Agent Orchestration Architecture

knowledge skill

Designs the structure of multi-agent workflows by selecting agent roles, communication protocols, and execution order. Produces a directed acyclic graph (DAG) defining agent interactions and dependencies.

Worked examples

• {"input": {"agents": ["data_loader", "preprocessor", "model_trainer", "evaluator"], "dependencies": {"preprocessor": ["data_loader"], "model_trainer": ["preprocessor"], "evaluator": ["model_trainer"]}, "protocol": "message_queue"}, "output": {"workflow_graph": {"nodes": ["data_loader", "preprocessor", "model_trainer", "evaluator"], "edges": [["data_loader", "preprocessor"], ["preprocessor", "model_trainer"], ["model_trainer", "evaluator"]]}, "execution_order": ["data_loader", "preprocessor", "model_trainer", "evaluator"], "communication_spec": {"queues": {"data_loader": "raw_data", "preprocessor": "processed_data", "model_trainer": "training_data", "evaluator": "metrics"}}}}
• {"input": {"agents": ["api_gateway", "auth_service", "user_service", "logging_service"], "dependencies": {"auth_service": ["api_gateway"], "user_service": ["auth_service"], "logging_service": ["api_gateway", "user_service"]}, "protocol": "http"}, "output": {"workflow_graph": {"nodes": ["api_gateway", "auth_service", "user_service", "logging_service"], "edges": [["api_gateway", "auth_service"], ["auth_service", "user_service"], ["api_gateway", "logging_service"], ["user_service", "logging_service"]]}, "execution_order": ["api_gateway", "auth_service", "user_service", "logging_service"], "communication_spec": {"endpoints": {"auth_service": "/auth", "user_service": "/users", "logging_service": "/logs"}}}}

Input

• agents: array<string> - List of agent identifiers to include in the workflow.
• dependencies: object<string, array<string>> - Maps each agent to its prerequisite agents.
• protocol: string - Communication method (e.g., 'message_queue', 'shared_memory', 'http')

Output

• workflow_graph: object - DAG representing agent interactions with nodes (agents) and edges (dependencies).
• execution_order: array<string> - Topologically sorted list of agents for sequential execution.
• communication_spec: object - Protocol-specific configuration (e.g., queue names, ports)

Details

• Skill type: knowledge skill
• Safety level: safe_public_research
• Version: 1.0.0