replicate-paper
安装方式
先用 Prompt 核对来源,也可切换到安装命令或下载可用的安装包。
用 AI 助手安装
复制提示词,让助手先核对固定版本和依赖。
查看 Prompt收起 Prompt
先检查当前 Agent 是否已安装 SkillPI Skill Finder。如果尚未安装,请按官方仓库的说明安装并启用:https://github.com/Findskills/skillpi。然后使用 Skill Finder 打开此 SkillPI 详情页并定位 "replicate-paper":https://skillpi.com/zh/skills/deepskill-78c4247629f0a1916cac15f8e2a6ba6dfc7d4570e46d747343f09a867c6909ef。安装前核对上游来源和版本:https://github.com/brycewang-stanford/Auto-Empirical-Research-Skills/blob/bfa0a6e5a843babbf45a1ef1663fd5dc4f07c0ba/skills/28-maxwell2732-paper-replicate-agent-demo/dot-claude/skills/replicate-paper/SKILL.md。来源不一致时停止并说明差异。核对通过后,用 Skill Finder 安装目标 Skill,并报告安装路径和来源版本。
此命令从 GitHub 来源仓库获取指定技能。运行前请先查看源文件。
npx skills add https://github.com/brycewang-stanford/Auto-Empirical-Research-Skills/tree/bfa0a6e5a843babbf45a1ef1663fd5dc4f07c0ba/skills/28-maxwell2732-paper-replicate-agent-demo/dot-claude/skills/replicate-paper --skill replicate-paper来源与配置
查看 GitHub 源码brycewang-stanford/Auto-Empirical-Research-Skills固定提交 bfa0a6e5a843babbf45a1ef1663fd5dc4f07c0ba
配置方式待核实
请查看原始来源中的适用工具、依赖和权限要求。
安装实测待补充
本条目面向文献与学术工具场景,提供“replicate-paper”技能的目录入口。公开标签:learning、论文复现、生物医学、流行病学。请通过固定上游版本核对实际功能、依赖与权限;此页为 Skillπ 独立编写的用途介绍和原站链接,不代表运行安全保证。原文与附件依各自许可开放。
文件浏览
---
name: replicate-paper
description: "Run a full 6-phase autonomous replication of a biomedical/epidemiology paper against UK Biobank or similar cohort data, producing Python and R scripts plus a validated replication report. Use when asked to replicate a paper end-to-end, or when invoked as /replicate-paper [paper.pdf] [data.csv|dta]."
allowed-tools: ["Read", "Write", "Edit", "Bash", "Glob", "Grep"]
---
# Skill: /replicate-paper
**Trigger:** `/replicate-paper [paper.pdf] [data.csv|dta]` or "replicate this paper"
**Purpose:** Full 6-phase autonomous replication of a biomedical/epidemiology paper using UK Biobank or similar data. Produces Python and R scripts plus a polished validation report.
---
## Invocation
```
/replicate-paper papers/AuthorYear.pdf data/ukb_extract.csv
```
Or with just: "replicate this paper" (Claude will ask for paths if not provided).
