Setup
On first use, read setup.md to define activation behavior and preferred engine order, then store the preference profile.
When to Use
Use this when one engine is not enough and you need broader coverage, faster cross-checking, and cleaner verification.
Architecture
Store minimal preferences in ~/multi-engine-web-search/. See memory-template.md.
CODEBLOCK0
Quick Reference
| Topic | File |
|---|
| Setup flow and activation policy | INLINECODE3 |
| Minimal memory schema |
memory-template.md |
Search Engines
Global Engines (12)
- - Google: INLINECODE5
- Google HK: INLINECODE6
- Bing: INLINECODE7
- Yahoo: INLINECODE8
- DuckDuckGo: INLINECODE9
- Brave: INLINECODE10
- Startpage: INLINECODE11
- Qwant: INLINECODE12
- Ecosia: INLINECODE13
- Mojeek: INLINECODE14
- Swisscows: INLINECODE15
- AOL Search: INLINECODE16
Regional Engines (9)
- - Baidu: INLINECODE17
- Bing CN: INLINECODE18
- Bing INT (CN endpoint): INLINECODE19
- Sogou: INLINECODE20
- 360 Search: INLINECODE21
- Yandex: INLINECODE22
- Naver: INLINECODE23
- Seznam: INLINECODE24
- CocCoc: INLINECODE25
Knowledge and Developer Engines (6)
- - WolframAlpha: INLINECODE26
- Wikipedia: INLINECODE27
- GitHub Search: INLINECODE28
- Stack Overflow Search: INLINECODE29
- Semantic Scholar: INLINECODE30
- PubMed: INLINECODE31
Engine Shortcuts
| Shortcut | Engine |
|---|
| INLINECODE32 | Google |
| INLINECODE33 |
Google HK |
|
!b | Bing |
|
!y | Yahoo |
|
!ddg | DuckDuckGo |
|
!br | Brave |
|
!sp | Startpage |
|
!qw | Qwant |
|
!eco | Ecosia |
|
!mj | Mojeek |
|
!sw | Swisscows |
|
!aol | AOL Search |
|
!ba | Baidu |
|
!bcn | Bing CN |
|
!sg | Sogou |
|
!360 | 360 Search |
|
!ydx | Yandex |
|
!nav | Naver |
|
!sz | Seznam |
|
!cc | CocCoc |
|
!wa | WolframAlpha |
|
!w | Wikipedia |
|
!gh | GitHub Search |
|
!so | Stack Overflow |
|
!ss | Semantic Scholar |
|
!pm | PubMed |
Quick Examples
CODEBLOCK1
Advanced Operators
| Operator | Example | Purpose |
|---|
| INLINECODE58 | INLINECODE59 | Limit to one domain |
| INLINECODE60 |
filetype:pdf model card | Find specific formats |
|
"" |
"context window" | Exact phrase |
|
- |
python -snake | Exclude noisy term |
|
OR |
llama OR mistral | Alternative terms |
|
intitle: |
intitle:benchmark llm | Keyword in page title |
|
inurl: |
inurl:docs authentication | Keyword in URL |
Time Filters
| Pattern | Example | Purpose |
|---|
| INLINECODE72 | Google past hour | Breaking updates |
| INLINECODE73 |
Google past day | Daily changes |
|
tbs=qdr:w | Google past week | Weekly updates |
|
before: |
ai act before:2026-03-01 | Upper date bound |
|
after: |
ai act after:2025-01-01 | Lower date bound |
Core Rules
1. Check Preferences First
Before searching, read memory preferences:
- - activation mode (always, on-request, or mixed)
- preferred engine order
- blocked engines
- output style (fast summary or evidence-heavy)
2. Always Use Multi-Engine Batches
Run at least 3 engines per request: one mainstream, one privacy engine, and one alternate index.
3. Add a Contradiction Query
For every important claim, run one query that looks for disagreement, failures, or corrections.
4. Prefer Primary Sources
If sources conflict, prioritize original docs, direct announcements, and first-party datasets.
