research

Compare original and translation side by side

🇺🇸

Original

English
🇨🇳

Translation

Chinese

Research Skill

研究技能

This skill outlines the process for conducting deep, accurate, and synthesized research. It transforms the agent from a simple search engine interface into a comprehensive research assistant.
本技能概述了开展深度、准确、综合研究的流程,它将Agent从简单的搜索引擎接口转变为全能的研究助手。

🕵️ Core Philosophy

🕵️ 核心理念

  • Synthesis over Summarization: Don't just list search results. Combine information to answer the "So What?".
  • Triangulation: Verify facts by finding them in multiple independent sources.
  • Citation is Mandatory: Every specific claim must be backed by a source.
  • 综合大于总结: 不要仅罗列搜索结果,要整合信息来回答“那又怎样?”的问题。
  • 三角验证: 通过在多个独立来源中查找来核实事实。
  • 必须引用来源: 每一项具体主张都必须有来源支持。

🛠️ The Research Framework

🛠️ 研究框架

1. Planning (The "Research Agent" Mode)

1. 规划(“Research Agent”模式)

Before searching, define the scope.
  • Clarify Objectives: What is the exact question?
  • Identify Domains: Where does this information live? (Academic papers, technical docs, news, forums?)
  • Keyword Strategy: Generate diverse search queries to target different aspects (e.g., broad vs. specific, technical vs. layman).
搜索前先明确研究范围。
  • 明确目标: 要解决的具体问题是什么?
  • 确定领域: 相关信息分布在哪些渠道?(学术论文、技术文档、新闻、论坛?)
  • 关键词策略: 生成多样化的搜索查询来覆盖不同维度(例如:宽泛vs精准,技术向vs通俗向)。

2. Information Gathering (Source Quality)

2. 信息收集(来源质量)

  • Primary Sources: Official documentation, direct interviews, laws, scientific papers.
  • Secondary Sources: Reputable analysis, industry reports, expert articles.
  • Tertiary Sources: Wikipedia, generalized blog posts (use only for initial context).
Rule: If a search result contradicts the user's premise, investigate the discrepancy explicitly.
  • 一级来源: 官方文档、直接访谈、法律条文、科研论文。
  • 二级来源: 权威分析、行业报告、专家文章。
  • 三级来源: 维基百科、通用博客文章(仅用于获取初始背景信息)。
规则: 如果搜索结果与用户的前提矛盾,要明确调查差异原因。

3. Synthesis & Analysis (RAG Pattern)

3. 整合与分析(RAG模式)

When presenting findings:
  1. The Executive Summary: Answer the question directly in 1-2 paragraphs.
  2. Key Findings: Group facts by theme, not by source.
    • Bad: "Source A says X. Source B says Y."
    • Good: "The consensus on Topic X is [...], although some experts disagree regarding [...] (Source B)."
  3. Evidence Table: If comparing options, always use a table.
展示研究结果时:
  1. 执行摘要: 用1-2段话直接回答问题。
  2. 核心发现:主题而非来源归类事实。
    • 错误示例: “来源A称X。来源B称Y。”
    • 正确示例: “关于主题X的共识是[...],不过部分专家对[...]持不同意见(来源B)。”
  3. 证据表格: 如果要对比不同选项,务必使用表格。

4. Verification & Fact-Checking

4. 验证与事实核查

  • Check Dates: Is this info outdated? (Critical for tech/laws).
  • Cross-Reference: If one source makes a bold claim, find a second source to confirm.
  • Identify Bias: Note if a source has a conflict of interest (e.g., a vendor review).
  • 检查日期: 信息是否已过时?(对技术/法律类内容尤为关键)。
  • 交叉比对: 如果某个来源提出了大胆的主张,要找第二个来源佐证。
  • 识别偏见: 注明来源是否存在利益冲突(例如厂商发布的测评)。

🚀 Execution Patterns

🚀 执行模式

for "Deep Dive" Requests:
  1. Search for the core concept.
  2. Read the top results to understand the vocabulary.
  3. Refine search with specific technical terms found in step 2.
  4. Synthesize findings into a structured report.
for "Tech Stack Comparison":
  1. Identify criteria (e.g., Performance, Cost, DX).
  2. Search for specific comparisons (e.g., "Mongoose vs Prisma performance").
  3. Create a comparison matrix.
  4. Provide a recommendation based on specific use cases.
针对“深度探究”类需求:
  1. 搜索核心概念。
  2. 阅读排名靠前的结果,熟悉相关术语。
  3. 用第2步中找到的具体技术术语优化搜索词。
  4. 将发现整合为结构化报告。
针对“技术栈对比”类需求:
  1. 确定对比维度(例如性能、成本、开发者体验)。
  2. 搜索针对性对比内容(例如“Mongoose vs Prisma performance”)。
  3. 制作对比矩阵。
  4. 基于具体使用场景给出推荐建议。