research-expert — скилл для Claude Code: методология и статистика
Скилл research-expert для ИИ-агентов: рабочий код для дизайна исследования и статистики плюс разделы про научное письмо.
Apache-2.0 256 установок ★ 45 обновлён 2026-10-12
npx skills add personamanagmentlayer/pcl --skill research-expert -a claude-codeЧто делает research-expert
Скилл даёт агенту заготовки для исследования: расчёт размера выборки, простая, блочная и стратифицированная рандомизация, описательные статистики, t-тест с Cohen's d, однофакторный ANOVA с eta-squared, линейная регрессия через statsmodels и поправка p-value на множественные сравнения. Рядом — разделы про типы исследований, от RCT до метаанализа, и про структуру рукописи.
Разрешённые инструменты — Read, Write, Edit. Код скилл пишет в файлы проекта, а запускать его вам: без numpy, scipy, pandas и statsmodels в окружении ни расчёт выборки, ни регрессия не пойдут.
Литературный обзор, заявка на исследование, оформление ссылок и публикационная этика идут текстом, без кода.
Какие задачи решает
- посчитать размер выборки до старта эксперимента
- разложить участников по группам блочной рандомизацией
- сравнить две группы t-тестом и оценить размер эффекта
- поправить p-value на множественные сравнения
- собрать рукопись по структуре IMRAD
Примеры запросов агенту
- Посчитай размер выборки для двухвыборочного t-теста с мощностью 0.8
- Сделай блочную рандомизацию участников по группам
- Прогони однофакторный ANOVA по группам и добавь eta-squared
Как установить research-expert
1Через Skills CLI (Vercel)
npx skills add personamanagmentlayer/pcl --skill research-expert -a claude-code2Вручную: положить файлы
~/.claude/skills/research-expert/SKILL.mdПапка скилла целиком; для одного проекта — .claude/skills/research-expert/
1Через Skills CLI (Vercel)
npx skills add personamanagmentlayer/pcl --skill research-expert -a codex2Вручную: положить файлы
.agents/skills/research-expert/SKILL.mdДля всех проектов — ~/.agents/skills/
1Через Skills CLI (Vercel)
npx skills add personamanagmentlayer/pcl --skill research-expert -a cursor2Вручную: положить файлы
.cursor/skills/research-expert/SKILL.mdCursor читает и .agents/skills/, и .claude/skills/
1GitHub CLI
gh skill install personamanagmentlayer/pcl stdlib/scientific/research-expert2Вручную: положить файлы
.github/skills/research-expert/SKILL.md1Через Skills CLI (Vercel)
npx skills add personamanagmentlayer/pcl --skill research-expert -a gemini-cli2Командой платформы
gemini skills install https://github.com/personamanagmentlayer/pcl --path stdlib/scientific/research-expertВручную: положить файлы
.devin/skills/research-expert/SKILL.mdСтарый путь .windsurf/skills/ тоже читается
Через Skills CLI (Vercel)
npx skills add personamanagmentlayer/pcl --skill research-expert -a opencodeClaude.ai
Упакуйте папку в zip и загрузите в Settings → Capabilities → Skills
По шагам, с командами для каждой платформы: Как установить скилл в Claude Code: пять способов
Вопросы о research-expert
Как установить research-expert в Claude Code?
Выполните npx skills add personamanagmentlayer/pcl --skill research-expert -a claude-code, затем скилл лежит в ~/.claude/skills/research-expert/SKILL.md.
Что делает research-expert?
Подсказывает агенту, как спланировать исследование и обработать данные: расчёт выборки, рандомизация, описательные статистики, тесты, регрессия. Отдельно идут советы по научному письму.
Что нужно, чтобы код из скилла заработал?
Python с numpy, scipy, pandas и statsmodels. Сам скилл умеет только читать и править файлы, счёт запускаете вы.
Исходный текст
SKILL.md · Apache-2.0
---
name: research-expert
version: 1.1.0
description: >-
Expert-level research methodology, academic writing, statistical analysis, and scientific
investigation. Use when the user mentions methodology, statistics, academic writing, or
experimental design, or when the task involves Research Design, Statistical Analysis, or
Data Analysis.
category: scientific
tags: [research, methodology, statistics, academic-writing, experimental-design]
allowed-tools:
- Read
- Write
- Edit
---
# Research Methodology Expert
Expert guidance for research methodology, experimental design, statistical analysis, and academic writing.
