Agent Skills
› brycewang-stanford/Awesome-Journal-Skills
› ecopol-replication-package
ecopol-replication-package
GitHub用于构建符合经济政策期刊标准的可复制数据包。整合原始数据、清洗与分析代码及主脚本，确保端到端复现所有结果。针对受限数据提供替代方案与验证路径，包含详细README和映射表，提升研究透明度与可信度。
Trigger Scenarios
需要使实证或模拟结果可访问且可复现时
代码分散且缺乏主脚本时
使用受限数据需制定验证计划时
期刊要求提交支撑图表的数据和程序时
会议前准备让讨论者即时核查数据时
Install
npx skills add brycewang-stanford/Awesome-Journal-Skills --skill ecopol-replication-package -g -y
SKILL.md
Frontmatter
{
"name": "ecopol-replication-package",
"description": "Use when assembling the data, code, and verification materials for an Economic Policy (EP) manuscript so results are accessible and replicable to the journal's standard. Builds the verification-ready package; it does not invent evidence or citations."
}
Replication Package (ecopol-replication-package)
When to trigger
- Empirical, experimental, or simulation results need to be made accessible and replicable
- Code is scattered across machines with hard-coded paths and no master script
- Data are restricted/proprietary and you need a verification plan that still works
- The Managing Editor / production process will ask for the datasets and programs underlying the figures
- You want the package ready before the conference so a discussant can probe a number on the spot
EP's data standard: accessible and replicable, verified by the journal
EP's guidance is that "all empirical, experimental and simulation results should, where possible, be accessible and replicable," and authors submit the datasets, programs, and sources for the journal to verify and publish alongside the article (检索于 2026-06；以官网为准). EP does not (as of this writing) run the AEA-style mandatory pre-acceptance Data Editor audit that AEJ:EP / QE use, so the bar is "where possible" rather than universal — but a paper feeding a policy recommendation lives or dies on whether its central number can be reproduced. Build the package as if a skeptical discussant will re-run it. Confirm the exact deposit location, README format, and any embargo rules in the live guidelines (待核实).
What goes in the package
| Component | Requirement |
|---|---|
| Raw / source data | included if licensing allows; otherwise a precise acquisition guide |
| Cleaning code | from raw to analysis dataset, one master script, no manual steps |
| Analysis code | reproduces every number, table, and figure in the paper |
| Master script | runs end-to-end with one command; relative paths only |
| README | data sources, software + versions, run instructions, expected runtime |
| Data citations | each dataset cited with provider, version, access date |
| Mapping | exhibit → script → output (so a discussant can find any number fast) |
Handling restricted policy data
EP papers often use confidential administrative or central-bank data. When you cannot deposit the raw data:
- Provide the full code plus a synthetic or simulated dataset with the same structure so the pipeline runs end-to-end.
- Document the exact access procedure (the agency, the application route, the version) so a determined replicator can obtain it.
- Offer the journal a verification path (e.g., code run on-site, or output verified against a secure enclave) rather than nothing.
- State the restriction openly in the data section — silence reads as evasion to a discussant.
Craft moves
- One-command reproducibility. A reviewer should clone, run
master, and get your exhibits. Hard-coded/Users/yourname/paths are the most common failure. - Pin software versions. "Stata 18.0", "R 4.4.1 with fixest 0.12" — not "recent version".
- Map every headline number to the line of code that produces it; the conference is live and you may be asked to show provenance.
- Seed every simulation / bootstrap and report the seed.
- Keep the package legible, not just runnable — a clear folder structure signals the same care the policy audience expects of the analysis.
Checklist
- Master script runs end-to-end with one command, relative paths only
- Every table/figure/number in the paper is reproduced by the code
- README lists data sources, software + exact versions, run instructions, runtime
- Restricted data handled: synthetic data + access guide + verification path
- Each dataset formally cited (provider, version, access date)
- Exhibit → script → output mapping included
- Simulations/bootstraps seeded and the seed reported
- Deposit location / README format confirmed against live guidelines (待核实)
Anti-patterns
- Hard-coded absolute paths that break on any other machine
- "Data available on request" with no code and no access procedure
- A package that produces some but not all of the paper's numbers
- Unpinned software versions, so the pipeline silently breaks on a newer release
- Treating replication as a post-acceptance afterthought when a discussant may probe it at the conference
Output format
【Journal】Economic Policy (EP)
【Skill】ecopol-replication-package
【One-command run】master reproduces all exhibits? Y/N
【Coverage】every number/table/figure reproduced? Y/N
【Restricted data plan】synthetic data + access guide + verification path
【Versions pinned】software + package versions listed? Y/N
【Exhibit→code map】present? Y/N
【Deposit spec】confirmed / 待核实
【Next skill】ecopol-referee-strategy
Version History
- 1839142 Current 2026-07-05 12:53
