Tavily 搜索

**Judgment:** A solid, practical Tavily-based web search skill that serves well as a Brave web_search alternative, with good format flexibility and straightforward setup, but reliant on an external API key and third‑party service. **Strengths & advantages** - Integrates cleanly into OpenClaw via a simple Python script; no custom server needed. - Multiple output formats (raw JSON, Brave‑style JSON, Markdown) support both machine consumption and human review. - Brave‑compatible schema (`{query, results:[{title,url,snippet}], answer?}`) makes it easy to swap in where Brave web_search is expected. - Encourages small `max-results` for efficiency, which helps manage token and bandwidth usage. **Key risks & limitations** - **External dependency:** Requires a valid Tavily API key and working Tavily service; outages, rate limits, or quota exhaustion will degrade reliability. - **Cost & quota:** Usage may incur Tavily costs; heavy search workloads must be monitored and rate‑limited. - **Content risk:** As with any web search, results may include low‑quality, biased, or unsafe content; downstream logic must enforce safety, filtering, and policy compliance. - **Error handling not fully specified:** Behavior on network failures, invalid keys, or malformed queries depends on the script’s internal error handling (not fully visible from the description). **Recommended scenarios** - Use as the **primary or fallback web search tool** when Brave web_search is unavailable or unsuitable. - Use in **tools that expect Brave-like search results**, leveraging the `--format brave` schema for easy integration. - Use where you need **quick URLs and snippets** (e.g., link-gathering, citation support), leaving heavy page fetching and parsing to other tools. - Avoid for **strictly offline, air‑gapped, or zero‑external‑dependency** environments, and be cautious in **high-compliance domains** without additional filtering and logging around the search results.