academic-scientific-research

作者: tavily-ai

查找、篩選並綜合學術論文、科學文獻、技術報告、預印本、臨床或生物醫學出版物,以及圍繞某一…的證據。

npx skills add https://github.com/tavily-ai/use-case-skills --skill academic-scientific-research

Academic Scientific Research

Workflow

Use search to discover scholarly sources, extract to verify source content, map or crawl for known scholarly sites and collections, and research for full literature-review synthesis. Keep this skill focused on research planning, source selection, extraction targets, and evidence synthesis; execution mechanics should come from companion endpoint skills.

Treat the guidance below as base guidance; adapt it to the user's request when appropriate.

  • Translate the user's question into search terms, synonyms, key entities, and likely source domains.
  • Break broad questions into short subqueries under 400 characters: core concept, synonyms, method names, target population or data, benchmark or dataset, author/lab, and year range.
  • Prefer scholarly and official sources when available: preprint servers, PubMed/NIH pages, journals, conference proceedings, professional societies, standards bodies, and official technical reports.
  • Extract from the strongest selected pages after screening titles, snippets, source type, recency, and relevance.
  • Reserve broad research synthesis for full literature reviews, research landscapes, or multi-method comparisons.
  • Distinguish primary studies, review papers, preprints, editorials, guidelines, standards, and news coverage.

Research Budget

  • Start with a small focused search set covering the main concept, synonyms, methods, and source type.
  • Extract only the strongest scholarly sources before drafting the evidence brief.
  • Add more searches only for named gaps, such as missing review papers, missing recent work, or missing primary studies.
  • Do not use map unless a known lab, journal, conference, repository, or documentation site has buried pages.
  • Do not use crawl unless the user needs broad collection from a known proceedings, publication list, or docs section.

Capability Guidance

  • Use search first when the user needs papers, methods, benchmarks, authors, institutions, or recent work.
  • Use extract when comparing a small set of papers, reading abstracts, or verifying study claims.
  • Use map only when a known lab, journal, conference, repository, or documentation site has useful but hard-to-find pages.
  • Use crawl only when the user needs many pages from a known source, such as proceedings, a lab publication list, or a technical docs section.
  • Use research only when the user explicitly needs a full literature review or research landscape.

Query And Source Guidance

  • Pair formal terms with common synonyms and acronyms.
  • Include population, dataset, benchmark, organism, intervention, model, metric, or method terms when relevant.
  • For recent literature, include explicit year ranges or recency language in the query.
  • Prefer primary literature for claims, review papers for landscape summaries, and guidelines/standards for practice recommendations.
  • Deduplicate preprints and final publications; cite the final version when available.
  • Treat abstracts, papers, supplemental pages, tables, and official repositories as separate evidence surfaces.
  • Report failed or inaccessible sources when they affect key papers, guidelines, datasets, or primary evidence.

Output Template

Use this markdown structure and adapt sections to the user's scope:

# Evidence Brief: <topic>

## Research Question
<brief framing of the question>

## Search Strategy
- Concepts searched:
- Source types prioritized:
- Inclusion logic:
- Exclusions or limitations:

## Key Sources
| Source | Type | Why it matters | URL |
| --- | --- | --- | --- |

## Findings
### <Theme>
- Claim:
- Evidence:
- Limitations:

### <Theme>
- Claim:
- Evidence:
- Limitations:

## Consensus And Disagreement
- Consensus:
- Disagreement:
- Evidence gaps:

## Follow-Up Queries
- <query idea>

For medical, clinical, or biomedical topics, be conservative: do not provide diagnosis or treatment advice, and prioritize primary literature, systematic reviews, guidelines, and official health sources.

來自 tavily-ai 的更多技能

crawl
tavily-ai
提取並將網站內容儲存為 Markdown 檔案,以供離線存取與分析。支援可設定的爬取深度(1-5層)、廣度限制與頁面上限,以平衡涵蓋範圍與效能。包含透過正則表達式模式進行路徑過濾,專注於特定區塊並排除無關內容。提供兩種模式:全頁提取用於資料收集,或語意分塊搭配自然語言指令,將結果導入 LLM 上下文。附帶配套的 Map API 用於 URL...
extract
tavily-ai
使用Tavily的提取API從特定URL提取乾淨內容。每次請求最多支援20個URL,可選基於查詢的重新排序以聚焦相關內容區塊。兩種提取模式:基本模式用於快速文字提取,進階模式用於JavaScript渲染頁面和結構化資料。首次運行時透過瀏覽器自動進行OAuth驗證,或手動在設定中配置API金鑰。返回Markdown或純文字格式,可選圖片URL,並可設定最長60秒的超時時間。
investment-research-briefs
tavily-ai
編寫簡明的投資研究简报、公司備忘錄、行業概覽、投資組合監控更新、財報/新聞摘要、風險與催化劑简报等……
product-competitor-intelligence
tavily-ai
研究競爭對手產品、SKU、定價、包裝、定位、功能比較、產品目錄、類別頁面、零售商列表、市場……
research
tavily-ai
針對任何主題進行全面研究,自動收集來源、分析並提供引用。執行多來源網路研究並附上明確引用,適合比較、時事、市場分析及詳細報告。提供三種模型選項:mini 針對單一主題的目標研究(約30秒)、pro 進行全面的多角度分析(約60-120秒),以及 auto 透過 API 驅動的複雜度偵測。透過 Tavily MCP 伺服器以 OAuth 進行驗證,並在...上自動執行基於瀏覽器的登入。
sales-account-intelligence
tavily-ai
為公司、潛在客戶、現有客戶、合作夥伴及目標買家建立銷售就緒的帳戶情報。當用戶要求銷售準備、帳戶簡報等時使用。
search
tavily-ai
使用LLM優化結果的網路搜尋,具備相關性評分與靈活篩選功能。支援四種搜尋深度模式(極速、快速、基本、進階),可配置延遲與相關性權衡。包含網域篩選、時間範圍限制、日期區間、國家加權及原始內容提取。回傳結果包含標題、網址、內容摘要與相關性評分;可選圖片結果與網站圖示。透過Tavily MCP伺服器或API金鑰配置自動進行OAuth驗證;...
tavily-best-practices
tavily-ai
專為LLM設計的網路搜尋API,具備即時資料存取、內容擷取、網站爬取及AI驅動研究功能。五大核心方法:search()用於搜尋網頁結果、extract()用於擷取URL內容、crawl()用於全站擷取、map()用於URL探索,以及research()用於端到端AI綜合分析。支援Python與JavaScript SDK,提供非同步客戶端以進行平行查詢,並可設定搜尋深度(極速/快速/基本/進階)。Crawl方法接受語意指令,以聚焦於特定內容的擷取...