web-research

作者: langchain-ai

協調多來源網路研究,透過委派子代理、綜合發現並產出附有引用的報告。將研究問題拆解為2至5個明確的子主題,建立研究計畫檔案,並平行啟動最多3個子代理以進行高效調查。每個子代理針對每個子主題執行3至5次網路搜尋,並將發現結果(包含關鍵事實、引用語句及來源網址)寫入本地檔案。最後透過讀取本地發現檔案,整合各子主題的見解,綜合產出結果。

npx skills add https://github.com/langchain-ai/deepagents --skill web-research

Web Research Skill

Research Process

Step 1: Create and Save Research Plan

Before delegating to subagents, you MUST:

  1. Create a research folder - Organize all research files in a dedicated folder relative to the current working directory:

    mkdir research_[topic_name]
    

    This keeps files organized and prevents clutter in the working directory.

  2. Analyze the research question - Break it down into distinct, non-overlapping subtopics

  3. Write a research plan file - Use the write_file tool to create research_[topic_name]/research_plan.md containing:

    • The main research question
    • 2-5 specific subtopics to investigate
    • Expected information from each subtopic
    • How results will be synthesized

Planning Guidelines:

  • Simple fact-finding: 1-2 subtopics
  • Comparative analysis: 1 subtopic per comparison element (max 3)
  • Complex investigations: 3-5 subtopics

Step 2: Delegate to Research Subagents

For each subtopic in your plan:

  1. Use the task tool to spawn a research subagent with:

    • Clear, specific research question (no acronyms)
    • Instructions to write findings to a file: research_[topic_name]/findings_[subtopic].md
    • Budget: 3-5 web searches maximum
  2. Run up to 3 subagents in parallel for efficient research

Subagent Instructions Template:

Research [SPECIFIC TOPIC]. Use the web_search tool to gather information.
After completing your research, use write_file to save your findings to research_[topic_name]/findings_[subtopic].md.
Include key facts, relevant quotes, and source URLs.
Use 3-5 web searches maximum.

Step 3: Synthesize Findings

After all subagents complete:

  1. Review the findings files that were saved locally:

    • First run list_files research_[topic_name] to see what files were created
    • Then use read_file with the file paths (e.g., research_[topic_name]/findings_*.md)
    • Important: Use read_file for LOCAL files only, not URLs
  2. Synthesize the information - Create a comprehensive response that:

    • Directly answers the original question
    • Integrates insights from all subtopics
    • Cites specific sources with URLs (from the findings files)
    • Identifies any gaps or limitations
  3. Write final report (optional) - Use write_file to create research_[topic_name]/research_report.md if requested

Note: If you need to fetch additional information from URLs, use the fetch_url tool, not read_file.

Best Practices

  • Plan before delegating - Always write research_plan.md first
  • Clear subtopics - Ensure each subagent has distinct, non-overlapping scope
  • File-based communication - Have subagents save findings to files, not return them directly
  • Systematic synthesis - Read all findings files before creating final response
  • Stop appropriately - Don't over-research; 3-5 searches per subtopic is usually sufficient

來自 langchain-ai 的更多技能

deepagents-thread-inspector
langchain-ai
檢查並解釋本地 Deep Agents Code SQLite 工作階段儲存庫中的對話。當 LangSmith 追蹤工具不可用時作為備用方案,用於…
deepagents-python-quickstart
langchain-ai
按照官方快速入門,在 Python 中搭建一個最小的本地 Deep Agent,使用提供者原生的網路搜尋而非 Tavily。當使用者想要……時使用。
deepagents-typescript-quickstart
langchain-ai
按照官方快速入門指南,以 TypeScript 搭建一個最小的本地 Deep Agent,使用供應商原生的網路搜尋而非 Tavily。當使用者……時使用
eval-engineering
langchain-ai
反覆檢查 agent 儲存庫與使用者提供的可選追蹤資料,訪談使用者,並逐一建立、執行及稽核 Harbor evals。用於……
LangChain RAG Pipeline
langchain-ai
在構建任何檢索增強生成(RAG)系統時,請調用此技能。涵蓋文檔加載器、遞迴字符文本分割器、嵌入(OpenAI)等。
LangChain Structured Output & HITL
langchain-ai
langchain-structured-output-&-hitl — 一個可安裝的 AI 代理技能,由 langchain-ai/langchain-skills 發布。
LangSmith Datasets
langchain-ai
當從追蹤建立評估資料集,或將資料集上傳至 LangSmith,或查詢資料集時,請調用此技能。涵蓋資料集類型(final_response、…)
langsmith-evaluator
langchain-ai
在為 LangSmith 建立評估管道時,請調用此技能。涵蓋三個核心組件:(1) 建立評估器 - LLM 作為評審、自訂程式碼;(2)…