---
## The 6-Phase Pipeline
### Phase 1: Intake
**Goal:** Understand exactly what needs to be replicated.
1. Read the paper PDF (all sections: Abstract, Methods, Results, Supplementary)
2. Identify **every table and figure** that presents empirical results
3. For each: record the gold standard values, SEs/CIs, sample sizes, and source location
4. Save targets to `quality_reports/[paper_name]_replication_targets.md`
5. Summarize: original software, data source, sample N, key methods, any replication package available
**Output:** `quality_reports/[paper_name]_replication_targets.md`
---
### Phase 2: Data Audit
**Goal:** Confirm what we can and cannot replicate given the available data.
1. Load the provided dataset (`data/[filename]`)
2. Compare to paper's described sample:
- Total N, exposed N, event counts
- Key variable distributions
- Missing data patterns
3. Apply inclusion/exclusion criteria as stated in Methods; document each step's effect on N
4. If variables are missing or differently named: document the gap; flag as a known discrepancy
5. Save audit summary to `quality_reports/[paper_name]_data_audit.md`
**Output:** `quality_reports/[paper_name]_data_audit.md`
---
### Phase 3: Code Analysis
**Goal:** Map the paper's methods to our dataset before writing a single line of code.
1. Read original Stata/R code (if provided in replication package)
2. Map each variable name in original code → corresponding variable in our dataset
3. Identify methodological steps: sample construction, covariate coding, model fitting, SE clustering
4. Flag any steps where original code differs from Methods text (use the paper, not the code, as ground truth)
5. Document the mapping in `quality_reports/[paper_name]_variable_map.md`
**Output:** `quality_reports/[paper_name]_variable_map.md`
---
### Phase 4: Translation
**Goal:** Produce clean, reproducible Python and R scripts that implement the paper's analysis.
**Rules:**
- Line-by-line translation first — **no improvements during replication**
- Follow `python-code-conventions.md` and `r-code-conventions.md` exactly
- Set seed: `random.seed(YYYYMMDD)` + `numpy.random.seed(YYYYMMDD)` (Python); `set.seed(YYYYMMDD)` (R)
- Use `pathlib.Path` for all Python paths; `here::here()` for all R paths
- Comment every non-obvious Stata→Python or Stata→R translation decision
- Refer to `replication-protocol.md` translation pitfall tables
**Python script:** `replications/[paper_name]/python/replicate.py`
Structure:
```python
# Replication: [Paper Author (Year)]
# Date: YYYY-MM-DD
# Original: Stata / R
# Python version: X.Y.Z
# Key packages: pandas X.X, statsmodels X.X, lifelines X.X
from pathlib import Path
import random
import numpy as np
import pandas as pd
# ... other imports
random.seed(YYYYMMDD)
np.random.seed(YYYYMMDD)
DATA_DIR = Path(__file__).parents[3] / "data"
RESULTS_DIR = Path(__file__).parent / "results"
RESULTS_DIR.mkdir(exist_ok=True)
# --- 1. Load Data ---
# --- 2. Sample Construction ---
# --- 3. Model Fitting ---
# --- 4. Save Results ---
```
**R script:** `replications/[paper_name]/R/replicate.R`
Structure:
```r
# Replication: [Paper Author (Year)]
# Date: YYYY-MM-DD
# Original: Stata / Python
# R version: X.Y.Z
# Key packages: survival X.X, fixest X.X
library(here)
library(tidyverse)
library(survival)
# ... other packages
set.seed(YYYYMMDD)
data_dir <- here("data")
results_dir <- here("replications", "[paper_name]", "R", "results")
dir.create(results_dir, recursive = TRUE, showWarnings = FALSE)
# --- 1. Load Data ---
# --- 2. Sample Construction ---
# --- 3. Model Fitting ---
# --- 4. Save Results ---
```
**Outputs:**
- `replications/[paper_name]/python/replicate.py`
- `replications/[paper_name]/R/replicate.R`
- `replications/[paper_name]/python/results/` (parquet/pkl files)
- `replications/[paper_name]/R/results/` (rds files)
---
### Phase 5: Validation
**Goal:** Run both scripts and compare results to gold standard targets.
1. Execute Python script: `python replications/[paper_name]/python/replicate.py`
2. Execute R script: `Rscript replications/[paper_name]/R/replicate.R`
3. Load results; compare to targets using tolerance thresholds from `replication-protocol.md`:
- Integers: exact
- Point estimates: ±0.01
- SEs: ±0.05
- P-values: same significance bracket
- Percentages: ±0.1pp
4. For each mismatch: investigate root cause before proceeding
5. Save `replications/[paper_name]/validation_report.md`
**Output:** `replications/[paper_name]/validation_report.md`
---
### Phase 6: Report
**Goal:** Produce a polished, self-contained replication report.
Report structure:
```markdown
# Replication Report: [Paper Author (Year)]
**Date:** [YYYY-MM-DD]
**Replicator:** Claude (domain-reviewer verified)
## Paper Summary
[1 paragraph: research question, population, exposure, outcome, key finding]
## Methods Summary
[Bullet list: sample, exclusions, covariates, model, SEs, software]
## Data
[Bullet list: our dataset, N after exclusions, any discrepancies vs. paper sample]
## Results Comparison
| Target | Table/Fig | Paper Value | Our Value (Python) | Our Value (R) | Diff | Status |
|--------|-----------|-------------|-------------------|---------------|------|--------|
## Discrepancies
[Each discrepancy: what, investigated how, resolved or not]
## Corrective Steps Taken
[Any adjustments made during validation and why]
## Verdict
**[REPLICATED / PARTIAL / FAILED]**
- Targets matched: N / Total
- Remaining discrepancies: [list or "none"]
## Reproducibility
- Python: X.Y.Z | pandas X.X | statsmodels X.X | lifelines X.X
- R: X.Y.Z | survival X.X | fixest X.X
- Data: [filename, UKB application ID if applicable]
- Seed: YYYYMMDD
```
Save to: `reports/[paper_name]_replication_report.md`
After saving: run domain-reviewer agent on the report.
---
## Quality Gate
After Phase 6, score the output. Minimum 80/100 to commit.
