AgentSkillsCN

google-search-skill

执行 Google 搜索以从网络获取最新信息。当用户需要当前信息、最新新闻、技术趋势、文档查找或一般网络搜索，且需要超出你知识范围的实时数据时使用。

SKILL.md
--- frontmatter
name: google-search-skill
description: Perform Google searches to retrieve up-to-date information from the web. Use when users need current information, latest news, technical trends, documentation lookups, or general web searches that require real-time data beyond your knowledge cutoff.
allowed-tools: Bash(python:*)

Google Search Skill

Overview

This Skill enables Claude to perform Google searches using the google_search_tool command-line interface. It retrieves current web information, making it ideal for answering questions that require up-to-date data, recent news, or information beyond Claude's knowledge cutoff.

When to Use This Skill

Use this Skill when:

  • •Users ask for current events, news, or recent information
  • •Questions require the latest technical documentation or API references
  • •Looking up current prices, statistics, or data
  • •Finding resources, tutorials, or guides on specific topics
  • •Verifying facts or getting multiple perspectives from web sources
  • •Users explicitly request a web search or Google search

Instructions

When performing a web search, follow these steps:

1. Identify the Search Query

  • •Extract the key search terms from the user's request
  • •Formulate a clear, concise search query
  • •Use specific keywords for better results
  • •Include relevant technical terms, product names, or specific phrases

2. Execute the Search

Run the search using the google_search_tool:

bash
python -m google_search_tool "your search query" --pretty

Options:

  • •Use --pretty for formatted JSON output (easier to read)
  • •Use -n <number> to specify number of results (1-10, default: 10)
  • •For focused searches, use -n 3 or -n 5

Examples:

bash
# General search with 5 results
python -m google_search_tool "Python 3.13 new features" -n 5 --pretty

# Focused search with 3 results
python -m google_search_tool "React hooks tutorial" -n 3 --pretty

# Full search with 10 results
python -m google_search_tool "machine learning frameworks comparison" --pretty

3. Parse and Present Results

After receiving the search results:

  1. •Check for errors: If the output contains "error", explain the error to the user
  2. •Extract key information: Parse the JSON output for title, link, and snippet
  3. •Summarize findings: Provide a concise summary of what you found
  4. •Present sources: List relevant results with titles and links
  5. •Answer the question: Use the search results to answer the user's original question

Presentation format:

code
Based on my search for "[query]", here's what I found:

[Summary of findings based on search results]

Key resources:
1. **[Title 1]** - [Brief description from snippet]
   [URL]

2. **[Title 2]** - [Brief description from snippet]
   [URL]

[Additional context or recommendations]

4. Handle Edge Cases

No results found:

json
{
  "results": [],
  "count": 0
}
  • •Inform the user that no results were found
  • •Suggest trying different search terms
  • •Offer to search with alternative queries

API error:

json
{
  "error": "Error message"
}
  • •Check if environment variables are set (GOOGLE_API_KEY, GOOGLE_CSE_ID)
  • •Explain the error to the user
  • •Suggest troubleshooting steps if applicable

Timeout or network error:

  • •Inform the user of the issue
  • •Offer to retry the search
  • •Suggest checking network connectivity

5. Follow-up Searches

If the initial search doesn't fully answer the question:

  • •Refine the search query based on initial results
  • •Perform additional targeted searches
  • •Combine information from multiple searches

Examples

Example 1: Current Events

User request: "What are the latest developments in AI this week?"

Steps:

  1. •Search: python -m google_search_tool "latest AI developments 2025" -n 5 --pretty
  2. •Parse the JSON results
  3. •Summarize key developments from the snippets
  4. •Present with source links

Example 2: Technical Documentation

User request: "How do I use React Server Components?"

Steps:

  1. •Search: python -m google_search_tool "React Server Components guide" -n 5 --pretty
  2. •Identify official documentation and tutorials
  3. •Extract key concepts from snippets
  4. •Provide overview with links to detailed resources

Example 3: Comparison Research

User request: "Compare PostgreSQL vs MySQL for large-scale applications"

Steps:

  1. •Search: python -m google_search_tool "PostgreSQL vs MySQL large scale comparison" -n 5 --pretty
  2. •Gather multiple perspectives from results
  3. •Synthesize comparisons from different sources
  4. •Present balanced view with source attribution

Important Notes

Environment Setup

The google_search_tool requires:

  • •GOOGLE_API_KEY: Google Cloud API key
  • •GOOGLE_CSE_ID: Custom Search Engine ID
  • •These should be set in the .env file in the project root

Limitations

  • •Maximum 10 results per search
  • •API rate limits apply (check Google CSE quotas)
  • •Results depend on Google's indexing and ranking
  • •Snippets are truncated (may not contain full context)

Best Practices

  1. •Be specific: Use precise search terms for better results
  2. •Verify sources: Check that URLs and snippets are relevant before citing
  3. •Cite properly: Always include source URLs when presenting information
  4. •Respect recency: Recent results may be more relevant for time-sensitive queries
  5. •Multiple searches: For complex topics, perform several targeted searches rather than one broad search

Troubleshooting

If searches fail:

  1. •Verify the google_search_tool is installed: pip show google-search-mcp
  2. •Check environment variables are set in .env
  3. •Test manually: python -m google_search_tool "test query" --pretty
  4. •Review API quota limits in Google Cloud Console
  5. •Check network connectivity

Integration with Other Tools

This Skill works well alongside:

  • •Read tool: After finding URLs, read local documentation
  • •WebFetch tool: Retrieve full content from found URLs
  • •Write tool: Save search results for later reference