科研 Skill

研究选题构思设计

brainstorming-research

Explores research intent, field state, methodology, and ethics, then writes an approved design doc. Use before starting a new research question or chapter.

发布者 Codar安全审核通过v0.1.0更新于 2026-11-18

SKILL.md

Brainstorming Research Ideas Into Designs

Turn a vague research interest into a written, approvable design doc. Act as an ideation partner, not an instructor. Probe, challenge, surface unexamined assumptions — then lock the design down on paper so later phases have something stable to build on.

If unmet: explain to the user what's missing, ask for a short reason for skipping (e.g. "mini-review, not a standalone project"), write it to knowledge/_meta/gate-overrides.log, and continue. Repeated skipping is a signal that the routine is fraying — the lint stats make that visible.

Anti-Pattern: "This Is Too Simple To Need A Design"

Every research project goes through this process. A single-source review, a short methodological note, a conference abstract — all of them. "Simple" projects are where unexamined assumptions cause the most wasted work. The design can be short (a half page for truly simple projects), but you MUST present it and get approval.

Checklist

Create a task-list item for each. Complete in order:

  1. Explore project context — read existing input/ideas/*, input/description/*, recent knowledge/_meta/log.md, check git for parallel work
  2. Ask clarifying questions — one at a time — what is the research question, what has been seen already, what would success look like. The user formulates the question first, in their own words, however rough; you sharpen it afterwards (see Phase 1)
  3. Ask for the AI-use rules — which institution / publisher / journal / funder will receive the output, and what do their rules say about AI tools (allowed uses, required disclosure form)? Unknown is a valid answer — record it as an open point. This goes into the design doc's AI-Use Rules section and is read again by finishing-a-research-project
  4. Assess scope — is this a single focused inquiry or an umbrella of independent sub-projects? If the latter, decompose first (each sub-project gets its own design doc)
  5. Propose 2–3 approaches — with trade-offs and your recommendation, grounded in disciplinary norms
  6. Surface hidden constraints — ethics (ancient human remains, epigraphic access permissions), data rights, linguistic reach, time horizon, collaborator availability
  7. Present design in sections — get user approval after each: question, state of the field (short), method, data, output target, risks
  8. Write design doc — save to input/ideas/<slug>-design.md using the template below
  9. Design self-review — scan for placeholders, contradictions, scope creep, ambiguous claims; fix inline
  10. User reviews the written doc — ask for sign-off on the file before handing off
  11. Transition to writing-research-plan — once approved, invoke that skill. Do NOT invoke anything else.

Process Flow

digraph brainstorming_research {
    "Explore project context" [shape=box];
    "Scope: single or umbrella?" [shape=diamond];
    "Decompose into sub-projects" [shape=box];
    "Ask clarifying questions" [shape=box];
    "Propose 2-3 approaches" [shape=box];
    "Surface hidden constraints" [shape=box];
    "Present design sections" [shape=box];
    "User approves each section?" [shape=diamond];
    "Write design doc" [shape=box];
    "Self-review (fix inline)" [shape=box];
    "User approves written doc?" [shape=diamond];
    "Invoke writing-research-plan" [shape=doublecircle];

    "Explore project context" -> "Scope: single or umbrella?";
    "Scope: single or umbrella?" -> "Decompose into sub-projects" [label="umbrella"];
    "Scope: single or umbrella?" -> "Ask clarifying questions" [label="single"];
    "Decompose into sub-projects" -> "Ask clarifying questions";
    "Ask clarifying questions" -> "Propose 2-3 approaches";
    "Propose 2-3 approaches" -> "Surface hidden constraints";
    "Surface hidden constraints" -> "Present design sections";
    "Present design sections" -> "User approves each section?";
    "User approves each section?" -> "Present design sections" [label="no, revise"];
    "User approves each section?" -> "Write design doc" [label="yes"];
    "Write design doc" -> "Self-review (fix inline)";
    "Self-review (fix inline)" -> "User approves written doc?";
    "User approves written doc?" -> "Write design doc" [label="changes requested"];
    "User approves written doc?" -> "Invoke writing-research-plan" [label="approved"];
}

Terminal state: invoke writing-research-plan. Do NOT invoke literature-review, ingest-source, or any execution skill directly — those belong after the plan is written.

Design Doc Template

Save to input/ideas/<slug>-design.md (slug = short, kebab-case, e.g. chronologie-levante-eisen-ii).

---
title: "<Research question in one sentence>"
type: design
created: YYYY-MM-DD
updated: YYYY-MM-DD
tags: [<field>, <region>, <method>]
status: draft
author: mixed
---

# <Title>

## Research Question
A precise, falsifiable or argumentatively testable guiding question.

## Relevance / Forschungsstand (brief)
2–5 sentences: why now, where the debate stands, which gap is addressed.
(Full literature work happens in literature-review — only positioning here.)

## Method
Which approach (*Quellenkritik*, stratigraphic comparison, quantitative
analysis, discourse-analytic reading …). Why this one and not the alternatives.

