博查 Web 搜索工具,使用 Bocha Web Search API。适用于在线查询、事实核查、获取时效性信息以及提供带引用的回答。(Default web search tool using Bocha Web Search API. Use for online lookup, verification,...
---
name: bocha-web-search
description: 博查 Web 搜索工具,使用 Bocha Web Search API。适用于在线查询、事实核查、获取时效性信息以及提供带引用的回答。(Default web search tool using Bocha Web Search API. Use for online lookup, verification, time-sensitive information, and citation-based answers.)
license: MIT-0
homepage: https://open.bocha.cn
metadata:
openclaw:
emoji: "🔎"
requires:
env:
- BOCHA_API_KEY
primaryEnv: BOCHA_API_KEY
---
# Bocha Web Search
本 Skill 使用 Bocha Web Search API 进行联网搜索。
This Skill performs web searches using the Bocha Web Search API.
它的主要设计目标是:
It is designed to:
- 检索最新、最及时的信息
Retrieve up-to-date information
- 核实事实声明
Verify factual claims
- 提供基于可靠来源的回答
Provide source-backed answers
- 支持带有引用的回复
Support citation-based responses
此版本避免了特定于 shell 的指令和系统级文件操作,以确保与 ClawHub 等安全环境的兼容性。
This version avoids shell-specific instructions and system-level file operations to ensure compatibility with secure environments such as ClawHub.
---
## 何时使用此 Skill (When to Use This Skill)
当用户的请求出现以下情况时,请使用此 Skill:
Use this skill whenever the user request:
- 需要对话历史中不存在的信息
Requires information not present in the conversation
- 涉及具有时效性或不断变化的数据,如新闻、政策、财务数据、发布信息等
Involves time-sensitive or changing data
- 需要进行事实核查或验证
Requires fact-checking or verification
- 要求提供链接、信息来源或引用
Requests links, sources, or citations
- 提及特定的组织、事件、人物或产品,并询问具体的事实细节
Mentions a specific organization, event, person, or product and asks for factual details
如果不确定是否需要进行在线查询,请直接执行搜索。
If uncertain whether online lookup is required, perform a search.
---
## API 规范 (API Specification)
**请求端点 (Endpoint)**:
POST `https://api.bocha.cn/v1/web-search`
**请求头 (Headers)**:
```http
Authorization: Bearer <BOCHA_API_KEY>
Content-Type: application/json
```
**请求体参数 (Request Body)**:
| 参数名 | 类型 | 必填 | 说明 | Description |
|--------|------|------|------|-------------|
| query | string | 是 | 搜索关键词 | Search query string |
| freshness | string | 否 | 时间范围,可选值:`noLimit`(默认), `oneDay`, `oneWeek`, `oneMonth`, `oneYear` | Time filter |
| summary | boolean | 否 | 是否返回网页原文摘要,默认 `true` | Whether to include web original content |
| count | integer | 否 | 返回结果数量,默认 `10`,最大 `50` | Number of returned results |
### 工作流程 (Workflow)
#### 第一步:提取搜索参数 (Step 1: Extract Search Parameters)
从用户请求中提取搜索关键词和相关参数:
Extract search keywords and related parameters from the user's request:
1. `query`: 从用户问题中提取核心搜索词,需要保留“最近”、“今年”、“今年1月”等时间范围描述。
Extract core search terms from the user's question, preserving time range descriptions like "recently", "this year", "January this year", etc.
2. `freshness`: 根据用户描述的时间范围确定,默认选择 `noLimit`,以便 web-search 结合 `query` 中的时间范围描述自动改写最合适的 freshness 值。
Determine based on the user's described time range, defaulting to `noLimit` so web-search can automatically rewrite the most appropriate freshness value.
3. `summary`: 默认 `true`。
Default is `true`.
4. `count`: 默认 `10`,如用户明确要求更多结果则进行调整。
Default is `10`, adjust if the user explicitly requests more results.
#### 第二步:发起资源请求 (Step 2: Send API Request)
向博查 API 发起 POST 请求:
Send a POST request to the Bocha API:
```json
{
"query": "<USER_QUERY>",
"freshness": "noLimit",
"summary": true,
"count": 10
}
```
---
## 响应结构 (Response Structure)
搜索结果位于以下路径:
Search results are located at:
`data.webPages.value[]`
每个结果通常包含以下字段:
Each result typically contains:
- `name` (标题 / Title)
- `url` (链接 / URL)
- `snippet` (内容片段 / Snippet)
- `summary` (原始内容摘要 / Summary)
- `siteName` (站点名称 / Site Name)
- `datePublished` (发布日期 / Date Published)
---
## 引用规则 (Citation Rules - Mandatory)
在生成最终回答时,必须遵守以下规则:
When generating the final answer, you must adhere to the following rules:
1. 使用返回的来源来支撑你的事实性陈述。
Support factual statements using returned sources.
2. 按出现顺序分配引用编号:[1], [2], [3]
Assign citation numbers in order of appearance.
3. 在回答的末尾添加“参考资料”部分。
End with a "References" section.
4. 每个参考文献必须包含标题、链接和站点名称(如果可用)。
Each reference must include the title, URL, and site name (if available).
**输出格式示例 (Example output format)**:
回答 (Answer):
<你的回答,包含内联引用,如 [1], [2]。>
<Your answer with inline citations like [1], [2].>
参考资料 (References):
[1] `<title>`
`<url>`
来源 (Source): `<siteName>`
[2] `<title>`
`<url>`
来源 (Source): `<siteName>`
如果未找到可靠的来源,请回复:
If no reliable sources are found, reply with:
"未找到可靠的来源。(No reliable sources found.)"
---
## 错误处理 (Error Handling)
常见的 API 错误代码:
Common API error codes:
- 400: 请求错误 (Bad request)
- 401: API Key 无效 (Invalid API key)
- 403: 余额不足 (Insufficient balance)
- 429: 请求过于频繁 (Rate limit exceeded)
- 500: 服务器错误 (Server error)
如需调试,请使用 API 响应中的 `log_id`。
Use `log_id` from API responses for debugging if needed.
don't have the plugin yet? install it then click "run inline in claude" again.