensive framework for designing and managing sophisticated multi-agent systems. It provides expert guidance on agent orchestration, task delegation, and the implementation of resilient architectural patterns such as Hub-and-Spoke and Pipelines to optimize complex AI workflows.

Use Cases

  • Architecting multi-agent systems where a central orchestrator decomposes complex user requests and delegates subtasks to specialized agents for research, analysis, or execution.
  • Implementing intelligent task routing mechanisms that dynamically assign work to agents based on their specific capabilities, skill sets, and current processing load.
  • Developing resilient AI infrastructures using advanced error-handling patterns like Circuit Breakers and Graceful Degradation to ensure system stability during agent or API failures.
  • Establishing structured inter-agent communication protocols using message buses and correlation IDs to maintain traceability and data integrity across distributed agent networks.
nameai-agent-orchestrator
descriptionЭксперт по оркестрации AI агентов. Используй для multi-agent systems, agent coordination, task delegation и agent workflows.

AI Agent Orchestrator Expert

Эксперт по проектированию и реализации многоагентных систем.

Основные принципы

Иерархия агентов

  • Оркестратор: Главный координатор, делегирует задачи
  • Специализированные агенты: Для конкретных доменов (исследования, анализ, код)
  • Утилитарные агенты: Вспомогательные функции (валидация, форматирование)
  • Мониторинг: Здоровье системы и обработка ошибок

Паттерны архитектуры

Hub and Spoke

class OrchestratorAgent:
    def __init__(self):
        self.agents = {
            'researcher': ResearchAgent(),
            'analyzer': AnalysisAgent(),
            'writer': WritingAgent(),
            'validator': ValidationAgent()
        }

    async def orchestrate_task(self, task):
        subtasks = self.decompose_task(task)

        results = []
        for subtask in subtasks:
            agent_type = self.route_task(subtask)
            result = await self.agents[agent_type].execute(subtask)
            results.append(result)

        return self.synthesize_results(results)

Pipeline Pattern

class AgentPipeline:
    def __init__(self):
        self.stages = [
            DataIngestionAgent(),
            ProcessingAgent(),
            AnalysisAgent(),
            OutputAgent()
        ]

    async def execute_pipeline(self, input_data):
        data = input_data
        for stage in self.stages:
            try:
                data = await stage.process(data)
            except Exception as e:
                return await self.handle_pipeline_error(stage, e, data)
        return data

Маршрутизация задач

class TaskRouter:
    def __init__(self):
        self.agent_capabilities = {
            'code_analysis': ['python', 'javascript', 'sql'],
            'research': ['web_search', 'document_analysis'],
            'writing': ['technical', 'creative', 'business']
        }
        self.agent_load = {}

    def route_task(self, task):
        required_skills = self.extract_skills(task)

        capable_agents = [
            agent_id for agent_id, skills in self.agent_capabilities.items()
            if self.has_required_skills(skills, required_skills)
        ]

        return self.select_least_loaded_agent(capable_agents)

Межагентная коммуникация

@dataclass
class AgentMessage:
    sender_id: str
    receiver_id: str
    message_type: MessageType
    payload: Dict[str, Any]
    correlation_id: str
    timestamp: float
    priority: int = 5

class MessageBus:
    async def send_message(self, message: AgentMessage):
        await self.validate_message(message)
        await self.route_message(message)
        await self.log_message(message)

Обработка ошибок

Circuit Breaker

class AgentCircuitBreaker:
    def __init__(self, failure_threshold=5, timeout=60):
        self.failure_count = 0
        self.failure_threshold = failure_threshold
        self.state = 'CLOSED'

    async def call_agent(self, agent, task):
        if self.state == 'OPEN':
            if time.time() - self.last_failure_time > self.timeout:
                self.state = 'HALF_OPEN'
            else:
                raise CircuitBreakerOpenError()

        try:
            result = await agent.execute(task)
            if self.state == 'HALF_OPEN':
                self.reset()
            return result
        except Exception as e:
            self.record_failure()
            raise

Graceful Degradation

class ResilientOrchestrator:
    def __init__(self):
        self.agent_priorities = {
            'primary': ['gpt-4', 'claude-3'],
            'fallback': ['gpt-3.5', 'local-model'],
            'emergency': ['rule-based-agent']
        }

    async def execute_with_fallback(self, task):
        for tier in ['primary', 'fallback', 'emergency']:
            for agent_id in self.agent_priorities[tier]:
                try:
                    if await self.is_agent_healthy(agent_id):
                        return await self.execute_on_agent(agent_id, task)
                except Exception:
                    continue
        raise AllAgentsFailedError()

Лучшие практики

  • Реализуйте комплексное логирование с correlation ID
  • Отслеживайте метрики производительности агентов
  • Используйте распределенную трассировку для сложных workflows
  • Валидируйте всю межагентную коммуникацию
  • Проектируйте агентов как stateless когда возможно
  • Используйте очереди сообщений для развязки