## Data Sources / Material Basis
Concrete archives, databases, text corpora, excavation publications, datasets.

## Output Target
☐ Book chapter   ☐ Journal article   ☐ Grant proposal   ☐ Talk   ☐ Other

## Expected Results
What would a meaningful positive result look like? What a null finding?

## Ethical / Methodological Risks
e.g. access rights to sources, handling of human remains, postcolonial
sensitivity, bias in transmission traditions.

## Feasibility
Time frame, tools (see CLAUDE.md §5), need for collaborators.

## AI-Use Rules
Target institution / venue: <university, journal, publisher, funder — or "open">.
What their rules allow and require (e.g. "AI permitted for language editing and
literature search; disclosure statement in acknowledgements required").
Required disclosure form: statement in text | tool overview page | AI Usage Card | not specified.
Open points: <rules not yet checked — ask supervisor / read author guidelines>.

## Scope Boundaries (what is NOT investigated)
Explicit mention of related questions that fall out of scope.

## Next Step
Hand off to writing-research-plan to produce `input/ideas/<slug>-plan.md`.

The Process

Phase 1 — Understanding the idea:

  • Read recent project activity first (existing design docs, log, commits)
  • Ask one question per message. Prefer multiple-choice where possible — except for the research question itself, which is open
  • Focus: what is the question, what has already been looked at, what would success look like?
  • The question comes from the user, Socratically. Before you offer any formulation of your own, ask the user to state what they want to understand and why ("I am investigating X, because I want to understand Y, in order to Z"). A rough sentence, a voice-note transcript or a paragraph of free writing is enough. Then work on their text: ask what a term means, what would count as an answer, which case they have in mind, what they would expect to find. Offer your own variants only after that, as alternatives to compare against theirs — never as the starting point. The reason: formulating the question is the thinking; a polished suggestion that the user merely approves looks like agreement but skips it (automation bias — a machine-made formulation reads as better than one's own and gets accepted unchecked)
  • If the user explicitly asks you to propose the question ("schlag du was vor"), do so, but say that the design doc will mark it as agent-proposed until they have reworded or confirmed it in their own words

Phase 2 — Divergent exploration (carried over from scientific-brainstorming):

  • Cross-domain analogies (e.g. transferring trade-network analysis from economic history to Iron Age ceramic distribution)
  • Assumption reversal ("What if the chronology is the other way around?")
  • Scale shifting (single find → site → region → macro-area)
  • Interdisciplinary fusion (archaeology + epigraphy + palaeogenetics)

Phase 3 — Critical evaluation:

  • What would it take to test this? What's the smallest meaningful sub-experiment?
  • Which existing data can be exploited instead of collecting new data?
  • Who else would need to be involved (institutions, collaborators)?

Phase 4 — Write, self-review, hand off:

  • Write the design doc, run self-review, get sign-off
  • Then — and only then — invoke writing-research-plan

Self-Review Checklist (inline)

After writing the document, look at it with fresh eyes:

  1. Placeholder scan: "TBD", "to be clarified", "relevant literature" without specifics — replace or delete
  2. Scope check: Does this fit into ONE research plan, or is it begging to be decomposed?
  3. Falsifiability: Could the question in principle also be answered in the negative? If not: rephrase
  4. Ambiguity: Could "Quellenkritik" mean two different methods here? Make it explicit
  5. Scope boundary: Is there ONE explicit line "not investigated: …"? If not, add one
  6. AI-use rules: Is the AI-Use Rules section filled — or its open points named?

Fix errors inline, no second review pass.

User Review Gate

After self-review:

"Design doc written and saved to <path>. Please look it over — if changes are needed, give them now. Once you green-light it, we move to writing-research-plan."

Wait for the reply. Only invoke writing-research-plan after approval.

Key Principles

  • The user's question first — elicit their formulation before offering yours; your variants sharpen theirs, they don't replace it
  • AI-use rules on record — the venue's rules are captured in the design doc, not discovered at submission
  • One question per message — don't overload
  • Multiple choice preferred — easier to answer
  • YAGNI — cut unneeded sub-questions
  • Explore alternatives — always offer 2–3 approaches
  • Incremental validation — get sign-off section by section
  • Stay flexible — go back when something is off

示例提示词

我想申请一个新课题,帮我梳理研究问题、方法可行性和伦理考量。

所属插件

人文学科面向历史、哲学、语言学、考古与艺术研究:TEI/OCR 与数字档案、论证分析、语料库与语音学、放射性碳测年、乐谱与音频分析,以及人文论证写作与审稿。

安装该插件会一并获得同领域的 62 个 Skill / MCP。

同插件的其他科研 Skill

常见问题

研究选题构思设计能做什么?

Explores research intent, field state, methodology, and ethics, then writes an approved design doc. Use before starting a new research question or chapter。

如何安装研究选题构思设计?

在科搭 Codar 左侧点击「插件」,在「组件」页签搜索「研究选题构思设计」并安装;在 Claude Code 或 Codex 中可以发送本页的安装提示词,或下载 Skill 包后解压到 ~/.claude/skills 或 ~/.codex/skills。

研究选题构思设计需要哪些权限?

该科研 Skill没有声明额外权限;执行具体操作时仍由 Agent 按自身规则请求确认。

研究选题构思设计经过官方审核吗?

是。SpeedSkills 收录的每个版本都会记录内容摘要与权限声明,当前版本已通过安全审核。