科研 Skill

研究计划预注册撰写

writing-research-plan

Converts an approved design into a pre-registered plan with a falsifiable hypothesis and task list. Use after brainstorming-research produces a design doc.

发布者 Codar安全审核通过v0.1.0更新于 2026-11-18

SKILL.md

Writing a Research Plan

Take an approved design doc (input/ideas/<slug>-design.md) and turn it into an executable research plan (input/ideas/<slug>-plan.md). The plan is the contract between idea and execution: it pre-registers the hypothesis, locks the method, lists bite-sized tasks, and defines verification criteria.

Announce at start: "Using writing-research-plan to convert the design into a pre-registered plan."

If unmet: explain to the user which condition is missing, ask for a short reason, write it to knowledge/_meta/gate-overrides.log, and continue.

Methodology and Pre-Registration

The project frontmatter (see CLAUDE.md) declares a methodology. This skill behaves differently depending on the value:

  • hermeneutic (default for theology / exegesis / Quellenkritik / interpretive archaeology): pre-registration is not enforced. Instead the plan requires a precise research question, a method sketch (which hermeneutic procedures, on which material), and a list of expected sources. Hypothesis revision through new reading is legitimate and gets recorded in knowledge/_meta/log.md — the hermeneutic circle is constitutive, not a methodological failure.

  • quantitative (for geostatistics, 14C Bayesian work, quantitative DH methods): full pre-registration. Hypothesis and Falsification Criteria are frozen from sign-off onward. Deviations are documented; the result is explicitly marked as exploratory rather than turned into a confirmation by silent plan edits.

  • mixed: pre-registration per sub-study. Quantitative tasks get a pre-registered: true block; hermeneutic tasks don't. The plan marks this per task in the task block's frontmatter.

The plan templates below show both variants.

Checklist

  1. Read the design doc — input/ideas/<slug>-design.md. If it does not exist or is not approved, stop and invoke brainstorming-research.
  2. Lock the hypothesis — one sentence, falsifiable or argumentatively prüfbar, in the user's own words after confirmation
  3. Define falsification criteria — what observation would make you reject the hypothesis
  4. Enumerate data sources — exact sources, with status (acquire pending / on disk / already in wiki)
  5. Decompose into tasks — an Acquire-sources task (before ingest), then Ingest tasks, Analysis tasks, Synthesis tasks, Draft tasks; each step 2–15 min. Give every Draft task a realistic, generous word count — it is a floor for development, not a quota; too-tight targets are the main cause of dense, bullet-reflowed prose (see drafting-manuscript)
  6. Specify verification criteria — how we know the project is done (lint green, N sources ingested, chapter renders, …)
  7. Write plan doc to input/ideas/<slug>-plan.md using the template below
  8. Self-review — no placeholders, type consistency across tasks, spec coverage from design doc
  9. User reviews and signs off on hypothesis + task list — this is the pre-registration moment
  10. Hand off to executing-research-plan

Process Flow

digraph writing_plan {
    "Read approved design doc" [shape=box];
    "Design exists & approved?" [shape=diamond];
    "Back to brainstorming" [shape=box];
    "Lock hypothesis" [shape=box];
    "Define falsification" [shape=box];
    "Enumerate data sources" [shape=box];
    "Decompose tasks" [shape=box];
    "Specify verification" [shape=box];
    "Write plan doc" [shape=box];
    "Self-review" [shape=box];
    "User pre-registers?" [shape=diamond];
    "Invoke executing-research-plan" [shape=doublecircle];

    "Read approved design doc" -> "Design exists & approved?";
    "Design exists & approved?" -> "Back to brainstorming" [label="no"];
    "Design exists & approved?" -> "Lock hypothesis" [label="yes"];
    "Lock hypothesis" -> "Define falsification";
    "Define falsification" -> "Enumerate data sources";
    "Enumerate data sources" -> "Decompose tasks";
    "Decompose tasks" -> "Specify verification";
    "Specify verification" -> "Write plan doc";
    "Write plan doc" -> "Self-review";
    "Self-review" -> "User pre-registers?";
    "User pre-registers?" -> "Write plan doc" [label="changes"];
    "User pre-registers?" -> "Invoke executing-research-plan" [label="signed off"];
}

Plan Document Template

Save to input/ideas/<slug>-plan.md. The frontmatter status: value depends on the methodology: pre-registered (quantitative or mixed with pre-registered parts) vs. ready (hermeneutic).

