AgentSkillsCN

web-search

针对只需聚焦事实查询的简单问题,采用快速网络搜索方法。由研究员的复杂度路由系统加载,适用于仅需 1–3 次精准检索的轻量级问题。

SKILL.md
--- frontmatter
name: web-search
description: Quick web search methodology for focused factual lookups. Loaded by the researcher's complexity router for simple questions requiring 1-3 targeted queries.

Quick Web Search Methodology

This skill provides the methodology for quick, focused web searches. Use when the complexity router classifies a question as a simple factual lookup, specific API/syntax question, or single-topic query.

When This Skill Applies

  • Simple factual lookups (e.g., "What's the default port for Redis?")
  • Specific API or syntax questions (e.g., "How to use useEffect cleanup?")
  • Single-topic queries with clear, direct answers
  • Version/compatibility checks
  • Error message lookups

Workflow

Step 1: Formulate Queries (1-3 queries max)

Craft 1-3 precise, targeted search queries. Each query should approach the topic from a slightly different angle:

javascript
web-research_multi_search({
  queries: [
    "exact technical question with key terms",
    "alternative phrasing or related aspect"  // optional 2nd query
  ],
  results_per_query: 5
})

Query crafting rules:

  • Use specific technical terms, not natural language
  • Include version numbers when relevant (e.g., "React 19 useEffect")
  • Include error codes/messages verbatim when debugging
  • Max 3 queries — if you need more, this should be deep research

Step 2: Scan Snippets

Review the search result snippets and titles. Identify the 2-3 most relevant URLs:

  • Official documentation (highest priority)
  • Stack Overflow answers with high votes
  • Recent blog posts from known authors
  • GitHub issues/discussions

Step 3: Fetch Top Pages (2-3 pages max)

javascript
web-research_fetch_pages({
  urls: ["top-result-url", "second-result-url"],
  max_chars: 10000  // shorter limit for quick searches
})

Page selection rules:

  • Prefer official docs over third-party
  • Prefer recent content over old
  • Skip pages whose snippets already answered the question
  • Max 3 pages — if you need more, this should be deep research

Step 4: Synthesize Answer

Combine findings into a concise response following the researcher's output format:

  • TL;DR (1-3 sentences)
  • Direct answer with code example if applicable
  • Source citations
  • Effort estimate

Quality Bar

  • Sufficient sources: 1-3 authoritative sources
  • Answer length: Concise — typically under 300 words
  • Citations: Every claim backed by a source
  • Confidence: High (clear answer found) or redirect to deep research if ambiguous

Escalation to Deep Research

If during quick search you discover:

  • The question is more complex than initially assessed
  • Multiple conflicting answers exist
  • The topic requires comparative analysis
  • More than 3 sources are needed for a complete answer

Stop and escalate: Load the deep-research skill instead and restart with the deep methodology.

Anti-Patterns

  • Over-researching: Don't fetch 10 pages for a simple question
  • Query sprawl: Don't issue 5+ queries — that's deep research territory
  • Skipping snippets: Always scan snippets first — the answer might be right there
  • Ignoring official docs: If official docs have the answer, don't keep searching