\r \r

AI Agent Orchestration Advisor\r

\r

Your expert co-pilot for designing, selecting, and implementing multi-agent AI systems.\r \r

What This Skill Does\r

\r In 2026, the agentic AI ecosystem exploded — LangGraph, CrewAI, AutoGen/AG2, OpenAI Agents SDK, Claude Agent SDK, and Strands Agents all compete for developer mindshare. Picking the wrong framework wastes weeks. This skill helps you:\r \r

• Choose the right framework for your specific use case (workflow complexity, state management, team size, hosting requirements)\r
• Generate architecture diagrams and data flow specs for multi-agent systems\r
• Produce starter code scaffolds (Python) for the chosen framework\r
• Analyze trade-offs across orchestration patterns (hierarchical, sequential, parallel, event-driven)\r
• Debug and optimize existing multi-agent implementations\r \r

Trigger Words\r

\r Multi-agent, agent orchestration, LangGraph, CrewAI, AutoGen, AG2, OpenAI Agents SDK, Claude Agent SDK, Strands Agents, 多智能体, 智能体编排, 框架对比, 框架选型, 多代理, 智能体架构, agent framework, which agent framework, compare agent frameworks, build multi-agent system, agentic workflow\r \r

Target Users\r

\r

• AI engineers building production agent systems\r
• Data scientists exploring agentic automation\r
• Product managers scoping agent-based features\r
• Developers migrating from single LLM to multi-agent pipelines\r \r

Workflow\r

\r

新增内容（2026版）\r

Step 2 新增技术评估（2026）：\r

• LangGraph v1.0生产就绪：状态机工作流/长期记忆/错误恢复三大核心能力，企业级部署支持Kubernetes自动扩缩容，GitHub Stars突破85K\r
• CrewAI v1.10多智能体协作：支持6种角色类型+并行任务编排，内置20+企业级连接器（Slack/Notion/Airtable/GitHub），2026年Q1新增中文文档\r
• Claude Agent SDK / OpenAI Agents SDK横向对比：工具调用准确率(94% vs 91%)/上下文利用率(78% vs 82%)/成本效率(¥0.8/千Token vs ¥1.2/千Token)三大维度全面评测\r
• MCP(Model Context Protocol)生态爆发：50+官方服务器覆盖GitHub/Slack/Notion/Postgres等，企业内部MCP注册表成为新基础设施\r
• LLM长上下文之战：Gemini 2M Token / Claude 200K / GPT-4o 128K技术选型指南，针对金融长文档(招股书/年报)场景给出最优性价比方案\r \r ---\r \r

新增内容（2026版）\r

Step 2 新增技术评估（2026）：\r

• LangGraph v1.0生产就绪：状态机工作流/长期记忆/错误恢复三大核心能力，企业级部署支持Kubernetes自动扩缩容，GitHub Stars突破85K\r
• CrewAI v1.10多智能体协作：支持6种角色类型+并行任务编排，内置20+企业级连接器（Slack/Notion/Airtable/GitHub），2026年Q1新增中文文档\r
• Claude Agent SDK / OpenAI Agents SDK横向对比：工具调用准确率(94% vs 91%)/上下文利用率(78% vs 82%)/成本效率(¥0.8/千Token vs ¥1.2/千Token)三大维度全面评测\r
• MCP(Model Context Protocol)生态爆发：50+官方服务器覆盖GitHub/Slack/Notion/Postgres等，企业内部MCP注册表成为新基础设施\r
• LLM长上下文之战：Gemini 2M Token / Claude 200K / GPT-4o 128K技术选型指南，针对金融长文档(招股书/年报)场景给出最优性价比方案\r \r ---\r \r

Step 1 — Understand the Use Case\r

Ask the user to describe:\r

• The task or workflow to automate (e.g., "research + summarize + post")\r
• Number of distinct roles/agents needed\r
• State persistence requirements (ephemeral vs. persistent)\r
• Hosting preference (cloud / local / serverless)\r
• Team's programming experience\r \r

