# Research Task Handoff: Scope, Sources, Stop Rules

> Reusable prompt that turns a research question into a bounded handoff with allowed sources, citation fields, a stop rule and a self-review list.

## Install

Copy the content below into your project:

# Research Task Handoff: Scope, Sources, Stop Rules

Reusable prompt that turns a research question into a bounded handoff with allowed sources, citation fields, a stop rule and a self-review list.

## Start here

Paste your research question and context into an ordinary AI chat that accepts text, then paste the full prompt (appended below) as the next message. You do not need any terminal, account setup, or API key.

Inputs to include, in this order:

1. **Question** — one to three sentences in your own words.
2. **Purpose** — what you will do with the answer and who reads it.
3. **Source rules** — allowed sources, off-limits sources, any date or geography limit.
4. **Budget** — for example, "up to 8 sources, 45 minutes."
5. **Known context** — what you already checked, so it is not repeated.
6. **Output you need** — for example, a memo, a comparison table, or a list of options.

If a field is empty, the prompt asks the AI to flag it as a needed input instead of inventing one. Fill in what you have; the draft will name the gaps.

How to check the output: read the review list at the end of the draft. Each item should be marked pass or fail with a one-line reason, the stop rule should copy a countable limit from your budget, and the handback format should ask for a source for every claim.

## Introduction

This is a drafting prompt. It produces a plan you can paste into a research agent, hand to a human assistant, or follow yourself. It does not browse, contact anyone, or send anything. Nothing in it is verified correct — it is a plan for you to approve and adjust.

## Permissions and limits

The prompt only writes a handoff draft and a review list. It must not add sources, examples, dates, or numbers that were not in your inputs. It must not instruct anyone to contact people, post, purchase, or decide on your behalf. If your inputs are too vague for a bounded handoff, it should say which specific input is missing and stop there.

Keep private or sensitive details out of the chat unless you are comfortable with your provider's data handling. Source reviewed; runtime not tested.

## FAQ

**What if I do not know a budget?** Leave it out. The draft should flag it as a needed input, since the stop rule depends on a countable limit.

**Can I send the draft to an agent as-is?** Read the review list first. Fix any failing checks, then approve it yourself.

## Source and thanks

Original TokRepo prompt, shared under [CC BY 4.0](<https://creativecommons.org/licenses/by/4.0/>). Context reference: [ChatGPT release notes](<https://help.openai.com/en/articles/6825453-chatgpt-release-notes>), reviewed 2026-10-04.

## Complete reusable prompt

# Research Task Handoff: Scope, Sources, Stopping Rules

You are helping me write a *research task handoff* that I can paste into a research agent, give to a human assistant, or follow myself manually. The handoff must be specific enough that the person or tool doing the research knows what to look for, where to look, when to stop, and how to show its work.

Important: you are drafting a plan and a set of instructions. You are not performing the research, not browsing, not contacting anyone, and not sending anything. Do not invent sources, findings, dates, or quotes.

## My inputs

I will paste the following, in this order. If any part is missing or thin, say so in your output rather than guessing.

1. **Question**: the research question in my own words (one to three sentences).
2. **Purpose**: what I will do with the answer and who will read it.
3. **Source rules**: which sources are allowed, which are off-limits, and any date range or geographic limit.
4. **Budget**: how many items or how much time is acceptable, e.g. "up to 8 sources, 45 minutes."
5. **Known context**: anything I already believe or have already checked, so it is not repeated.
6. **Output I need**: a memo, a comparison table, a list of options, etc.

Example of a filled input (fictional, for shape only):
- Question: Which two or three low-cost ways do small offices reduce paper waste?
- Purpose: a one-page internal note for a 12-person office; the office manager will decide.
- Source rules: allowed: government waste-reduction pages, university facilities pages, nonprofit guides, vendor documentation for cost figures; not allowed: forum posts, undated blog listicles, social media. Date range: 2022 or later.
- Budget: up to 6 sources, about 40 minutes.
- Known context: we already recycle paper; single-sided printing is the current default.
- Output: a short comparison with a rough cost range and one caveat per option.

## What to produce

Write the handoff in these six sections, in this order, using short prose or simple bullets. Keep it compact enough to paste into one message.