Quantitative / Mixed (with pre-registration)

---
title: "<Plan title — same slug as the design>"
type: plan
created: YYYY-MM-DD
updated: YYYY-MM-DD
status: pre-registered
author: mixed
design: "[[<slug>-design]]"
---

# <Title> — Research Plan

> **For agentic workers:** REQUIRED SUB-SKILL: Use the `executing-research-plan` skill (`research-superpowers:executing-research-plan` in Claude Code) to run this plan task by task. Tasks use checkbox (`- [ ]`) syntax so each maps onto one task-list item.

**Research Question:** <precise guiding question>
**Hypothesis (pre-registered, immutable):** <one sentence>
**Falsification Criteria:** <what would falsify H>
**Stop Criterion:** <when is data collection done — sample size, time window, data availability>
**Method:** <2–3 sentences on the procedure>
**Output target:** book-chapter | article | grant-proposal | conference-talk
**Language:** de | en | mixed

---

Hermeneutic (no pre-registration)

---
title: "<Plan title — same slug as the design>"
type: plan
created: YYYY-MM-DD
updated: YYYY-MM-DD
status: ready
author: mixed
design: "[[<slug>-design]]"
---

# <Title> — Research Plan

> **For agentic workers:** REQUIRED SUB-SKILL: Use the `executing-research-plan` skill (`research-superpowers:executing-research-plan` in Claude Code) to run this plan task by task.

**Research Question:** <precise guiding question>
**Method sketch:** <which hermeneutic / source-critical / interpretive procedures, on which material>
**Expected source corpus:** <primary sources, secondary literature in keywords — may broaden through reading>
**Iteration expectation:** <which phases the plan will iterate — e.g. "repeated switching between ingest and synthesis until the source picture stabilises">
**Output target:** book-chapter | article | grant-proposal | conference-talk
**Language:** de | en | mixed

---

In both variants, Data Sources, Tasks, and Verification follow as shown below. For mixed, quantitative tasks mark pre-registered: true in the block frontmatter.

Data Sources

  • [[finkelstein-2003]] — acquire pending (not yet on disk)
  • [[mazar-2011]] — on disk, ingest pending
  • Dataset: input/data/megiddo-stratigraphy.csv
  • Archive: Israel Antiquities Authority, IAA-Nr. XXX (access: on-site visit required)

Tasks

Task 1: Acquire source PDFs

Files: <library>/pdf/*.pdf, input/bibliography/acquisition-todo.md

  • Run acquire-sources on the A+B set from literaturguide.md (auto-downloads Open-Access PDFs, writes the manual-download worklist)
  • Download any paywalled originals manually (university VPN / library proxy) into <library>/pdf/ under the exact <bibkey>.pdf filename
  • Re-run acquire-sources to reconcile until the required originals are present (or explicitly deferred)

Task 2: Ingest foundational sources

Files: knowledge/sources/*.md, output/bibtex/references.bib

  • Ingest Finkelstein 2003 via ingest-source skill → knowledge/sources/finkelstein-2003.md
  • Ingest Mazar 2011 via ingest-source skill → knowledge/sources/mazar-2011.md
  • Verify BibTeX keys match citation style (chicago-author-date.csl)

Task 3: Analyze stratigraphic sequence

Files: output/data-analysis/stratigraphy.py, output/data-analysis/results/phases.json