5. Verify Date Context
For current topics, verify publication date and event date before final conclusions.
6. Return Evidence First
Output must include direct answer, best links, and confidence in one concise block.
Common Traps
- - Using only one engine and assuming top results are correct.
- Forgetting contradiction queries on high-impact decisions.
- Treating copied news rewrites as independent confirmation.
- Ignoring date filters for rapidly changing topics.
- Returning link dumps without a clear recommendation.
External Endpoints
| Endpoint Family | Data Sent | Purpose |
|---|
| Public search engines listed above | query text | Multi-engine retrieval and cross-checking |
| Knowledge and developer engines listed above |
query text | Technical, scientific, and code verification |
No other data is sent externally.
Related Skills
Install with
clawhub install <slug> if user confirms:
- -
analysis - Turn search findings into clear conclusions - INLINECODE81 - Compare options side by side with tradeoffs
- INLINECODE82 - Inspect pages deeply after initial retrieval
- INLINECODE83 - Expand into long-form investigations
- INLINECODE84 - Build custom search backends when needed
Feedback
- - If useful: INLINECODE85
- Stay updated: INLINECODE86
设置
首次使用时,请阅读 setup.md 以定义激活行为及首选引擎顺序,然后存储偏好配置文件。
使用时机
当单一引擎不足以满足需求,需要更广泛的覆盖范围、更快速的交叉验证以及更清晰的核实结果时使用。
架构
将最小偏好设置存储在 ~/multi-engine-web-search/ 目录下。详见 memory-template.md。
text
~/multi-engine-web-search/
-- memory.md # 激活模式、引擎优先级、屏蔽引擎及输出风格
快速参考
| 主题 | 文件 |
|---|
| 设置流程与激活策略 | setup.md |
| 最小内存模式 |
memory-template.md |
搜索引擎
全球引擎(12个)
- - Google:https://www.google.com/search?q={keyword}
- Google HK:https://www.google.com.hk/search?q={keyword}
- Bing:https://www.bing.com/search?q={keyword}
- Yahoo:https://search.yahoo.com/search?p={keyword}
- DuckDuckGo:https://duckduckgo.com/html/?q={keyword}
- Brave:https://search.brave.com/search?q={keyword}
- Startpage:https://www.startpage.com/sp/search?query={keyword}
- Qwant:https://www.qwant.com/?q={keyword}
- Ecosia:https://www.ecosia.org/search?q={keyword}
- Mojeek:https://www.mojeek.com/search?q={keyword}
- Swisscows:https://swisscows.com/web?query={keyword}
- AOL Search:https://search.aol.com/aol/search?q={keyword}
区域引擎(9个)
- - Baidu:https://www.baidu.com/s?wd={keyword}
- Bing CN:https://cn.bing.com/search?q={keyword}&ensearch=0
- Bing INT(CN 端点):https://cn.bing.com/search?q={keyword}&ensearch=1
- Sogou:https://www.sogou.com/web?query={keyword}
- 360 Search:https://www.so.com/s?q={keyword}
- Yandex:https://yandex.com/search/?text={keyword}
- Naver:https://search.naver.com/search.naver?query={keyword}
- Seznam:https://search.seznam.cz/?q={keyword}
- CocCoc:https://coccoc.com/search?query={keyword}
知识与开发者引擎(6个)
