К основному содержанию
скилл CX Codex

research-craft — скилл для Claude Code: что делает и как установить

Скилл research-craft для Claude Code помогает формулировать научные проблемы, проверять гипотезы и структурировать диссертации или статьи.

MIT 153 установок ★ 274 обновлён 2026-10-30

Установить в Claude Code · Через Skills CLI (Vercel) Все способы и платформы ↓
npx skills add curiositech/some_claude_skills --skill research-craft -a claude-code
Открыть на GitHub

Что делает research-craft

Скилл формализует академическую работу через систему строгих шагов. Он помогает превратить общую тему в исследовательскую проблему по схеме «условие пробела в знаниях — практические последствия — доступные доказательства». Проблемы делятся на концептуальные и практические, что исключает бессистемное накопление данных.

Внутри заложены четыре дерева решений: формулирование проблемы, построение аргументации (тезис, основания, свидетельства, возражения), оценка источников (первичные, вторичные, третичные) и пятиуровневая вычитка черновика от общей композиции до формулировок предложений. Набор инструментов ограничен операцией Read.

Методология создана исключительно для научных публикаций, диссертаций и академических обзоров литературы. Для художественного текста, журналистики или бизнес-кейсов она непригодна из-за строгих требований к академической аргументации.

Какие задачи решает

  • сформулировать проблему исследования через пробел в знаниях и последствия
  • построить доказательную цепочку из тезиса, оснований и свидетельств
  • провести пятиуровневую вычитку структуры научной статьи
  • оценить надёжность и первичность академических источников

Примеры запросов агенту

  • Помоги сформулировать проблему для диссертации по теме распределённых систем
  • Проверь аргументацию во второй главе статьи по дереву решений research-craft
  • Оцени черновик введения по структуре контекст-проблема-ответ

Как установить research-craft

1Через Skills CLI (Vercel)

npx skills add curiositech/some_claude_skills --skill research-craft -a claude-code

2Вручную: положить файлы

~/.claude/skills/research-craft/SKILL.md

Папка скилла целиком; для одного проекта — .claude/skills/research-craft/

По шагам, с командами для каждой платформы: Как установить скилл в Claude Code: пять способов

Вопросы о research-craft

Как установить research-craft в Claude Code?

Выполните команду npx skills add curiositech/some_claude_skills --skill research-craft -a claude-code. Файл навыка запишется в ~/.claude/skills/research-craft/SKILL.md.

Что делает скилл research-craft?

Он предоставляет пошаговые деревья решений для академического письма: помогает поставить научный вопрос, сбалансировать эмпирические данные с интерпретацией и выстроить логику статьи.

Для каких текстов скилл не подходит?

Скилл не предназначен для художественной литературы, журналистских репортажей, маркетинговых материалов и бизнес-аналитики.

Исходный текст

SKILL.md · MIT
---
name: research-craft
description: Systematic research methodology for formulating problems, building evidence-based arguments, and communicating findings in academic contexts. For scholarly writing, thesis development, and
  knowledge contribution. NOT for creative writing, journalistic reporting, or business analytics.
allowed-tools: Read
metadata:
  category: Research & Analysis
  tags:
  - research
  - craft
  pairs-with:
  - skill: research-analyst
    reason: Research craft provides the methodology that research analysis applies to specific domains
  - skill: technical-writer
    reason: Research findings are communicated through technical writing with proper citations
  - skill: munger-worldly-wisdom
    reason: Mental model frameworks complement systematic research methodology for deeper analysis
---

# Research Craft
Master the systematic process of formulating meaningful research problems, constructing evidence-based arguments, and communicating findings that contribute to scholarly conversations.