## Core Concepts
### Research Design
- Experimental vs observational studies
- Randomized controlled trials (RCTs)
- Cross-sectional, longitudinal, cohort studies
- Case-control studies
- Systematic reviews and meta-analysis
- Sample size determination
### Statistical Analysis
- Descriptive statistics
- Hypothesis testing
- Confidence intervals
- Regression analysis
- ANOVA and t-tests
- Non-parametric tests
- Multiple testing correction
### Academic Writing
- Literature review
- Research proposals
- Manuscript structure (IMR AD)
- Citation management
- Peer review process
- Publishing ethics
## Experimental Design
```python
from dataclasses import dataclass
from typing import List, Optional
import numpy as np
from scipy import stats
@dataclass
class Study:
name: str
design_type: str # 'RCT', 'observational', 'cohort'
sample_size: int
groups: List[str]
primary_outcome: str
secondary_outcomes: List[str]
class SampleSizeCalculator:
"""Calculate required sample size for studies"""
@staticmethod
def two_sample_ttest(effect_size: float, alpha: float = 0.05,
power: float = 0.8) -> int:
"""Calculate sample size for two-sample t-test"""
from statsmodels.stats.power import tt_ind_solve_power
n = tt_ind_solve_power(
effect_size=effect_size,
alpha=alpha,
power=power,
alternative='two-sided'
)
return int(np.ceil(n))
@staticmethod
def proportion_test(p1: float, p2: float, alpha: float = 0.05,
power: float = 0.8) -> int:
"""Calculate sample size for comparing proportions"""
from statsmodels.stats.power import zt_ind_solve_power
effect_size = (p2 - p1) / np.sqrt(p1 * (1 - p1))
n = zt_ind_solve_power(
effect_size=effect_size,
alpha=alpha,
power=power,
alternative='two-sided'
)
return int(np.ceil(n))
class ExperimentalDesign:
"""Design and randomize experimental studies"""
def __init__(self, n_subjects: int, n_groups: int):
self.n_subjects = n_subjects
self.n_groups = n_groups
def simple_randomization(self) -> List[int]:
"""Simple random assignment to groups"""
return np.random.choice(self.n_groups, size=self.n_subjects)
def block_randomization(self, block_size: int) -> List[int]:
"""Block randomization for balanced groups"""
n_blocks = self.n_subjects // block_size
assignments = []
for _ in range(n_blocks):
block = np.repeat(range(self.n_groups),
block_size // self.n_groups)
np.random.shuffle(block)
assignments.extend(block)
# Handle remaining subjects
remainder = self.n_subjects % block_size
if remainder > 0:
extra = np.random.choice(self.n_groups, size=remainder)
assignments.extend(extra)
return assignments
def stratified_randomization(self, strata: List[str]) -> List[int]:
"""Stratified randomization by covariates"""
assignments = np.zeros(self.n_subjects, dtype=int)
for stratum in set(strata):
stratum_indices = [i for i, s in enumerate(strata) if s == stratum]
stratum_n = len(stratum_indices)
stratum_assignments = np.random.choice(
self.n_groups,
size=stratum_n,
replace=True
)
for idx, assignment in zip(stratum_indices, stratum_assignments):
assignments[idx] = assignment
return assignments
```
## Statistical Analysis
```python
import pandas as pd
from scipy import stats
import statsmodels.api as sm
from statsmodels.stats.multitest import multipletests
class StatisticalAnalysis:
"""Perform statistical analyses"""
@staticmethod
def descriptive_stats(data: pd.Series) -> dict:
"""Calculate descriptive statistics"""
return {
"mean": data.mean(),
"median": data.median(),
"std": data.std(),
"min": data.min(),
"max": data.max(),
"q25": data.quantile(0.25),
"q75": data.quantile(0.75),
"skewness": stats.skew(data),
"kurtosis": stats.kurtosis(data)
}
@staticmethod
def independent_ttest(group1: np.ndarray, group2: np.ndarray) -> dict:
"""Perform independent samples t-test"""
statistic, pvalue = stats.ttest_ind(group1, group2)
# Calculate effect size (Cohen's d)
pooled_std = np.sqrt((group1.var() + group2.var()) / 2)
cohens_d = (group1.mean() - group2.mean()) / pooled_std
return {
"t_statistic": statistic,
"p_value": pvalue,
"cohens_d": cohens_d,
"mean_diff": group1.mean() - group2.mean(),
"significant": pvalue < 0.05
}
@staticmethod
def one_way_anova(groups: List[np.ndarray]) -> dict:
"""Perform one-way ANOVA"""
f_statistic, p_value = stats.f_oneway(*groups)
# Calculate effect size (eta-squared)
grand_mean = np.mean(np.concatenate(groups))
ss_between = sum(len(g) * (g.mean() - grand_mean)**2 for g in groups)
ss_total = sum(((g - grand_mean)**2).sum() for g in groups)
eta_squared = ss_between / ss_total
