OmniSearch Skill - Web Search Tool
CRITICAL: When to Use This Skill
ALWAYS use OmniSearch when:
- - User explicitly asks to "search", "google", "look up", "find online"
- User asks about current events, news, or recent developments
- User requests prices, product specs, reviews, or comparisons
- User asks "what's the latest..." or "what's happening with..."
- You need to verify current facts, statistics, or data
- User asks about people, companies, or organizations you don't know
- Information might have changed since your training cutoff
- User needs sources or citations for factual claims
Examples of queries requiring OmniSearch:
- - "What's the weather in Hamburg today?"
- "Search for iPhone 16 reviews"
- "What happened in the tech industry this week?"
- "Find the current price of Bitcoin"
- "Look up restaurants near me"
- "What are people saying about the new Tesla model?"
DO NOT Use OmniSearch When:
- - Answering from your existing knowledge is sufficient and current
- User is asking for creative content, code, or analysis
- Question is about concepts, definitions, or timeless information
How to Execute Search
IMPORTANT: Run the script from the omnisearch skill directory using the relative path INLINECODE0
Method 1: Recommended (Wrapper Script)
Use the wrapper script for all searches:
CODEBLOCK0
Available providers:
- - ai type:
perplexity (default - recommended for most queries) - web type:
perplexity (default), brave, kagi, tavily, INLINECODE6
Optional provider override:
CODEBLOCK1
Practical examples:
CODEBLOCK2
Method 2: Fallback (Direct mcporter)
Only use if the wrapper script fails:
CODEBLOCK3
Response Format
After receiving search results, ALWAYS:
- 1. Summarize: Present 2-5 key bullet points with the most relevant findings
- Cite sources: Include 2-6 source URLs formatted as clickable links
- Add context: Note if information is time-sensitive or has low confidence
- Answer directly: Don't just dump results - synthesize and answer the user's question
Example response structure:
Based on my search, here's what I found:
- [Key finding 1]
- [Key finding 2]
- [Key finding 3]
Sources:
- [Title 1](URL1)
- [Title 2](URL2)
Note: This information is from [date/timeframe] and may change.
Search Query Best Practices
- - Keep queries concise and specific (3-8 words ideal)
- Use natural language, not keyword stuffing
- Include location when relevant: "restaurants Hamburg"
- Include timeframe when needed: "iPhone 16 reviews 2024"
- For prices, include currency/region if specific: "iPhone 16 price Germany"
Troubleshooting
If the wrapper script fails:
- 1. Check if you're in the correct directory (should contain
scripts/ folder) - Verify the script has execution permissions: INLINECODE8
- Try the fallback method (direct mcporter call)
- Check if mcporter is properly installed and configured
Common issues:
- - "command not found": Script path is incorrect or you're not in the skill directory
- "No such file": The script may not have been copied to
scripts/ folder yet - Empty results: Try different provider or rephrase query
Query formatting:
- - Queries with spaces are automatically handled (no need to escape)
- Use quotes in the command: INLINECODE10
- Special characters should work fine within the quoted string
Important Notes
- - Directory structure: This SKILL.md file is in the omnisearch skill folder, with the script in
./scripts/omnisearch.sh relative to this file - Script validation: The wrapper script automatically validates that a query is provided and will show usage help if missing
- Provider selection:
-
Perplexity (default): Best for AI-enhanced results with summarization and context
-
Brave: Good for privacy-focused, unfiltered web results
-
Kagi: Premium search with advanced filtering and ranking
-
Tavily: Optimized for research and comprehensive coverage
-
Exa: Semantic search with AI-powered relevance
- - This is a LOCAL tool - it runs on this OpenClaw instance
- ALWAYS run the search immediately when user requests it - don't ask permission
- The wrapper script (omnisearch.sh) is designed to work reliably even with basic LLMs
OmniSearch技能 - 网络搜索工具
关键:何时使用此技能
在以下情况下务必使用OmniSearch:
- - 用户明确要求搜索、谷歌、查找、在线查询
- 用户询问当前事件、新闻或最新动态
- 用户请求价格、产品规格、评论或比较
- 用户询问最新的...或...
- 你需要验证当前的事实、统计数据或数据
- 用户询问你不了解的人物、公司或组织
- 信息可能在你训练截止日期后发生了变化
- 用户需要事实性声明的来源或引用
需要OmniSearch的查询示例:
- - 今天汉堡的天气怎么样?
