tavily-search

What This Skill Does

The tavily-search skill is a specialized search engine integration designed specifically for OpenClaw AI agents. Unlike standard search engines that return cluttered UI-heavy results, Tavily provides clean, parsed text extracted from the web, optimized directly for LLM consumption. It retrieves relevant content snippets, source URLs, and relevance scores, allowing your agent to synthesize information, fact-check claims, or research emerging topics without leaving the terminal or the AI environment.

Installation

To integrate this skill, ensure you have the OpenClaw environment set up. Run the following command in your terminal: clawhub install openclaw/skills/skills/15914355527/liang-tavily-search-v2. After installation, you must configure your authentication by providing a valid API key from tavily.com. This can be added to your OpenClaw JSON config file under the skills object or injected via the TAVILY_API_KEY environment variable. Ensure the key starts with 'tvly-' for proper authorization.

Use Cases

• Research & Documentation: Quickly find technical documentation or library patterns without manual browser navigation.
• Fact Checking: Verify statements or news reports by searching for recent, authoritative sources.
• Content Aggregation: Build agents that monitor specific industries (e.g., AI news) and report summaries based on the --topic news flag.
• Targeted Data Extraction: Use domain filtering to look exclusively within developer platforms like GitHub, StackOverflow, or official documentation sites.

Example Prompts

1. "Search for the latest best practices for implementing async patterns in Python and summarize the top three recommendations."
2. "Find recent news regarding advancements in large language models from this week and provide a bulleted list of findings."
3. "Search for 'React hook tutorials' but restrict results to official react.dev domains and return 5 results."

Tips & Limitations

To get the most out of your searches, keep your queries concise—aim for under 400 characters and frame them as specific search strings rather than conversational prompts. If a topic is complex, break it into smaller sub-queries to avoid overloading the context window. Use the --deep flag for research tasks where precision is paramount, and use --raw-content if you need the full text for deeper analysis. Note that results are subject to the limitations of the web index; always verify critical information.