## When to Use
✅ Use for: Academic research papers, thesis/dissertation work, scholarly article writing, literature reviews, evidence-based argumentation, knowledge synthesis, participating in disciplinary conversations
❌ NOT for: Creative writing, journalistic reporting, business case studies, technical documentation, marketing content, personal essays, opinion pieces

## Core Process

### Problem Formulation Decision Tree
```
START: Do you have a clear research problem?
├─ NO: Do you have a specific topic (not just broad subject)?
│  ├─ NO → Find specific topic within general interest area
│  └─ YES → Question topic systematically
│     └─ Ask "So what?" to test consequences
│        ├─ Readers won't care → Reframe problem from reader perspective
│        └─ Readers will care → Verify problem structure:
│           ├─ Condition (knowledge gap) stated? 
│           ├─ Cost/consequence stated?
│           └─ Evidence type available?
│              ├─ ALL YES → PROCEED to argument building
│              └─ ANY NO → Revise problem formulation
└─ YES → Is problem practical or conceptual?
   ├─ PRACTICAL (tangible costs requiring action)
   │  └─ Solution must be actionable
   └─ CONCEPTUAL (understanding gap)
      ├─ Pure research → Solution improves understanding
      └─ Applied research → Solution has practical consequences
```

### Argument Construction Decision Tree
```
START: Building argument for research claim
├─ State CLAIM (answers research question)
├─ Develop REASONS (assertions from thinking)
│  └─ For each reason, do you have EVIDENCE (data from world)?
│     ├─ NO → Insufficient support; gather evidence or drop reason
│     └─ YES → Is evidence representative, accurate, sufficient?
│        ├─ NO → Cherry-picking risk; expand sample or qualify claim
│        └─ YES → Continue
├─ Anticipate reader objections
│  ├─ Plausible objections exist?
│  │  ├─ YES → ACKNOWLEDGE and RESPOND with subordinate argument
│  │  │  ├─ Can you refute? → Provide counter-evidence
│  │  │  ├─ Must concede? → Acknowledge limitation candidly
│  │  │  └─ Alternative view? → Show why yours is stronger
│  │  └─ NO → Proceed to warrants
│  └─ Will readers question reason-claim connection?
│     ├─ YES → STATE WARRANT (general principle)
│     └─ NO (expert audience) → Omit warrant
└─ Test complete argument
   └─ Evidence comprises ~1/3 of section?
      ├─ NO: Too little → Data dump; too much → Add interpretation
      └─ YES → ARGUMENT COMPLETE
```

### Source Evaluation Decision Tree
```
START: Evaluating potential source
├─ Assess RELIABILITY
│  ├─ Peer-reviewed? → +reliability
│  ├─ Frequently cited? → +reliability  
│  ├─ Reputable publisher? → +reliability
│  ├─ Author credentialed in domain? → +reliability
│  └─ Current for field? → Check temporal standards
├─ Classify SOURCE TYPE (relative to YOUR project)
│  ├─ Provides raw data on subject? → PRIMARY
│  ├─ Analyzes primary sources? → SECONDARY
│  └─ Synthesizes secondary sources? → TERTIARY
│     └─ Can you access more primary sources?
│        ├─ YES → Get closer to primary; chain is too long
│        └─ NO → Acknowledge limitation
├─ Evaluate EVIDENCE quality
│  ├─ Accurate? → Verify important claims
│  ├─ Precise? → Sufficient detail for replication?
│  ├─ Sufficient? → Sample size adequate?
│  ├─ Representative? → Generalizable?
│  └─ Authoritative? → Within author's expertise?
└─ USAGE decision
   ├─ Quote? → Exact words matter for analysis
   ├─ Paraphrase? → Specific ideas in your words
   └─ Summarize? → General overview sufficient
```