**Auto-commit if score >= 80:**
```
git add replications/[paper_name]/ reports/[paper_name]_replication_report.md quality_reports/[paper_name]_*.md
git commit -m "Replicate [Paper Author (Year)] -- [VERDICT]: N/Total targets matched"
```
---
## Failure Modes & Recovery
| Failure | Recovery |
|---------|---------|
| Script syntax error | Fix before proceeding |
| N mismatch > 5% | Stop, audit inclusion/exclusion criteria |
| All point estimates off by same factor | Check unit conversion (HR vs. log-HR, OR vs. log-OR) |
| SEs systematically too large | Check clustering level |
| Cannot install package | Document, note in report, use closest alternative |
| Data variable missing | Document gap; attempt proxy; flag as ASSUMED in report |
版本历史
-
v2026.07
GitHub ↗
原版更新说明
## Highlights - **First tagged release.** Every gate is green: `make check` (validate + 183 unit tests + eval-harness + numeric benchmarks), validate-catalog, quality-evals, and OpenSSF Scorecard on `main`. - **Rigor double loop at 13 method families.** Each family closes taxonomy tag → eval scenario (does the agent avoid the method's known traps?) → deterministic numeric benchmark (does it recover the true effect?). New this cycle: heterogeneous treatment effects (CATE) and quantile treatment effects (QTE). - **Paper-WorkFlow flagship at 29/29 executable rigor gates**, now merged to its `main` with a drift-gated README badge — the credibility artifact is checked by CI, so it can never silently go stale ([RIGOR.md](https://github.com/brycewang-stanford/Paper-WorkFlow/blob/main/RIGOR.md)). - **1,150 vendored skills / 69 collections**, all license-audited, security-scanned, tracked in a machine-readable catalog with CI-enforced freshness, and importable as a single whole-repo router skill. - **Hardened weekly upstream syncs**: synced trees must pass the full gate before merge; this cycle both sync PRs were verified, patched (StatsPAI's SkillOpt execution-gate card restored), and merged. - **First-party plugin marketplace** (Python / R / Stata empirical-analysis plugins + guide) shipped in the previous cycle, now under a tagged version. ## Known Follow-Ups - Ecosystem benchmark comparison (AERS vs econometrics agents on the shared benchmark suite) — [PLAN-2026-07 week 3](https://github.com/brycewang-stanford/Auto-Empirical-Research-Skills/blob/main/docs/PLAN-2026-07.md). - Decomposition (Oaxaca–Blinder) and shift-share (Bartik) method families (weeks 3–4). - GitHub Pages catalog site, README demo screenshots, five-language README consistency script (week 4). ## Catalog - Top-level collections: **69** - `SKILL.md` files: **1150** - Cataloged agent tools: **335** ([`tools/CATALOG.md`](../tools/CATALOG.md)) ## Methodological rigor - Method families in the coverage map: **13** ([`RIGOR_COVERAGE.md`](RIGOR_COVERAGE.md)) - Eval scenarios: **26** (138 rubric items, 122 auto-checkable) ([`../eval-harness/`](../eval-harness/README.md)) - Benchmark tasks: **13** (70 gold checks, 58 required) ([`../benchmark/`](../benchmark/README.md)) ### Method families (skills tagged) - `iv`: 27 - `rdd`: 25 - `panel-fe`: 19 - `staggered-did`: 17 - `bayesian`: 13 - `did`: 10 - `event-study`: 10 - `matching`: 10 - `synthetic-control`: 10 - `cate`: 8 - `dml`: 8 - `survival`: 3 - `quantile`: 2 ### Eval scenarios by category - `causal-identification`: 14 - `citation-hygiene`: 1 - `reproducibility`: 2 - `research-integrity`: 1 - `runtime-safety`: 1 - `statistical-validity`: 2 - `writing-compliance`: 3 - `writing-style`: 2 ## License buckets - AGPL-3.0: 1 - Apache-2.0: 1 - CC-BY-NC-4.0 (non-commercial): 1 - CC-BY-SA-4.0: 1 - CC-BY-SA-4.0 (repository default): 4 - GPL-3.0: 1 - MIT: 35 - MIT Non-Commercial: 1 - Mixed (per-skill; see folder): 2 - UNKNOWN - check upstream: 22 ## Quality gates - `make check` — catalog/provenance/audit/eval freshness, repo hygiene, Paper-WorkFlow gates, stdlib unit tests, eval-harness lint + graded smoke, and the strict numeric benchmark (reference-gold drift fails the build). - External links and OpenSSF Scorecard run on their scheduled workflows.
当前收录版本
0.0.1
质量评测
满分 100。前四项来自 DeepSkill 收录的质量报告。
评分依据
提供完整的6阶段生物医学论文复制流程,解决科研可重复性这一高价值需求
评分依据
触发词清晰明确:/replicate-paper [paper.pdf] [data.csv|dta] 或自然语言指令
评分依据
6个阶段划分清晰,每阶段有明确目标和输出文件
评分依据
工具来源明确,仅使用平台基础工具(Read/Write/Edit/Bash/Glob/Grep)
计算方式
取前四项原始分值的算术平均,四舍五入取整。可通过收录报告核对各项评分与说明,运行记录需单独查证。
权限与使用前检查
根据当前主文档自动整理,命中内容可能是示例或防护说明;不是实际权限声明,也未覆盖附件、外部依赖或运行行为。
未从主文档识别到上述常见操作,不能据此判断无需权限或没有风险。
安全评级与访问范围
risk · 静态证据核验,不代表运行安全保证。
使用前请由宿主核对并授权实际文件读写、外部服务和联网目的地;本站的核验不会授予任何运行权限。
用户讨论
还没有已核实的点评。