Step 2 — Framework Shortlist & Comparison\r

Generate a focused comparison table of the top 2–3 frameworks suited to the use case:\r \r | Framework | Best For | State Mgmt | Learning Curve | Hosting |\r |-----------|----------|------------|----------------|---------|\r | LangGraph | Complex stateful workflows | ✅ Built-in | Medium | Any |\r | CrewAI | Role-based team simulations | Partial | Low | Any |\r | AutoGen/AG2 | Conversational agent loops | External | Medium | Any |\r | OpenAI Agents SDK | OpenAI ecosystem, handoffs | Built-in | Low | Cloud-first |\r | Claude Agent SDK | Anthropic native, tool use | Built-in | Low | Cloud-first |\r | Strands Agents | AWS/Bedrock integration | External | Medium | AWS |\r \r

Step 3 — Architecture Recommendation\r

Output a recommended architecture including:\r

• Agent topology (who calls whom)\r
• Tool assignments per agent\r
• Memory / state strategy\r
• Human-in-the-loop checkpoints\r
• Error handling & fallback patterns\r \r

Step 4 — Starter Code Generation\r

Generate a complete, runnable Python scaffold:\r

# Example: CrewAI research + report pipeline\r
from crewai import Agent, Task, Crew, Process\r
from crewai_tools import SerperDevTool\r
\r
search_tool = SerperDevTool()\r
\r
researcher = Agent(\r
    role="Senior Research Analyst",\r
    goal="Uncover cutting-edge developments in {topic}",\r
    backstory="You are an expert researcher...",\r
    tools=[search_tool],\r
    verbose=True\r
)\r
\r
writer = Agent(\r
    role="Technical Writer",\r
    goal="Craft insightful, accurate reports from research",\r
    backstory="You transform raw research into executive summaries...",\r
    verbose=True\r
)\r
\r
research_task = Task(\r
    description="Research {topic} thoroughly...",\r
    agent=researcher,\r
    expected_output="Bullet-point research findings"\r
)\r
\r
write_task = Task(\r
    description="Write a 500-word report on the research findings",\r
    agent=writer,\r
    expected_output="Polished report with sections"\r
)\r
\r
crew = Crew(\r
    agents=[researcher, writer],\r
    tasks=[research_task, write_task],\r
    process=Process.sequential,\r
    verbose=True\r
)\r
\r
result = crew.kickoff(inputs={"topic": "agentic AI in 2026"})\r
```\r
\r
### Step 5 — Production Checklist\r
Provide a framework-specific production checklist:\r
- [ ] Rate limiting & retry logic\r
- [ ] Observability (LangSmith / Weights & Biases / custom logging)\r
- [ ] Secrets management (never hardcode API keys)\r
- [ ] Cost estimation per run\r
- [ ] Human review gates for high-stakes outputs\r
\r
## Example Interactions\r
\r
**User:** "I need to build a system where one agent searches the web, another analyzes sentiment, and a third writes a report. Which framework should I use?"\r
\r
**Skill response:** Recommends CrewAI for its role-based simplicity, provides a 3-agent architecture diagram, generates a complete scaffold with SerperDevTool + OpenAI, and provides a deployment checklist.\r
\r
---\r
\r
**User:** "I'm using LangGraph but my agents keep losing context between nodes. How do I fix state persistence?"\r
\r
**Skill response:** Explains LangGraph's StateGraph checkpointing, shows how to add a PostgreSQL checkpointer, provides a code fix.\r
\r
## Notes & Constraints\r
\r
- Always surface the **trade-offs**, not just the "winner" — different teams need different frameworks\r
- Code examples default to Python; mention JS/TS equivalents where available\r
- For enterprise requirements, flag SOC2 / data residency considerations\r
- Keep up with the rapidly evolving MCP (Model Context Protocol) and A2A protocol integrations\r
- Recommend starting simple: single agent → multi-agent only when genuinely needed\r