### Revision Decision Tree
```
START: Draft complete, begin revision (TOP-DOWN only)
├─ LEVEL 1: FRAME (Introduction/Conclusion)
│  ├─ Introduction has Context→Problem→Response?
│  │  ├─ NO → Restructure using three-step pattern
│  │  └─ YES → Problem framed from reader perspective?
│  │     ├─ NO → Rewrite emphasizing costs to readers
│  │     └─ YES → Continue
│  └─ Conclusion has Main Point→New Significance→Future Research?
│     ├─ NO → Add missing elements
│     └─ YES → Proceed to Level 2
├─ LEVEL 2: ARGUMENT STRUCTURE  
│  ├─ 4-5 key concepts run throughout paper?
│  │  ├─ NO → Identify concepts; track through all sections
│  │  └─ YES → Continue
│  └─ Main claim clearly supported by reasons?
│     ├─ NO → Revise logical architecture
│     └─ YES → Proceed to Level 3
├─ LEVEL 3: SECTION ORGANIZATION
│  ├─ Each section has point sentence (beginning or end)?
│  │  ├─ NO → Add or clarify point sentences
│  │  └─ YES → Continue
│  ├─ Evidence comprises ~1/3 of each section?
│  │  ├─ NO → Adjust evidence/interpretation ratio
│  │  └─ YES → Continue
│  └─ Ordering principle clear (simple→complex, familiar→unfamiliar)?
│     ├─ NO → Reorganize sections logically
│     └─ YES → Proceed to Level 4
├─ LEVEL 4: PARAGRAPH STRUCTURE
│  ├─ Each paragraph advances single point?
│  │  ├─ NO → Split or merge paragraphs
│  │  └─ YES → Continue
│  └─ Transitions guide readers between paragraphs?
│     ├─ NO → Add connective tissue
│     └─ YES → Proceed to Level 5
└─ LEVEL 5: SENTENCE CLARITY
   ├─ Subjects = main characters?
   ├─ Verbs = crucial actions (not nominalizations)?
   └─ Hedges appropriately qualify certainty?
      ├─ ALL YES → REVISION COMPLETE
      └─ ANY NO → Fix sentence-level issues
```

## Anti-Patterns

### Data Dump
**Novice approach:** Accumulates information without guiding question, creating pastiche of vaguely related facts. Presents evidence excessively without interpretation.

**Expert approach:** Formulates clear research question before gathering evidence. Balances evidence (~1/3 of section) with reasons and interpretation showing how evidence supports claims.

**Timeline to mastery:** Internalized after 2-3 major research projects where early aimless collection causes crisis, forcing recognition that questions must precede answers.

---

### Warrant-Based Factual Claims  
**Novice approach:** Supports factual claims with elaborate reasoning and logical principles alone, without presenting hard evidence from the world.

**Expert approach:** Distinguishes sharply between reasons (assertions from thinking) and evidence (data from world). Bases reasons on evidence, never evidence on reasons. Reserves warrant-heavy arguments for conceptual claims only.

**Timeline to mastery:** Requires shift from high school essay mode (where reasoning suffices) to research mode (where data is king). Typically 1-2 semesters of feedback cycles.

---

### Arrogant Certainty
**Novice approach:** States claims with absolute confidence using "all," "never," "always," "proves." Hides weaknesses hoping readers won't notice. Thinks forceful = credible.

**Expert approach:** Paradoxically strengthens arguments by modestly acknowledging limitations. Uses hedges appropriately. States claims precisely matching what evidence supports. Recognizes that candid acknowledgment builds ethos.

**Timeline to mastery:** Counter-intuitive reversal that clicks when first experiencing how acknowledged limitations increase rather than decrease reader trust. Often requires seeing own overreaching claims rejected.

---

### Patch Writing
**Novice approach:** Stitches together source passages with minimal connecting sentences. Organizes paper following major source's structure. Demonstrates knowledge acquisition without original synthesis.

**Expert approach:** Uses sources to engage ongoing conversations through agreement, disagreement, or extension. Develops independent organizational structure. Shows what you can DO with information, not just that you found it.

**Timeline to mastery:** Requires role reversal from student-as-examiner to researcher-as-expert. Typically emerges when first making genuine discovery that contradicts or extends existing literature.

