web-search

name	web_search
description	Search the web using search engines (Baidu, Google, Bing) with result ranking and snippet extraction
category	search
version	1.2.0
author	DocSys Team
permissions	["read"]
tags	["web","search","search-engine","baidu","google","research"]

Web Search

Search the web using search engines when local documents don't have relevant results. Supports multi-engine search, result ranking, snippet extraction, and citation guidance.

Triggers

Web search, 网络搜索, 网上搜索, 百度搜索, search the web, web-search, search web, 搜索网页, search online, google search, 网上查, 在线搜索, 搜索互联网, 帮我查一下, 帮我搜一下, look up online, find on the internet

Workflow

1. Validate & reformulate query: Ensure query is non-empty. If vague or ambiguous → try to reformulate before searching (e.g., "latest iPhone" → "iPhone latest release date 2024"). If truly empty → report error.
2. Determine engine: Infer from keywords ("百度" / "中文" → baidu, "google" / "英语" → google, "英文资料" → google/bing). Default: baidu. For international queries → recommend google.
3. ⚠️ Sensitive content check: If query involves medical/financial/legal advice → add disclaimer: "以下信息仅供参考，不构成专业建议。"
4. Call CLI: docsys web-search <query> [--engine <engine>] [--limit <N>]
5. Process results: For each result, extract: title (bold), URL, and a 1-2 sentence snippet showing why this result is relevant. Remove duplicate domains.
6. Rank and present: Order by relevance (title match > snippet relevance > domain authority). Show top 5-10 results with snippets.
7. Rate limit handling: If rate limited → wait 30s and retry once. If still blocked → suggest trying a different engine.
8. Citation guidance: After presenting results, optionally include: "来源: title" for each result the user references.
9. Next steps: If 0 results → suggest broadening query or trying a different engine. If few results → suggest related searches.

CLI Command

docsys web-search <query> [--engine baidu] [--limit 10]

Tip: For academic queries → use --engine google. For Chinese queries → baidu is best. For real-time stock/price data → results are snapshots only, not live data.

Parameters

Parameter	Type	Required	Description
query	string	yes	Search query
engine	string	no	Search engine: baidu (default), google, bing
limit	number	no	Max results (default: 10, max: 50)

Examples

Example 6 (Tip: engine selection)

User: "find academic papers on quantum computing" Skill triggers → Step 2: academic query → recommends --engine google → calls → results Output: Academic papers with snippets. Tip shown: "For academic queries → use --engine google."

Example 7 (Local-first guidance)

User: "search for project notes about Q3 planning" Skill triggers → Before searching: "This query sounds like it might be in your local repositories. Try search_doc or search_in_repo first. If no local results → proceed with web search." Output: Web search results only (after local search suggestion)

User: "search the web for Claude AI features" Skill triggers → Step 1: query clear → Step 2: engine=baidu (default) → Step 3: no sensitive content → calls: docsys web-search Claude AI features --limit 10 Step 5: extracts snippets from each result. Step 6: presents top results ranked. Output: [Result 1] Claude AI Official Site — https://claude.ai — Anthropic's next-generation AI assistant... [Result 2] Claude AI Features Overview — https://docs.anthropic.com — Comprehensive guide to Claude capabilities...

Example 2

User: "百度搜索 人工智能发展趋势" Skill triggers → Step 1: query clear → Step 2: engine=baidu (inferred) → Step 3: no sensitive content → calls: docsys web-search 人工智能发展趋势 --engine baidu --limit 10 Output: 搜索结果列表，含标题、URL、摘要

Example 3

User: "帮我查一下 特斯拉最新股价" Skill triggers → Step 1: query clear → Step 2: engine=baidu → Step 3: ⚠️ financial content → adds disclaimer → calls: docsys web-search 特斯拉最新股价 --limit 5 Output: Results + "⚠️ 股价信息仅供参考，不构成投资建议。"

Example 4 (no results → next steps)

User: "search for xyzabc123 nonexistent query" Skill triggers → calls: docsys web-search xyzabc123 --limit 10 → 0 results → Step 9: suggests broadening query or trying different engine Output: No results found. Try: (1) broader keywords, (2) different engine (e.g. --engine google), (3) check spelling.

Example 5 (multi-engine)

User: "find academic papers on machine learning" Skill triggers → Step 1: query reformulated → Step 2: suggests --engine google (academic) → calls → presents ranked results Output: Academic results with snippets showing relevance to machine learning

Output Format

Success:

[Result 1] <title> — <URL>
  摘要: <1-2 sentence snippet explaining relevance>

[Result 2] <title> — <URL>
  摘要: <...>

⚠️ Sensitive queries add disclaimer: "以下信息仅供参考，不构成XX建议。"

Zero results: "No results found. Suggestions: (1) broaden keywords, (2) try --engine google, (3) check spelling."

Error: "Query is required", "Browser not available", "Rate limit exceeded"

Error Handling

Error	Cause
Query is required	Empty query
Browser not available	Playwright not initialized
Search engine blocked	Anti-bot protection
Rate limit exceeded	Too many searches
Connection timeout	Network issue
Zero results	Query too specific or engine blocked

Related Skills

If user searches for content that may be in local repos → suggest search_doc or search_in_repo first.

See references/ for related skills and API docs.