1. **Restated question and purpose** — one or two sentences, in plain language, plus what decision or draft this will feed. If my question is too broad, propose one narrower version and mark it as a suggestion.
2. **Scope and boundaries** — what is in scope, what is out of scope, the allowed and disallowed source types from my inputs, and any date or geography limit. Call out topics that look adjacent but should not be included.
3. **Source requirements** — for each source the researcher should record: title, publisher, publication or last-updated date, link, and which allowed rule it satisfies. State plainly that a source without a verifiable date or publisher must be listed as *unverified* rather than silently used.
4. **Method steps** — a short numbered sequence the researcher follows: search, screen against the source rules, extract only claims that answer the question, note the exact supporting sentence or figure for each claim, then consolidate. Include one instruction that claims appearing in only one weak source must be labeled *single-source, low confidence*.
5. **Stopping rules** — when to stop. Use my budget. Stop when the budget is reached, when two additional searches return no new allowed source, or when further searching would require breaking a source rule. Add that the researcher must stop and report rather than widen scope on its own.
6. **Handback format** — the exact structure of the returned work: one-line answer at the top; then a table with columns Claim | Source (title, date) | Confidence | Note; then a short "open questions and gaps" list; then a "what I did not check" line. Ask for the total count of sources actually used.

## Checks to run before I send it

After drafting, run this review on your own draft and report results as a short list, each marked pass or fail with a one-line reason:

1. Every claim in the handback format maps to a recorded source field — no uncited claims allowed.
2. The stopping rule names a concrete, countable limit taken from my inputs, not a vague "when enough."
3. Disallowed source types from my inputs appear verbatim in the boundaries section.
4. No source, finding, statistic, or date was invented in the draft.
5. The draft never instructs the researcher to contact people, post, purchase, or make a decision on my behalf.
6. If I left an input field empty, the draft flags it as a needed input instead of filling it in.

## Boundaries

- You prepare one handoff draft and its review list. You do not execute the research or the plan.
- Do not add sources, examples, or numbers that were not in my inputs.
- Do not present the draft as verified correct; it is a plan for me to approve and adjust.
- If my inputs are too vague to write a bounded handoff, say what specific input is missing and stop there.


## References and reuse

- [ChatGPT release notes](https://help.openai.com/en/articles/6825453-chatgpt-release-notes) · Reviewed 2026-10-04

Original TokRepo prompt · [CC BY 4.0](https://creativecommons.org/licenses/by/4.0/). Reference documents retain their own rights.

---

<!-- ZH -->

# 研究任务交接单：范围、来源与停止规则

一个可复用的提示词，把研究问题整理成有边界的任务交接单，包含允许的来源、引用字段、停止条件和自查清单。

## 开始使用

把你的研究问题和背景粘贴到一个能接受文本输入的普通 AI 对话中，然后把完整提示词（附在下方）作为下一条消息粘贴进去。无需终端、账号配置或 API 密钥。

按以下顺序提供输入：

1. **问题** —— 用你自己的话写一到三句。
2. **用途** —— 你会拿这个答案做什么，谁会阅读它。
3. **来源规则** —— 允许哪些来源，禁止哪些来源，以及任何日期或地域限制。
4. **预算** —— 例如“最多 8 个来源、45 分钟”。
5. **已知背景** —— 你已经核查过的内容，避免重复。
6. **你需要的产出** —— 例如备忘录、对比表格或选项清单。

如果某个字段是空的，提示词会要求 AI 将其标注为“缺少的输入”，而不是自行编造。把你已有的信息填进去，草稿会指出缺口。

如何检查输出：先读草稿末尾的自查清单。每一项都应标注“通过”或“未通过”并附一行理由；停止规则应引用你预算中可计数的上限；交接格式应为每条结论都要求一个来源。

## 简介

这是一个起草用的提示词。它产出一份计划，你可以粘贴给研究智能体、交给人类助手，或自己按步骤执行。它不会浏览网页、联系任何人或发送任何内容。其中内容都未经核实为正确——它是一份供你批准和调整的计划。

## 权限与限制

该提示词只负责撰写交接草稿和自查清单。它不得添加你输入中没有的来源、示例、日期或数字。它不得指示任何人代替你联系他人、发帖、购买或做决定。如果你的输入过于模糊、无法写出有边界的交接单，它应指出缺少哪个具体输入，并就此停止。

除非你认可所用服务商的数据处理方式，否则不要把隐私或敏感信息放进对话。来源已查阅，运行时未测试。

## 常见问题

**如果我定不出预算怎么办？** 可以先不写。草稿应把它标注为缺少的输入，因为停止规则需要一个可计数的上限。

**我能把草稿原样发给智能体吗？** 先读自查清单，修正未通过的项，然后由你自己批准。

## 来源与致谢

TokRepo 原创提示词，依据 [CC BY 4.0](<https://creativecommons.org/licenses/by/4.0/>) 共享。背景参考资料：[ChatGPT release notes](<https://help.openai.com/en/articles/6825453-chatgpt-release-notes>)，查阅日期 2026-10-04。

## 完整可复制提示词

# 研究任务交接：范围、来源、停止规则

你正在帮助我撰写一份*研究任务交接单*，以便我将其粘贴给研究智能体、交给人类助手，或自己手动执行。交接单必须足够具体，让执行研究的人或工具知道要查找什么、去哪里查找、何时停止，以及如何展示工作过程。