---

### Cherry-Picking Evidence
**Novice approach:** Uses one or two convenient examples appearing representative but actually anecdotal. Presents only supporting evidence, ignoring contradictory data. Hopes unrepresentativeness goes unnoticed.

**Expert approach:** Demonstrates representativeness through adequate samples. Acknowledges contradictory evidence and explains why interpretation remains valid, or candidly admits limitations. Recognizes arguments are only as strong as weakest evidence.

**Timeline to mastery:** Internalized after first experience having reviewer demolish argument by citing contradictory evidence you knew existed but ignored. Painful but permanent lesson.

---

### Student-Teacher Dynamic
**Novice approach:** Writes to demonstrate knowledge to teacher-as-examiner. Explains what assignment asked for. Uses phrases like "In this paper I will show..." Treats research as proving you did the work.

**Expert approach:** Positions self as expert sharing findings with interested peers. Reverses roles so researcher is authority with something new. Readers are colleagues who care about discoveries, not examiners grading performance.

**Timeline to mastery:** Critical threshold crossed when making first genuine contribution—discovering something even instructor doesn't know. Often coincides with first conference presentation or publication.

## Mental Models & Shibboleths

### "Research is conversation spanning millennia"
True researchers recognize they're joining ongoing dialogue, not discovering from scratch. They trace bibliographic trails backward (what sources cite) and forward (what cites sources), understanding each contribution responds to previous voices while anticipating future responses.

**Shibboleth usage:** "I'm following the bibliographic trail from Smith's primary sources" (expert) vs. "I found five sources on my topic" (novice).

---

### "Evidence is bedrock; reasons are built structures"
Evidence must be found in the world and made visible to readers—it stands alone. Reasons are assertions thought up by minds that require their own foundation. We base reasons on evidence, never evidence on reasons.

**Shibboleth usage:** "My reason needs stronger evidence" (expert) vs. "My evidence shows that..." when citing an assertion (novice confusing terms).

---

### "Thickening arguments through acknowledgment"
Strong arguments aren't thin straight lines from claim to evidence—they're thickened by acknowledging objections, alternative views, and limitations. Paradoxically, acknowledging what you don't know makes claims about what you do know more credible.

**Shibboleth usage:** "I need to thicken this section by acknowledging the Copenhagen interpretation" (expert understanding ethos) vs. "I won't mention that counterexample" (novice thinking concealment = strength).

---

### "Research as zigzagging through fog"
Research doesn't follow linear path from topic→thesis→notes→draft. It's recursive, uncertain, messy—searching for something you won't recognize until you see it. Plans guide but don't constrain.

**Shibboleth usage:** "My problem shifted as I discovered..." (expert accepting non-linearity) vs. "I can't start writing until I know exactly what I'll say" (novice expecting false certainty).

---

### "Primary/secondary sources are relative, not absolute"
Source classification depends on YOUR research project, not inherent properties. Same source can be primary for one study, secondary for another. Key question: How many steps removed from phenomena you're studying?

**Shibboleth usage:** "For my purposes, this is a primary source because..." (expert understanding relativity) vs. "This is a secondary source" (novice treating as fixed category).

---

### "Problem significance measured by reader's cost, not researcher's interest"
The worst response isn't "I disagree" but "I don't care." Problems must be framed from reader's perspective showing consequences THEY care about, not just explaining your curiosity.

**Shibboleth usage:** "Readers in my field care about this because..." (expert framing) vs. "I'm interested in this topic because..." (novice framing).

## Temporal Knowledge

**19th century:** Primary/secondary source distinction originated with historians establishing evidence reliability; later became standard across disciplines as research professionalized.

**Early experimental science (~1700s):** Researchers relied on witnesses to attest experimental accuracy before standardized methodologies emerged as modern replacement for verification.

**Pre-Internet era (pre-1990s):** Citation indexing required hours/days with printed indexes. Card catalogs were standard. Bibliographic trail-following was laborious physical process.

