deep-research-firecrawl

Compare original and translation side by side

🇺🇸

Original

English
🇨🇳

Translation

Chinese

Deep Research (Firecrawl)

深度研究(Firecrawl)

Purpose

目的

Deliver verifiable research reports with explicit citations, clear fact-vs-analysis boundaries, and user-selectable depth.
交付可验证的研究报告,包含明确引用、清晰的事实与分析界限,且支持用户选择研究深度。

Tooling requirements

工具要求

  • Use Firecrawl MCP for web retrieval:
    • firecrawl:firecrawl_search
      for discovery
    • firecrawl:firecrawl_scrape
      for page extraction
    • firecrawl:firecrawl_map
      for URL discovery on known domains
    • firecrawl:firecrawl_crawl
      for controlled multi-page capture
    • firecrawl:firecrawl_agent
      for autonomous broad collection when scope is unclear
  • Prefer parallel retrieval whenever tasks are independent.
  • Hard-stop policy: if Firecrawl MCP is unavailable, pause and ask the user to re-enable it. Do not switch to other web retrieval tools.
  • 使用Firecrawl MCP进行网页检索:
    • firecrawl:firecrawl_search
      用于资源发现
    • firecrawl:firecrawl_scrape
      用于页面内容提取
    • firecrawl:firecrawl_map
      用于在已知域名下发现URL
    • firecrawl:firecrawl_crawl
      用于可控的多页面捕获
    • firecrawl:firecrawl_agent
      用于范围不明确时的自主广泛收集
  • 只要任务相互独立,优先采用并行检索方式。
  • 强制停止规则:若Firecrawl MCP不可用,需暂停操作并请求用户重新启用。不得切换至其他网页检索工具。

When to use

适用场景

Use this skill when the request needs multiple sources, synthesis, and evidence quality controls.
Do not use this skill for:
  • Simple one-off lookups
  • Debugging or code-only tasks
  • Questions answerable with 1-2 sources
当需求涉及多源信息、综合分析及证据质量管控时,使用本技能。
请勿将本技能用于:
  • 简单单次查询
  • 调试或纯代码类任务
  • 1-2个来源即可解答的问题

Mode selection

模式选择

If the user does not specify a mode, default to
standard
.
ModeTypical runtimeSource targetOutput depth
quick
3-6 min6-10 sources800-1,500 words
standard
8-15 min12-20 sources2,000-4,000 words
deep
15-30 min25-40 sources4,000-8,000 words
ultradeep
30-60+ min40-80 sources8,000-15,000+ words
若用户未指定模式,默认使用
standard
模式。
模式典型运行时长目标数据源数量输出深度
quick
3-6分钟6-10个来源800-1500词
standard
8-15分钟12-20个来源2000-4000词
deep
15-30分钟25-40个来源4000-8000词
ultradeep
30-60+分钟40-80个来源8000-15000+词

Execution workflow

执行流程

1) Scope

1) 范围界定

  • Restate the research question.
  • Capture explicit constraints (time window, geography, sector, audience).
  • Define assumptions only when missing.
  • 重述研究问题。
  • 记录明确的约束条件(时间范围、地域、行业、受众)。
  • 仅在信息缺失时定义假设。

2) Retrieval plan

2) 检索计划

  • Break the topic into 5-10 search angles.
  • Include opposing viewpoints and recent developments.
  • Mark which angles need domain deep-dives.
  • 将研究主题拆解为5-10个搜索角度。
  • 涵盖对立观点及最新动态。
  • 标记需要深入挖掘的领域方向。

