data-analysis-report

中文     English

中文

一个 Claude Skill：读取 Excel/CSV 数据文件，结合业务背景和分析目标，自动完成数据质量审计、统计检验（Z 检验 / 置信区间 / 效应量 / 多重比较校正），并生成一份包含图表、结论建议、方法说明的完整 HTML 分析报告。

这个 Skill 做什么

只要你提供表格数据（xlsx/csv）并希望做分析、看趋势、比较分组、评估效果、产出报告或可视化结论，这个 Skill 就会被触发——即使你没有明确说”报告”两个字。

核心流程：

  1. 需求澄清（信息缺失时才问，最多一轮）——确认分析目标、核心指标口径
  2. 数据质量审计——重复行、缺失率、类型陷阱、脏值归一化、极端值处理，全部披露而非静默跳过
  3. 分析规划与指标计算——纯 pandas 计算，覆盖分组比较、趋势、转化/漏斗、结构分析、驱动因素定位等场景，包含统计检验、置信区间、效应量、小样本标记
  4. 生成 HTML 报告——用 Chart.js 4.x 在浏览器端渲染图表，遵循统一的视觉规范（”柔和纸墨邻近色” Harmonious Paper & Inky Ink 设计系统）
  5. 生成后自检——确保每个结论都能回指到具体图表或检验结果，每个数字都来自 Python 计算
  6. 自动启动本地预览

质量底线

依赖

使用方法

将本仓库的 SKILL.md 放入 Claude 的 skills 目录（例如 ~/.claude/skills/data-analysis-report/SKILL.md，具体路径以你所用的 Claude 客户端/环境为准），Claude 在检测到相关数据分析需求时会自动加载并遵循其中的执行规范。

SKILL.md 的执行指令主体是中文（开发与测试语言），生成的报告本身语言不受限——见下方示例中的英文版报告。

示例

examples/telco-churn-report/ —— 基于公开的 IBM Telco Customer Churn 数据集生成的完整英文报告，展示了这个 skill 的核心能力：Z 检验 + Wilson 置信区间、Benjamini–Hochberg 多重比较校正、”效率 + 结构”双面板图表（避免辛普森悖论式误读）、混淆因素识别与谨慎因果表述。

👉 直接查看报告

目录结构

data-analysis-report/
├── SKILL.md              # skill 的完整定义与执行规范
├── README.md
└── examples/
    └── telco-churn-report/
        ├── index.html            # 示例报告（英文）
        ├── telco-customer-churn.csv
        └── README.md

English

A Claude Skill that reads Excel/CSV files and, combined with business context and an analysis goal, automatically performs data-quality auditing, statistical testing (Z-tests, confidence intervals, effect sizes, multiple-comparison correction), and produces a complete HTML analysis report with charts, findings, and a methodology note.

What this skill does

Triggers whenever you provide tabular data (xlsx/csv) and want analysis, trend-spotting, group comparisons, impact evaluation, a report, or visualized conclusions — even if you never say the word “report”.

Core pipeline:

  1. Clarification (only when information is missing, at most one round) — confirms the analysis goal and how core metrics are defined
  2. Data-quality audit — duplicate rows, missing rates, type traps, dirty categorical values, outliers; all issues are disclosed, never silently skipped
  3. Analysis planning & metric computation — pure pandas, covering group comparisons, trends, conversion/funnels, structural analysis, and driver identification, with statistical tests, confidence intervals, effect sizes, and small-sample flags
  4. HTML report generation — charts rendered client-side with Chart.js 4.x, following a consistent visual system (“Harmonious Paper & Inky Ink”)
  5. Post-generation self-check — every conclusion must trace back to a chart or test result; every number must come from Python computation
  6. Automatic local preview

Quality bar

Requirements

Usage

Place this repository’s SKILL.md into Claude’s skills directory (e.g. ~/.claude/skills/data-analysis-report/SKILL.md, the exact path depends on your Claude client/environment). Claude will automatically load and follow it when it detects a relevant data-analysis request.

The operating instructions inside SKILL.md are written in Chinese (the language it was developed and tested in). The language of the generated report is not constrained by that — see the English example below.

Example

examples/telco-churn-report/ — a full English-language report generated on the public IBM Telco Customer Churn dataset, showcasing the skill’s core capabilities: two-proportion Z-tests with Wilson confidence intervals, Benjamini–Hochberg correction for multiple comparisons, a dual-panel “efficiency + structure” chart (to avoid a Simpson’s-paradox-style misread), and explicit confounder callouts with cautious causal language.

👉 View the report

Directory structure

data-analysis-report/
├── SKILL.md              # full skill definition and execution rules
├── README.md
└── examples/
    └── telco-churn-report/
        ├── index.html            # example report (English)
        ├── telco-customer-churn.csv
        └── README.md