**Contemporary era (post-2000):** Online databases make citation indexing instant, but information overload and unmonitored Internet content paradoxically make libraries and expert curation MORE essential, not less. Google ubiquity creates false confidence in research reliability.

**Field-specific temporal standards:** Currency requirements vary dramatically—molecular biology may require sources from last 2-3 years, while history may heavily cite centuries-old primary sources. Recognize your field's temporal norms.

## References
- Source: *The Craft of Research* (4th edition) by Wayne C. Booth, Gregory G. Colomb, Joseph M. Williams, Joseph Bizup, and William T. FitzGerald

© curiositech · MIT · curiositech/some_claude_skills

Похожие скиллы

Все скиллы по теме «Наука и исследования» →
research-expert personamanagmentlayer Скилл research-expert для ИИ-агентов: рабочий код для дизайна исследования и статистики плюс разделы про научное письмо. CX Codex +1 256установок academic-research-suite Imbad0202 Скилл academic-research-suite маршрутизирует задачи академического пайплайна: от обзора литературы и проверки цитирования до симуляции рецензирования. Запускается по слэш-командам и ключевым фразам. CX Codex +1 2 200установок thesis-advisor genspark-ai Скилл thesis-advisor собирает план диплома, черновики глав или каркас обзора литературы в документ Word на основе ваших материалов. Данные и выводы остаются за вами. CX Codex +1 ★ 12 746звёзд репо academic-research akillness Скилл academic-research маршрутизирует запрос по теме исследования или тексту статьи в одну из четырёх смежных методик — от поиска литературы до рецензирования. Решение, какую ветку и режим запускать, принимает сам скилл. CX Codex +1 148установок research-writing-literature hanhuark Скилл research-writing-literature помогает выстраивать строгую научную аргументацию, готовить обзоры литературы, описывать методы и разбирать графики в статьях. CX Codex +1 ★ 26звёзд репо research-agent-skills Kalaris Labs Плагин research-agent-skills добавляет 281 скилл для научных исследований в Claude Code и Codex. Охватывает весь цикл: от поиска литературы и выдвижения гипотез до проверки цитат и ответов рецензентам. CX Codex ★ 1 596звёзд репо

Из того же набора180

Весь набор →
task-decomposer Скилл task-decomposer разбирает текстовое описание проблемы на подзадачи для узлов DAG, находя зависимости и этапы для параллельного запуска. CX Codex +1 155установок caching-strategies Скилл caching-strategies помогает проектировать многоуровневое кэширование веб-приложений и выбирать стратегию для Redis, CDN, браузера или памяти. CX Codex +1 154установок checklist-discipline Скилл checklist-discipline для Claude Code превращает разрозненные знания в рабочие чек-листы для сложных процессов, где сбои внимания неизбежны. CX Codex +1 154установок mermaid-graph-writer Скилл mermaid-graph-writer генерирует синтаксис схем Mermaid по текстовым описаниям или коду. Помогает визуализировать переходы состояний, архитектуру и потоки данных. CX Codex +1 154установок database-design-patterns Скилл database-design-patterns проектирует схемы реляционных БД: нормализует таблицы до 3NF, подбирает типы индексов и планирует миграции без простоя сервиса. CX Codex +1 152установок mermaid-graph-renderer Скилл mermaid-graph-renderer для рендера Mermaid-диаграмм в SVG, PNG и PDF. Подскажет, как настроить ленивую загрузку на сайте, экспорт из CLI и выбор библиотеки. CX Codex +1 152установок logging-observability Скилл logging-observability настраивает структурированное логирование, сквозную трассировку через OpenTelemetry и пороги алертов. Помогает связать логи сервисов и убрать лишний шум из оповещений. CX Codex +1 152установок websocket-streaming Скилл websocket-streaming для Claude Code настраивает двусторонний обмен данными по WebSocket между DAG-движком и дашбордом. Передаёт события выполнения и живое состояние графа. CX Codex +1 152установок

Спонсоры проекта

Компании, которые помогают держать аналитику открытой