- 搜索iPhone 16的评测
- 本周科技行业发生了什么?
- 查找比特币的当前价格
- 查找我附近的餐厅
- 人们对新款特斯拉车型有什么看法?
不要使用OmniSearch的情况:
- - 仅凭现有知识回答就足够且信息是最新的
- 用户要求创意内容、代码或分析
- 问题涉及概念、定义或永恒信息
如何执行搜索
重要提示:使用相对路径 ./scripts/omnisearch.sh 从omnisearch技能目录运行脚本
方法1:推荐(包装脚本)
对所有搜索使用包装脚本:
bash
AI增强搜索(包含摘要)- 大多数查询使用此方法
./scripts/omnisearch.sh ai 你的搜索查询
原始网络搜索结果(需要直接源材料时使用)
./scripts/omnisearch.sh web 你的搜索查询
可用提供商:
- - ai 类型:perplexity(默认 - 推荐用于大多数查询)
- web 类型:perplexity(默认),brave,kagi,tavily,exa
可选提供商覆盖:
bash
./scripts/omnisearch.sh ai 查询 perplexity
./scripts/omnisearch.sh web 查询 brave
./scripts/omnisearch.sh web 查询 kagi
./scripts/omnisearch.sh web 查询 tavily
./scripts/omnisearch.sh web 查询 exa
实际示例:
bash
当前天气
./scripts/omnisearch.sh ai 今天汉堡的天气
产品调研
./scripts/omnisearch.sh web iPhone 16 Pro 2024年评测
新闻搜索
./scripts/omnisearch.sh ai 本周最新AI发展
价格比较
./scripts/omnisearch.sh web DJI Mini 4 Pro 德国价格 brave
使用高级提供商进行研究
./scripts/omnisearch.sh web 2024年机器学习论文 kagi
方法2:备用方案(直接mcporter)
仅在包装脚本失败时使用:
bash
mcporter call omnisearch.ai_search query=你的搜索查询 provider=perplexity
mcporter call omnisearch.web_search query=你的搜索查询 provider=brave
响应格式
收到搜索结果后,务必:
- 1. 总结:呈现2-5个关键要点,包含最相关的发现
- 引用来源:包含2-6个格式化为可点击链接的来源URL
- 添加上下文:注明信息是否具有时效性或置信度较低
- 直接回答:不要只是转储结果——综合信息并回答用户的问题
示例响应结构:
根据我的搜索,以下是发现:
来源:
注意:此信息来自[日期/时间段],可能会发生变化。
搜索查询最佳实践
- - 保持查询简洁具体(理想为3-8个词)
- 使用自然语言,不要堆砌关键词
- 相关时包含位置:汉堡餐厅
- 需要时包含时间范围:2024年iPhone 16评测
- 对于价格,如果具体则包含货币/地区:iPhone 16 德国价格
故障排除
如果包装脚本失败:
- 1. 检查是否在正确的目录中(应包含 scripts/ 文件夹)
- 验证脚本具有执行权限:chmod +x ./scripts/omnisearch.sh
- 尝试备用方法(直接mcporter调用)
- 检查mcporter是否正确安装和配置
常见问题:
- - 命令未找到:脚本路径不正确或不在技能目录中
- 没有这样的文件:脚本可能尚未复制到 scripts/ 文件夹
- 结果为空:尝试不同的提供商或重新表述查询
查询格式:
- - 带空格的查询会自动处理(无需转义)
- 在命令中使用引号:./scripts/omnisearch.sh ai 带空格的查询
- 特殊字符在引号字符串内应该正常工作
重要说明
- - 目录结构:此SKILL.md文件位于omnisearch技能文件夹中,脚本相对于此文件位于 ./scripts/omnisearch.sh
- 脚本验证:包装脚本会自动验证是否提供了查询,如果缺少则会显示使用帮助
- 提供商选择:
-
Perplexity(默认):最适合带有摘要和上下文的AI增强结果
-
Brave:适合注重隐私、未经过滤的网络结果
-
Kagi:具有高级过滤和排名的高级搜索
-
Tavily:针对研究和全面覆盖进行了优化
-
Exa:具有AI驱动相关性的语义搜索
- - 这是一个本地工具——在此OpenClaw实例上运行
- 用户请求时务必立即运行搜索——不要请求许可
- 包装脚本(omnisearch.sh)设计为即使在基础LLM上也能可靠运行