How do agents share data with each other?

Agents communicate using a shared session state. One agent can save results to the context state, which the next agent in the orchestration flow can then retrieve and process.

When should I use the Coordinator pattern?

You should use the Coordinator pattern when a single agent's prompt becomes too complex or monolithic. It allows a central 'manager' agent to route tasks to specialized 'worker' agents, improving accuracy and maintainability.

Can this skill help with parallel processing?

Yes, it includes patterns for ParallelAgents which can execute multiple independent sub-agents simultaneously, significantly reducing the total execution time for complex analysis tasks.

What is the difference between an LlmAgent and a WorkflowAgent?

LlmAgents use dynamic LLM reasoning to decide which tools to call at runtime, making them ideal for unpredictable tasks. WorkflowAgents follow a hardcoded, deterministic sequence (Sequential, Parallel, or Loop), making them better for repeatable automation pipelines.

Multi-Agent Orchestration Patterns

Name: Multi-Agent Orchestration Patterns
Author: Agentient

byAgentient

0•

Productivity & Workflow

Implements robust multi-agent systems using coordinator patterns, specialized sub-agents, and deterministic workflows.

This skill provides a comprehensive framework for building complex agentic systems by decomposing monolithic tasks into focused, specialized agent teams. It enables developers to strategically choose between dynamic LlmAgents for reasoning-heavy tasks and deterministic WorkflowAgents (Sequential, Parallel, or Loop) for predictable automation pipelines. By utilizing the 'agent-as-tool' pattern and shared session state, it ensures that Claude can build modular, maintainable, and scalable AI architectures that outperform single-agent prompts.

Key Features

01Agent-as-tool pattern for modular and recursive agent architectures

02Dynamic LlmAgent reasoning for adaptive, open-ended problem solving

030 GitHub stars

04Deterministic WorkflowAgents for predictable Sequential, Parallel, and Loop execution

05Inter-agent communication via managed session state and context sharing

06Root Coordinator pattern for intelligent task delegation to specialists

Use Cases

01Building research assistants that delegate to specialized search, analysis, and synthesis agents

02Developing high-performance data processing workflows using parallel analysis agents

03Creating automated code review pipelines that sequentially lint, test, and analyze security

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add agentient/vibekit agent-orchestration

For use in Claude.ai and ChatGPT

Download Skill