3) Retrieve with Firecrawl

3) 借助Firecrawl进行检索

  • Run parallel
    firecrawl:firecrawl_search
    queries for all angles.
  • Scrape top results with
    firecrawl:firecrawl_scrape
    for primary evidence.
  • For known high-value sites:
    • discover URLs via
      firecrawl:firecrawl_map
    • crawl key paths with bounded
      firecrawl:firecrawl_crawl
      limits
  • Use
    firecrawl:firecrawl_agent
    when the topic is open-ended or highly fragmented.
  • 针对所有搜索角度并行运行
    firecrawl:firecrawl_search
    查询。
  • 使用
    firecrawl:firecrawl_scrape
    抓取顶级搜索结果作为核心证据。
  • 对于已知高价值站点:
    • 通过
      firecrawl:firecrawl_map
      发现URL
    • 以受限的
      firecrawl:firecrawl_crawl
      规则爬取关键路径
  • 当研究主题开放或高度分散时,使用
    firecrawl:firecrawl_agent

4) Triangulate

4) 交叉验证

  • Verify important claims across at least 3 independent sources.
  • Prefer primary sources over commentary when possible.
  • Flag conflicts and unresolved uncertainty.
  • 重要事实主张需通过至少3个独立来源验证。
  • 优先采用原始来源而非评论性内容。
  • 标记冲突点及未解决的不确定性。

5) Synthesize

5) 综合分析

  • Separate facts from interpretation.
  • Explain what is known, what is likely, and what is uncertain.
  • Include alternative explanations for contested topics.
  • 区分事实与解读内容。
  • 说明已知信息、大概率结论及不确定点。
  • 针对有争议的主题提供替代解释。

6) Critique and refine (required for
deep
/
ultradeep
)

6) 审核与优化(
deep
/
ultradeep
模式必填)

  • Stress-test weak claims.
  • Check for recency bias and survivorship bias.
  • Add missing counterevidence.
  • 对薄弱主张进行压力测试。
  • 检查是否存在时效性偏差及幸存者偏差。
  • 补充缺失的反证。

7) Package output

7) 输出打包

Use
templates/report_template.md
and include:
  • Executive Summary
  • Method and Scope
  • Key Findings
  • Counterevidence and Risks
  • Recommendations
  • Bibliography
使用
templates/report_template.md
模板,内容需包含:
  • 执行摘要
  • 方法与范围
  • 核心发现
  • 反证与风险
  • 建议
  • 参考文献

Citation and quality rules

引用与质量规则

  • Every major factual claim must include a citation marker like
    [1]
    .
  • Bibliography entries must map one-to-one with cited markers.
  • If evidence is weak or contradictory, say so explicitly.
  • Never invent citations.
  • 每个重要事实主张必须包含
    [1]
    这类引用标记。
  • 参考文献条目需与引用标记一一对应。
  • 若证据薄弱或存在矛盾,需明确说明。
  • 严禁伪造引用。

Stop conditions

停止条件

Pause and report limitations when:
  • Required Firecrawl MCP tools are unavailable (hard-stop policy).
  • Source count stays below minimum for the selected mode after broad retrieval attempts.
  • Core claims cannot be verified from reliable sources.
出现以下情况时,需暂停并报告限制:
  • 所需的Firecrawl MCP工具不可用(遵循强制停止规则)。
  • 经过广泛检索尝试后,数据源数量仍未达到所选模式的最低要求。
  • 核心主张无法通过可靠来源验证。

Progressive disclosure

渐进式披露

Load only what is needed:
  • Method details:
    reference/methodology.md
  • Output template:
    templates/report_template.md
仅加载必要内容:
  • 方法细节:
    reference/methodology.md
  • 输出模板:
    templates/report_template.md

Quick invocation examples

快速调用示例

  • "Use
    deep-research-firecrawl
    in
    quick
    mode on edge AI chip trends in 2025."
  • "Run
    deep-research-firecrawl
    in
    deep
    mode comparing Auth0, Clerk, and Supabase Auth."
  • "Use
    deep-research-firecrawl
    in
    ultradeep
    mode for US longevity biotech funding and risks."
  • "以
    quick
    模式使用
    deep-research-firecrawl
    研究2025年边缘AI芯片趋势。"
  • "以
    deep
    模式运行
    deep-research-firecrawl
    对比Auth0、Clerk和Supabase Auth。"
  • "以
    ultradeep
    模式使用
    deep-research-firecrawl
    研究美国长寿生物技术的融资情况及风险。"