  • Create stratigraphy.py — loads input/data/megiddo-stratigraphy.csv
  • Compute Bayesian phase overlaps via PyMC
  • Save posterior to results/phases.json
  • Write assumption-check note in knowledge/synthesis/stratigraphy-assumptions.md

Task 4: Synthesize into a stable knowledge page

Files: knowledge/synthesis/chronologie-debatte.md

  • Pull entities [[tel-megiddo]], [[tel-rehov]], [[finkelstein]], [[mazar]]
  • Argument structure: Position A, Position B, own reading, evidence weight
  • Cite ≥ 10 sources from BibTeX
  • Set status: stable once reviewed by user

Task 5: Draft chapter "Forschungsstand — Chronology"

Files: output/book/text/03-forschungsstand.md

  • Confirm wiki-lint is green
  • Draft 2500–3500 words via drafting-manuscript skill
  • Citations [@bibkey] from output/bibtex/references.bib

Verification

The project is complete when:

  • scripts/lint-wiki.py exits 0
  • ≥ 15 sources ingested, all with status: stable or review
  • knowledge/synthesis/chronologie-debatte.md has status: stable
  • make render in output/book/ produces PDF without errors
  • Peer review round completed via requesting-peer-review
  • Hypothesis explicitly confirmed, refuted, or marked "inconclusive — exploratory"

Deviations Log

Append to knowledge/_meta/log.md whenever scope, method, or data sources change after pre-registration. Reference the plan slug.


## No Placeholders (plan failures — never write these)

- "TBD", "see later", "add relevant literature"
- "more tasks as needed" (make concrete or delete)
- Tasks without file path ("an analysis somewhere")
- Citations without BibTeX key ("probably Finkelstein")
- Verification criteria without measurability ("the chapter feels good")

## Self-Review Checklist (inline)

1. **Spec coverage:** every point from `input/ideas/<slug>-design.md` — is it covered by a task?
2. **Placeholder scan:** remove the red flags (see above)
3. **Type consistency:** if task 2 writes `phases.json`, does task 3 reference exactly that file?
4. **Falsifiability:** is the hypothesis actually testable, or just a thesis?
5. **Granularity:** are all steps 2–15 min? Split longer ones
6. **No gaps between Acquire → Ingest → Analysis → Synthesis → Draft**

Fix errors inline.

## Sign-off

After self-review:

- For **quantitative** or **mixed with pre-registered tasks**:
  > "Plan written and saved at `<path>`. Please explicitly confirm the
  > **Hypothesis** and the **Falsification Criteria** — once confirmed they
  > are frozen (pre-registered). Deviations later go into log.md. OK?"
  >
  > Only after explicit confirmation set `status: pre-registered` and
  > invoke `executing-research-plan`.

- For **hermeneutic**:
  > "Plan written and saved at `<path>`. Research question, method sketch,
  > and expected sources are recorded. Hypothesis remains open — revision
  > through reading is logged in `_meta/log.md`. Does the plan look right?
  > Then I'll set `status: ready`."
  >
  > After confirmation set `status: ready` and invoke
  > `executing-research-plan`.

## Key Principles

- **Hypothesis before data** — pre-registered
- **Bite-sized tasks** — 2–15 min, exact file path
- **Verification measurable** — not "good enough"
- **Deviation log** — document deviations, don't hide them
- **YAGNI** — only tasks that follow from the design

示例提示词

帮我把这份设计方案转化为带可证伪假设和任务清单的预注册研究计划。

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

研究计划预注册撰写能做什么?

Converts an approved design into a pre-registered plan with a falsifiable hypothesis and task list. Use after brainstorming-research produces a design doc。

如何安装研究计划预注册撰写?

在科搭 Codar 左侧点击「插件」,在「组件」页签搜索「研究计划预注册撰写」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。

研究计划预注册撰写需要哪些权限?

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。

研究计划预注册撰写经过官方审核吗?

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。