tavily-web

Tavily Web 技能

触发条件与端点选择

根据用户意图选择 Tavily 端点：

• search：需要“搜索网页／最新信息／查找来源／查找链接”
• extract：给定 URL，需要提取／总结内容
• crawl：需要按照说明遍历网站并抓取页面内容
• map：需要发现网站的页面列表／结构（不含完整内容，或仅获取元数据）
• research：需要按照给定的 output_schema 生成结构化研究结果

推荐架构（主技能 + 子技能）

此技能采用两阶段架构：

1. 主技能（当前上下文）：理解用户问题 → 选择端点 → 组装 JSON 负载
2. 子技能（分叉上下文）：仅负责执行 HTTP 调用，避免对话历史浪费 token

执行方式

使用 Task 工具调用 tavily-fetcher 子技能，并传入命令和 JSON（stdin）：

Task parameters:
- subagent_type: Bash
- description: "Call Tavily API"
- prompt: cat <<'JSON' | node .claude/skills/tavily-web/tavily-api.cjs <search|extract|crawl|map|research>
  { ...payload... }
  JSON

负载示例（基于提供的 curl）

1）搜索网页

cat <<'JSON' | node .claude/skills/tavily-web/tavily-api.cjs search
{
  "query": "who is Leo Messi?",
  "auto_parameters": false,
  "topic": "general",
  "search_depth": "basic",
  "chunks_per_source": 3,
  "max_results": 1,
  "time_range": null,
  "start_date": "2025-02-09",
  "end_date": "2025-12-29",
  "include_answer": false,
  "include_raw_content": false,
  "include_images": false,
  "include_image_descriptions": false,
  "include_favicon": false,
  "include_domains": [],
  "exclude_domains": [],
  "country": null,
  "include_usage": false
}
JSON

2）提取网页内容

cat <<'JSON' | node .claude/skills/tavily-web/tavily-api.cjs extract
{
  "urls": "https://en.wikipedia.org/wiki/Artificial_intelligence",
  "query": "<string>",
  "chunks_per_source": 3,
  "extract_depth": "basic",
  "include_images": false,
  "include_favicon": false,
  "format": "markdown",
  "timeout": "None",
  "include_usage": false
}
JSON

3）抓取网页

cat <<'JSON' | node .claude/skills/tavily-web/tavily-api.cjs crawl
{
  "url": "docs.tavily.com",
  "instructions": "Find all pages about the Python SDK",
  "chunks_per_source": 3,
  "max_depth": 1,
  "max_breadth": 20,
  "limit": 50,
  "select_paths": null,
  "select_domains": null,
  "exclude_paths": null,
  "exclude_domains": null,
  "allow_external": true,
  "include_images": false,
  "extract_depth": "basic",
  "format": "markdown",
  "include_favicon": false,
  "timeout": 150,
  "include_usage": false
}
JSON

4）映射网页

cat <<'JSON' | node .claude/skills/tavily-web/tavily-api.cjs map
{
  "url": "docs.tavily.com",
  "instructions": "Find all pages about the Python SDK",
  "max_depth": 1,
  "max_breadth": 20,
  "limit": 50,
  "select_paths": null,
  "select_domains": null,
  "exclude_paths": null,
  "exclude_domains": null,
  "allow_external": true,
  "timeout": 150,
  "include_usage": false
}
JSON

5）创建研究任务

cat <<'JSON' | node .claude/skills/tavily-web/tavily-api.cjs research
{
  "input": "What are the latest developments in AI?",
  "model": "auto",
  "stream": false,
  "output_schema": {
    "properties": {
      "company": {
        "type": "string",
        "description": "The name of the company"
      },
      "key_metrics": {
        "type": "array",
        "description": "List of key performance metrics",
        "items": {
          "type": "string"
        }
      },
      "financial_details": {
        "type": "object",
        "description": "Detailed financial breakdown",
        "properties": {
          "operating_income": {
            "type": "number",
            "description": "Operating income for the period"
          }
        }
      }
    },
    "required": [
      "company"
    ]
  },
  "citation_format": "numbered"
}
JSON

环境变量与 API 密钥

配置 API 密钥有两种方式（优先级：环境变量 > .env）：

1. 环境变量：TAVILY_API_KEY
2. .env 文件：放置于 .claude/skills/tavily-web/.env，可从 .env.example 复制