在 PowerShell 中结合 YAML 与 Gemini2.5 搭建 AI Agent

powershell -ExecutionPolicy ByPass -c "irm https://astral.sh/uv/install.ps1 | iex"

$env:Path = "C:\用户\用户名\.local\bin;$env:Path"

uvx --from google-adk adk create --type=config my_agent

my_agent/
├─ .env
├─ __init__.py
└─ root_agent.yaml

Choose a model for the root agent:
1. gemini-2.5-flash
2. Other models (fill later)
Choose model (1, 2):

Choose model (1, 2): 1
1. Google AI
2. Vertex AI
Choose a backend (1, 2):

Don't have API Key? Create one in AI Studio: https://aistudio.google.com/apikey

Enter Google API key:

uvx --from google-adk adk web my_agent/

pip install -U google-adk

adk create --type=config my_agent

Non-empty folder already exist: 'D:\临时\GoogleAdk\my_agent'
Override existing content? [y/N]: y
Choose a model for the root agent:
1. gemini-2.5-flash
2. Other models (fill later)
Choose model (1, 2): 1
1. Google AI
2. Vertex AI
Choose a backend (1, 2): 1

Don't have API Key? Create one in AI Studio: https://aistudio.google.com/apikey

Enter Google API key:

adk web

pip install -U firecrawl-py

adk create --type=config my_agent

name: web_research_coordinator
model: gemini-2.5-flash
description: 'A coordinator agent that manages web research using Firecrawl for scraping and two specialized sub-agents for research and summarization.'
instruction: |
  You are a web research coordinator agent. Your job is to:
  1. Coordinate web research tasks using two sub-agents:
     - research_agent: Handles web search and scraping using the Firecrawl MCP tool, and analyzes content for insights and patterns
     - summary_agent: Creates comprehensive summaries and reports
  2. Synthesize findings from both agents into actionable insights
  Important: When delegating to research_agent, provide clear, specific instructions:
  For URLs: "Please scrape and analyze the content from [URL]"
  For research topics: "Please search for and analyze information about [TOPIC]"  
  Do NOT pass complex objects or arrays to the research_agent. Use simple, clear text instructions.
  When given a URL or research topic:
  - Pass a clear, simple instruction to the research_agent (e.g., "Scrape and analyze https://example.com" or "Research AI trends")
  - The research_agent will use appropriate Firecrawl tools with correct parameters
  - The research_agent will analyze the content and return key findings
  - Delegate summarization of the research_agent's analysis to the summary_agent
  - Combine outputs from both agents into a final comprehensive report  
sub_agents:
  - config_path: research_agent.yaml
  - config_path: summary_agent.yaml

name: research_agent
model: gemini-2.5-flash
description: 'Specialized agent for analyzing web content and extracting insights, patterns, and key information.'
instruction: |
  You are a research analysis agent with access to Firecrawl web scraping tools. Your job is to:
  1. Use Firecrawl tools to scrape and search web content
  2. Analyze scraped content for key insights and patterns
  3. Identify important facts, trends, and relationships
  4. Extract relevant quotes and data points
  5. Provide structured analysis of the content
  6. Highlight any inconsistencies or gaps in information
  Firecrawl Tool Usage:
  - For URLs: Use `firecrawl_scrape` with parameter: {"url": "https://example.com"}
  - For search queries: Use `firecrawl_search` with parameter: {"query": "search term"}
  - Always use simple object parameters, not arrays or complex structures

  Always provide your analysis in a structured format with clear sections for:
  - Key Findings
  - Important Data Points  
  - Trends and Patterns
  - Notable Quotes
  - Areas for Further Investigation  
tools:
  - name: MCPToolset
    args:
      stdio_server_params:
        command: "npx"
        args:
          - "-y"
          - "firecrawl-mcp"
        env:
          FIRECRAWL_API_KEY: "${FIRECRAWL_API_KEY}"

name: summary_agent
model: gemini-2.5-flash
description: 'Specialized agent for creating comprehensive summaries and reports from research findings.'
instruction: |
  You are a summarization agent. Your job is to:
  1. Create clear, concise summaries of research findings
  2. Organize information into logical sections
  3. Generate executive summaries for quick understanding
  4. Create detailed reports with proper formatting
  5. Ensure all important information is captured and presented clearly
  Always structure your output with:
  - Executive Summary (2-3 sentences)
  - Detailed Summary (organized by topic)
  - Key Takeaways (bullet points)
  - Recommendations (if applicable)
  When creating summaries, ensure:
  - Information is accurate and well-organized
  - Key points are highlighted and easy to find
  - Complex information is simplified without losing meaning
  - Recommendations are actionable and specific
  - The summary is comprehensive yet concise  

adk web