return {
"f_statistic": f_statistic,
"p_value": p_value,
"eta_squared": eta_squared,
"significant": p_value < 0.05
}
@staticmethod
def linear_regression(X: pd.DataFrame, y: pd.Series) -> dict:
"""Perform linear regression"""
X_with_const = sm.add_constant(X)
model = sm.OLS(y, X_with_const).fit()
return {
"coefficients": model.params.to_dict(),
"r_squared": model.rsquared,
"adj_r_squared": model.rsquared_adj,
"f_statistic": model.fvalue,
"p_value": model.f_pvalue,
"summary": model.summary()
}
@staticmethod
def multiple_testing_correction(p_values: List[float],
method: str = 'fdr_bh',
alpha: float = 0.05) -> dict:
"""Apply multiple testing correction"""
reject, pvals_corrected, alphacSidak, alphacBonf = multipletests(
p_values,
alpha=alpha,
method=method
)
return {
"rejected": reject,
"corrected_pvalues": pvals_corrected,
"n_significant": reject.sum(),
"method": method
}
class EffectSize:
"""Calculate effect sizes"""
@staticmethod
def cohens_d(group1: np.ndarray, group2: np.ndarray) -> float:
"""Cohen's d for two groups"""
pooled_std = np.sqrt((group1.var() + group2.var()) / 2)
return (group1.mean() - group2.mean()) / pooled_std
@staticmethod
def hedges_g(group1: np.ndarray, group2: np.ndarray) -> float:
"""Hedges' g (corrected effect size)"""
n1, n2 = len(group1), len(group2)
df = n1 + n2 - 2
correction = 1 - (3 / (4 * df - 1))
d = EffectSize.cohens_d(group1, group2)
return d * correction
@staticmethod
def r_squared_to_cohens_f(r_squared: float) -> float:
"""Convert R² to Cohen's f"""
return np.sqrt(r_squared / (1 - r_squared))
```
## Literature Review
```python
from typing import Dict, List
from dataclasses import dataclass
@dataclass
class Citation:
authors: List[str]
year: int
title: str
journal: str
volume: Optional[int] = None
pages: Optional[str] = None
doi: Optional[str] = None
def format_apa(self) -> str:
"""Format citation in APA style"""
authors_str = self._format_authors_apa()
citation = f"{authors_str} ({self.year}). {self.title}. {self.journal}"
if self.volume:
citation += f", {self.volume}"
if self.pages:
citation += f", {self.pages}"
if self.doi:
citation += f". https://doi.org/{self.doi}"
return citation + "."
def _format_authors_apa(self) -> str:
"""Format authors in APA style"""
if len(self.authors) == 1:
return self.authors[0]
elif len(self.authors) == 2:
return f"{self.authors[0]} & {self.authors[1]}"
else:
return f"{self.authors[0]} et al."
class LiteratureReview:
"""Manage literature review"""
def __init__(self):
self.citations: List[Citation] = []
self.themes: Dict[str, List[Citation]] = {}
def add_citation(self, citation: Citation, themes: List[str]):
"""Add citation and categorize by themes"""
self.citations.append(citation)
for theme in themes:
if theme not in self.themes:
self.themes[theme] = []
self.themes[theme].append(citation)
def get_bibliography(self, style: str = 'apa') -> List[str]:
"""Generate bibliography"""
if style == 'apa':
return [c.format_apa() for c in sorted(
self.citations,
key=lambda x: (x.authors[0], x.year)
)]
def get_summary_by_theme(self, theme: str) -> List[Citation]:
"""Get citations for specific theme"""
return self.themes.get(theme, [])
```
## Best Practices
### Research Design
- Pre-register studies when possible
- Calculate adequate sample sizes
- Use appropriate controls
- Randomize when applicable
- Blind assessors to reduce bias
- Consider confounding variables
- Document protocol deviations
### Data Analysis
- Pre-specify analysis plan
- Check statistical assumptions
- Report effect sizes, not just p-values
- Apply multiple testing corrections
- Use appropriate statistical tests
- Report confidence intervals
- Make data and code available
### Academic Writing
- Follow journal guidelines
- Use clear, precise language
- Report methodology in detail
- Discuss limitations openly
- Acknowledge conflicts of interest
- Properly cite all sources
- Use reference management software
## Anti-Patterns
❌ P-hacking and data dredging
❌ HARKing (Hypothesizing After Results are Known)
❌ Cherry-picking results
❌ Inadequate sample sizes
❌ Ignoring failed experiments
❌ No pre-registration
❌ Selective reporting of outcomes
## Resources
- ClinicalTrials.gov: https://clinicaltrials.gov/
- CONSORT Statement: http://www.consort-statement.org/
- Cochrane Handbook: https://training.cochrane.org/handbook
- APA Style: https://apastyle.apa.org/
- StatsModels: https://www.statsmodels.org/
© personamanagmentlayer · Apache-2.0 · personamanagmentlayer/pcl