business-analysis-learning-agent-skill

This is a Python-based tool designed to help business students and data analysis beginners learn how to analyze real datasets, not just see the results. The project is written primarily in Chinese and intended for Chinese-speaking learners, but the code and pipeline architecture are language-agnostic. You provide a CSV or Excel file, and the tool walks through an analysis while explaining the thinking at each step: what each data field likely means, which business metrics can be calculated from the available columns, and what the findings do and do not prove.

The analysis pipeline runs in a fixed sequence: it reads the data, inspects field names to guess their business meaning, matches those fields against a library of common metrics (things like click-through rate, average order value, or profit margin), then runs ranking, trend, and distribution analysis. Each step is rule-driven and auditable. The tool explicitly is not a machine learning system and does not fit statistical models automatically. It stays in the territory of descriptive business analysis, which is appropriate for the audience.

After each run, the tool produces four types of output: an Excel workbook with charts and a dashboard, an HTML report designed for reading in a browser, a PDF version of that same report, and a JSON audit log that records what was executed, what was skipped, and why. These files are meant to be kept and reviewed later, not just read once during the session.

The project can be used through Claude Code (via a .claude/ adapter), Claude desktop or web, ChatGPT, or Gemini. It packages its behavior as an agent skill file, so any AI assistant that can read instruction files can follow the same workflow. The command-line tool also works standalone without an AI agent: one Python command with an input path and an output mode (beginner summary, standard report, or audit report).

Sample outputs for a retail sales dataset and a multi-table dataset are included in the repo so you can see what the Excel, PDF, and audit log look like before running anything yourself. The project README and all user-facing text default to Chinese.