重要提示：你起草的是一份计划和一套指令。你不执行研究、不浏览网页、不联系任何人、不发送任何内容。不得编造来源、发现、日期或引语。

## 我的输入

我会按以下顺序粘贴内容。如果有任何部分缺失或过于单薄，请在输出中说明，而不是猜测。

1. **问题**：用我自己的话表述的研究问题（一到三句）。
2. **用途**：我会拿这个答案做什么，谁会阅读它。
3. **来源规则**：允许哪些来源、禁止哪些来源，以及任何日期范围或地域限制。
4. **预算**：可以接受多少个条目或多少时间，例如“最多 8 个来源、45 分钟”。
5. **已知背景**：我已经相信或已经核查过的任何内容，以免重复。
6. **我需要的产出**：备忘录、对比表格、选项清单等。

已填写输入的示例（虚构，仅作形式参考）：
- 问题：小型办公室减少纸张浪费的两三种低成本方法是什么？
- 用途：给一个 12 人办公室的一页内部说明；由办公室经理决定。
- 来源规则：允许：政府减废页面、大学设施页面、非营利指南、用于成本数字的供应商文档；不允许：论坛帖子、无日期的博客清单、社交媒体。日期范围：2022 年或之后。
- 预算：最多 6 个来源，约 40 分钟。
- 已知背景：我们已经回收纸张；单面打印是当前默认设置。
- 产出：一份简短对比，包含大致成本范围和每个选项的一条注意事项。

## 需要产出的内容

按以下顺序，用这六个部分撰写交接单，使用简短散文或简单项目符号。保持紧凑，足够粘贴进一条消息。

1. **重述的问题与用途** —— 用平实语言写一到两句，并说明这会用于什么决定或草稿。如果我的问题过于宽泛，提出一个更窄的版本，并将其标注为建议。
2. **范围与边界** —— 什么在范围内、什么在范围外，我的输入中允许和不允许的来源类型，以及任何日期或地域限制。指出看起来相邻但不应包含的主题。
3. **来源要求** —— 对于每个来源，研究者应记录：标题、发布者、发布日期或最后更新日期、链接，以及它满足哪条允许规则。明确说明：没有可核实日期或发布者的来源必须列为*未核实*，而不是静默使用。
4. **方法步骤** —— 研究者遵循的简短编号序列：搜索、按来源规则筛选、只提取回答问题的结论、为每条结论记下确切的支撑句或数字，然后整合。加入一条指令：只出现在一个弱来源中的结论必须标注为*单一来源，低置信度*。
5. **停止规则** —— 何时停止。使用我的预算。当预算用尽、当额外两次搜索没有返回新的允许来源，或当进一步搜索会需要违反来源规则时，就停止。补充说明：研究者必须停止并报告，而不是自行扩大范围。
6. **交接格式** —— 返回工作的确切结构：顶部一行答案；然后是一个表格，列为 结论 | 来源（标题、日期） | 置信度 | 备注；然后是一个简短的“未决问题与缺口”清单；然后是一行“我未核查的内容”。要求提供实际使用的来源总数。

## 发送前要运行的检查

起草完成后，对你自己产生的草稿运行以下自查，并以简短清单形式报告结果，每一项都标注“通过”或“未通过”并附一行理由：

1. 交接格式中的每条结论都对应一个已记录的来源字段——不允许出现未引用的结论。
2. 停止规则引用的是从我的输入中取得的、具体且可计数的上限，而不是含糊的“当足够时”。
3. 我在输入中禁止的来源类型，逐字出现在边界部分。
4. 草稿中没有编造任何来源、发现、统计数据或日期。
5. 草稿从不指示研究者代替我联系他人、发帖、购买或做决定。
6. 如果我留空了某个输入字段，草稿应将其标注为“缺少的输入”，而不是自行补上。

## 边界

- 你只准备一份交接草稿及其自查清单。你不执行研究或计划。
- 不要添加我的输入中没有的来源、示例或数字。
- 不要把草稿呈现为已核实正确；它是供我批准和调整的计划。
- 如果我的输入过于模糊、无法写出有边界的交接单，请说明缺少哪个具体输入，并就此停止。

## 参考资料与复用

- [ChatGPT release notes](https://help.openai.com/en/articles/6825453-chatgpt-release-notes) · Reviewed 2026-10-04

TokRepo 原创提示词 · [CC BY 4.0](https://creativecommons.org/licenses/by/4.0/)。参考资料保留各自原有权利。

---
Source: https://tokrepo.com/en/workflows/research-task-handoff-scope-sources-stop-rules-1fbdd911
Author: Prompt Lab