- - WolframAlpha:https://www.wolframalpha.com/input?i={keyword}
- Wikipedia:https://en.wikipedia.org/w/index.php?search={keyword}
- GitHub Search:https://github.com/search?q={keyword}
- Stack Overflow Search:https://stackoverflow.com/search?q={keyword}
- Semantic Scholar:https://www.semanticscholar.org/search?q={keyword}
- PubMed:https://pubmed.ncbi.nlm.nih.gov/?term={keyword}
引擎快捷键
Google HK |
| !b | Bing |
| !y | Yahoo |
| !ddg | DuckDuckGo |
| !br | Brave |
| !sp | Startpage |
| !qw | Qwant |
| !eco | Ecosia |
| !mj | Mojeek |
| !sw | Swisscows |
| !aol | AOL Search |
| !ba | Baidu |
| !bcn | Bing CN |
| !sg | Sogou |
| !360 | 360 Search |
| !ydx | Yandex |
| !nav | Naver |
| !sz | Seznam |
| !cc | CocCoc |
| !wa | WolframAlpha |
| !w | Wikipedia |
| !gh | GitHub Search |
| !so | Stack Overflow |
| !ss | Semantic Scholar |
| !pm | PubMed |
快速示例
javascript
// 1) 基础多引擎搜索
web_fetch({url: https://www.google.com/search?q=llm+agent+framework})
web_fetch({url: https://duckduckgo.com/html/?q=llm+agent+framework})
web_fetch({url: https://search.brave.com/search?q=llm+agent+framework})
// 2) 特定站点验证
web_fetch({url: https://www.bing.com/search?q=site:github.com+fastapi+auth})
// 3) 文件类型查询
web_fetch({url: https://www.google.com/search?q=rag+evaluation+filetype:pdf})
// 4) 时效性查询
web_fetch({url: https://www.google.com/search?q=ai+policy+2026&tbs=qdr:m})
// 5) 知识查询
web_fetch({url: https://www.wolframalpha.com/input?i=150+USD+to+EUR})
高级运算符
| 运算符 | 示例 | 用途 |
|---|
| site: | site:arxiv.org agentic ai | 限定在单一域名内 |
| filetype: |
filetype:pdf model card | 查找特定格式 |
| | context window | 精确短语匹配 |
| - | python -snake | 排除干扰词 |
| OR | llama OR mistral | 替代术语 |
| intitle: | intitle:benchmark llm | 关键词在页面标题中 |
| inurl: | inurl:docs authentication | 关键词在URL中 |
时间过滤器
| 模式 | 示例 | 用途 |
|---|
| tbs=qdr:h | Google 过去一小时 | 突发更新 |
| tbs=qdr:d |
Google 过去一天 | 每日变化 |
| tbs=qdr:w | Google 过去一周 | 每周更新 |
| before: | ai act before:2026-03-01 | 日期上限 |
| after: | ai act after:2025-01-01 | 日期下限 |
核心规则
1. 优先检查偏好设置
搜索前,读取内存偏好设置:
- - 激活模式(始终、按需或混合)
- 首选引擎顺序
- 屏蔽引擎
- 输出风格(快速摘要或证据密集型)
2. 始终使用多引擎批次
每次请求至少运行3个引擎:一个主流引擎、一个隐私引擎和一个备用索引。
3. 添加矛盾查询
对于每个重要主张,运行一次寻找分歧、失败或纠正的查询。
4. 优先使用原始来源
如果来源冲突,优先选择原始文档、直接公告和第一方数据集。
5. 验证日期上下文
对于当前话题,在最终结论前验证发布日期和事件日期。
6. 先返回证据
输出必须包含直接答案、最佳链接和置信度,以简洁的区块呈现。
常见陷阱
- - 仅使用一个引擎并假设顶部结果正确。
- 在高影响决策中忘记矛盾查询。
- 将转载的新闻改写视为独立确认。
- 忽略快速变化话题的日期过滤器。
- 返回链接堆砌而无明确建议。
外部端点
| 端点系列 | 发送数据 | 用途 |
|---|
| 上述列出的公共搜索引擎 | 查询文本 | 多引擎检索与交叉验证 |
| 上述列出的知识与开发者引擎 |
查询文本 | 技术、科学及代码验证 |
不向外部发送其他数据。
相关技能
如果用户确认,使用 clawhub install 安装:
- - analysis - 将搜索结果转化为清晰结论
- compare - 并排比较选项并权衡利弊
- web - 初始检索后深入检查页面
- in-depth-research - 扩展为长篇调研
- elasticsearch - 在需要时构建自定义搜索后端
反馈
- - 如果有用:clawhub star multi-engine-web-search
- 保持更新:clawhub sync