DeepResearch 6D Prompt

Overview

Use this skill to convert a research need into a complete Deep Research prompt. The output should be a prompt that another AI research agent can execute, not the research answer itself.

The generated prompt must contain exactly six major sections: D1-D6. Do not create separate top-level sections named "研究主题", "Assumptions", or "质量要求". Those requirements must be embedded inside the relevant D1-D6 sections.

Source Methodology

Extracted 6D Deep Research Workflow:

1. D1 Define - 定义目标
• 研究什么？
• 为什么研究？
• 用于什么决策？
• 研究主题、默认假设、范围边界都放在本节。
2. D2 Decompose - 拆解结构
• 要拆成哪些维度、地区、对象或问题？
3. D3 Design Fields - 设计字段
• 每一项要收集哪些字段？
• 什么算完整？
4. D4 Discover - 有组织搜索
• 去哪里搜？
• 怎么多源检索？
• 来源质量、检索渠道、排除规则都放在本节。
5. D5 Digest - 消化汇总
• 如何去重、清洗、归纳、求证？
• 事实核验、冲突处理、引用规范都放在本节。
6. D6 Deliver - 交付输出
• 最终交付表格、报告还是建议方案？
• 输出格式、结论表达、风险和下一步都放在本节。

Core principles:

• 流程设计优先于提示词编写。
• 先有交付物，后有搜索。
• 先有业务结构，后有检索任务。
• 最终提示词只保留 D1-D6 六个主标题，避免额外顶层模块分散执行重点。

Workflow

When the user provides a topic or rough research request:

1. Identify the likely decision, audience, and expected deliverable.
2. Rewrite the user's task into a clear research target and place it inside D1, not as a standalone "研究主题" section.
3. If essential details are missing, make conservative assumptions and place them inside D1 under "默认假设与边界". Ask a clarifying question only when the missing detail changes the research scope materially.
4. Design the deliverable before designing the search plan.
5. Decompose the business or research structure before writing source queries.
6. Generate one complete prompt with exactly six major 6D sections.
7. Embed concrete requirements for source quality, verification, output format, and citations into D4, D5, and D6. Do not add a standalone "质量要求" section.

Output Requirements

Always output in Chinese unless the user requests another language.

Start with a short line:

以下是根据 6D Deep Research Workflow 生成的可执行提示词：

Then provide a fenced prompt block. The prompt inside the block must be written as direct instructions to the future Deep Research agent.

Use this structure:

你是一名 Deep Research 研究员。请严格按照 6D Deep Research Workflow 完成系统性研究，并在每一步明确说明依据、范围、证据和不确定性。

D1 Define｜定义目标
- 研究对象与主题：[清晰改写用户任务；不要另起“研究主题”标题]
- 研究背景：[说明为什么需要研究该问题]
- 核心问题：[列出需要回答的 3-6 个关键问题]
- 使用场景/决策场景：[说明研究结果将服务于什么判断、方案或行动]
- 默认假设与边界：[列出必要假设；如不需要假设，写“暂无额外假设”；同时说明时间、地域、对象、行业或资料范围边界]
- 成功标准：[说明什么样的结果算完成研究]

D2 Decompose｜拆解结构
- 研究维度：[按业务、技术、市场、政策、学术、用户、场景等合适维度拆解]
- 地区/市场/人群/对象拆分：[根据任务选择适用拆分方式]
- 关键子问题：[把 D1 的核心问题拆成可检索、可验证的子问题]
- 对比对象或基准：[说明需要横向比较的对象、标准或历史基线]
- 优先级：[说明先查什么、后查什么，以及优先级理由]

D3 Design Fields｜设计字段
请为每个研究对象或维度收集以下字段，并说明缺失字段如何处理：
| 字段 | 含义 | 必填/可选 | 完整性标准 | 推荐来源 |
|---|---|---|---|---|
| [field] | [meaning] | [required/optional] | [completion rule] | [source type] |

D4 Discover｜有组织搜索
- 优先来源：[优先使用一手来源、官方文件、权威数据库、行业报告、学术资料或可信媒体]
- 检索渠道：[列出具体数据库、官网、搜索引擎、机构网站、论文库、报告库或数据平台]
- 中英文/多语言关键词：[列出可直接搜索的关键词组合]
- 检索顺序：[先查权威定义和原始数据，再查报告分析、案例和新闻验证]
- 多源交叉验证方式：[关键事实至少用两个独立来源核验；时间敏感信息必须标注发布日期或数据时间]
- 排除来源或低质量信号：[排除无来源转载、营销软文、无法追溯数据、过期材料、明显利益相关但未披露的方法]

D5 Digest｜消化汇总
- 去重规则：[合并重复来源、重复案例、重复数据口径]
- 清洗规则：[统一时间、地区、单位、分类口径和术语定义]
- 可信度评估：[按来源权威性、数据透明度、时效性、方法可复核性评估证据等级]
- 冲突信息处理：[保留冲突点，解释可能原因，并优先采用更原始、更权威、更新的数据]
- 归纳框架：[说明如何从事实归纳出模式、趋势、原因、影响或判断]
- 引用要求：[明确区分事实、推断和建议；关键结论必须附来源链接或可追溯出处]

D6 Deliver｜交付输出
- 最终交付物类型：[表格、结构化报告、对比矩阵、研究简报、建议方案等]
- 输出结构：[给出章节或表格结构]
- 表格字段：[列出最终表格需要呈现的字段]
- 结论与建议：[基于证据输出结论，避免把未经验证的信息写成事实]
- 风险、不确定性与信息缺口：[列出证据不足、口径冲突、资料缺失和可能影响]
- 可执行下一步：[给出后续验证、访谈、实验、采购、政策跟踪或决策建议]
- 最终质量检查：[确认来源质量、交叉验证、时间标注、事实/推断/建议区分、引用链接完整性均已满足]

Adaptation Rules

Tailor each 6D section to the user's domain:

• For market or competitor research, emphasize market segmentation, players, pricing, channels, customer groups, and trend evidence.
• For product or user research, emphasize user segments, scenarios, pain points, jobs-to-be-done, feature comparisons, and evidence confidence.
• For policy or legal research, emphasize jurisdiction, effective dates, primary legal texts, enforcement bodies, and compliance implications.
• For academic or technical research, emphasize definitions, methods, benchmark papers, datasets, evaluation metrics, and reproducibility.
• For soil, geotechnical, civil engineering, environmental, or material research, emphasize material type, test conditions, spatial scale, engineering scenario, standards, parameters, datasets, laboratory/field methods, uncertainty, and reproducibility.
• For internal business research, emphasize stakeholder decision, available data, operating constraints, risks, and recommended action.

Guardrails

• Do not output standalone top-level sections named "研究主题", "Assumptions", or "质量要求".
• Place the user's research topic in D1.
• Place assumptions in D1.
• Place source quality, verification, and citation requirements across D4, D5, and D6.
• Do not add decorative commentary after the prompt.
• If useful, add a brief "可选补充问题" section before the fenced prompt with no more than three questions.