agentsclimarketplace

06 repro logger

Skill lianxhcn/PXa2026a/skills/core/06-repro-logger

Lecture

Install
npx -y skills add lianxhcn/PXa2026a --skill 06-repro-logger

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

3 things to look at

  • 20 days oldThe repository was created 20 days ago. New is not bad, but a brand new repository carrying a familiar-sounding name is the shape a typosquat arrives in, and there has been no time for anyone else to find a problem with it.
  • no licenseNo license file was found in the repository. Code published without one is not open source by default, so using it at work is a question for whoever answers licensing questions where you are.
  • 3 stars3 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its author says it does

Copied from the file, not written here

Use this skill to turn a data-analysis run into reproducible records. Given the scripts you ran, the data you used and the outputs you got, it drafts a reproduction log (input data → processing script → output), a variable-construction note, a sample-selection log (row counts at each filter), a results explanation and an open-questions checklist; it also abstracts one analysis run into a reusable project template. Records only what the user provides; marks steps that could not run for missing data or license; never fabricates numbers, file names or results. The human checks the draft for errors and omissions.

SKILL.md

4.4 KB, as published. Nobody here has run it

repro-logger(复现日志员)

把一次分析或复现,落成别人照着能重跑的记录。它不替你做研究判断,而是根据你实际跑过的脚本、用过的数据和得到的产出,起草复现日志、变量构造说明、样本筛选日志、结果说明和待检查问题清单,并能把一次流程抽象成可复用的项目模板——初稿由它写,错漏由你查。对应第 2 讲「留痕与可复现」和第 6 讲「整理复现日志、把流程整理成模板」的 AI 协作。

何时用

  • 跑完一段清洗或复现,想把「从哪份数据、经过哪些步骤、得到哪个结果」记成一份日志;
  • 样本一路筛下来行数在变,想留一张「每一步筛了什么、还剩多少」的样本筛选日志;
  • 要给关键变量写一份构造说明(怎么来的、口径是什么),补进变量字典;
  • 想把这次分析流程抽象成模板,下次拿到新任务照着套;
  • 复现里有些步骤因为缺数据或缺许可跑不了,要如实标注、不留假账。

使用方式(提示词)

生成「从数据到结果」复现日志

以下是我跑过的脚本、用到的数据和得到的产出:[粘贴]。请整理成一份复现日志,按步骤记录「输入数据 → 处理脚本 → 产出(中间数据 / 图 / 表)」,并标出哪些步骤因为缺数据或缺许可没跑成。只依据我提供的信息,不要补全我没给的内容。

生成样本筛选日志

这是我的样本筛选步骤和每步后的样本量:[粘贴,如「剔除缺失 126→124」]。请整理成一张样本筛选日志:每一行写「筛选条件、剔除多少、剩余多少、为什么这样筛」,末尾提示样本量变化里有没有需要我再核对的异常。

生成变量构造说明 / 结果说明

这些是我构造的变量和它们的定义:[粘贴]。请写成一份变量构造说明,每个变量写「原始来源、构造公式或规则、单位、口径注意点」;口径拿不准的地方标出来让我确认,不要编造定义。

整理成可复用的项目模板

请把我们这次分析的完整流程,抽象成一份可复用的项目模板:包括标准的文件夹结构、每一步做什么、每步建议调用哪个 skill、每步要人工核对的要点。目标是我下次拿到新任务,照着这份模板就能走一遍。

输出约定

  • 日志按步骤组织,每步「输入 → 处理 → 产出」三段齐全,缺一不补;
  • 跑不了的步骤如实标注原因(缺数据 / 缺许可 / 未运行),绝不假装跑过;
  • 样本筛选日志给出每步剔除数与剩余数,能对上最终分析样本;
  • 只依据用户提供的脚本、数据与产出,不编造行数、文件名、变量口径或结果;拿不准的显式标出。

最小使用示例

用 repro-logger 帮我把这段整理成复现日志:我先 use firm_finance_2021.dta(84 行),append 2022 年那份得到 126 行,再剔除 rd 缺失的 2 行剩 124 行,最后 reg rd_ratio size 出了一张回归表。

人工检查清单

  • 日志里每一步的输入输出,回到你实际跑的脚本逐条对一遍,数字(行数、匹配率)要对得上;
  • 跑不了的步骤有没有被如实标注——一份诚实的日志,比一份漂亮但注水的更有价值
  • 变量口径、清洗理由与筛选依据要与真实数据一致,别让 AI 替你「圆」一个说法;
  • AI 辅助、不替代判断:记录是否可信、能否复现,最终由你把关。

来源与许可

  • 原创:本课程为「连享会 2026 暑期班·初级班」编写,Agent 无关的开放 Skill,MIT(随仓库 LICENSE-CODE)。
  • 最后验证日期:<!-- TODO: T7 彩排 Claude Code 与 Codex 双端实测后填写 -->

Keep looking

Skills are one crate of 328,083. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.