cabbage2000-lab/data-analysis-skills
Developer toolsThis page was translated automatically. Technical details may differ from the original. View the original
Turns data files (CSV/Excel/JSON/TSV) into storytelling analysis reports (supporting Claude Code, Codex, WorkBuddy): first confirm the industry and report audience, then clean and analyze to produce single-file HTML interactive reports—a six-paragraph format that leads conclusions, each chart is interpreted, and data processing is traceable; No fabrication of industry benchmarks, improper causality, or small sample sizes to draw conclusions
Overview
This is a data analysis tool — the same package can fit multiple programming agents like Claude Code, Codex CLI, WorkBuddy, etc., without binding to a single tool. Throw data files (CSV / Excel / JSON / TSV) to it and say "help me analyze," and it will work like a data analyst who understands your industry: first confirm your business background, then complete cleaning and analysis, and finally deliver a ** — not isolated charts and numbers, but data stories with insights, conclusions, and direct reporting. : Open the examples/sample_report.html in a browser — a real, generated SaaS business analysis report (36 months of subscription data, test data not from real customers). Must be connected to the internet, and the chart library is loaded via CDN. Cross-host **: Five-stage process, Stage 1 confirmation access control, statistical discipline, report six-part structure—these don't change with the host. : Use a structured question tool if the host has one (e.g., AskUserQuestion); if not, downgrade to numbered text multiple-choice questions and wait for your response; If you have a task list tool (TodoWrite / update_plan), use it; if not, print a plain text stage checklist. - "This is our Q2 order data, give me an analysis report" → A complete narrative report organized from a retail perspective: GMV, average transaction value, repurchases, returns, each chart with interpretation, conclusions first - "Help me see if Plan B is really better in the A/B experiment" → Directly provide rigorous statistical conclusions (p-value, confidence interval, premise check) without forcing you to go through the full reporting process - "Analyze my account's note data from the past three months" → Interpret engagement rate, streams, and follower growth efficiency from a self-media perspective - "This budget execution sheet helps me do a business analysis" → Analyze revenue, gross profit, expense ratio, and budget achievement from a financial FP&A perspective Update using /plugin update data-analysis-skills. WorkBuddy follows the prompt path of Method A (this repository has not yet provided a list of tested WorkBuddy plugins). Method A and Method B are equipped with the same set of skills,—— resulting in two copies and repeated triggers. The plugin form only changes how it appears in the host, not the analysis behavior. ——Codex and WorkBuddy don't read ~/.claude/skills/, Claude Code doesn't read ~/.codex/skills/. If you use all of these, install one copy for each. If you install them in the wrong location, it means 'skill doesn't appear at all, and there are no errors.' Replace the user-level path with the project-level directory under the project's root directory: Claude Code uses .
Overview (machine translated)
Chinese Simplified | English
This is an Agent Skill for data analysis—the same package can fit multiple programming agents like Claude Code, Codex CLI, WorkBuddy, and more, without binding to a single tool. Throw the data file (CSV / Excel / JSON / TSV) to it and say “Help me analyze,” and it will work like a data analyst who understands your industry: first confirm your business background, then complete cleaning and analysis, and finally deliver a single-file HTML interactive report—not isolated charts and numbers, but data stories with insights, conclusions, and direct reporting.
First, check the results: Open examples/sample_report.html in your browser — a real-life generated SaaS business analysis report (36 months of subscription data, test data not from real customers). Requires internet connection, and the chart library is loaded via CDN.
Which agents are supported
| Host | User-level installation location | To what extent is the validation | |------|----------------|----------------|
| Claude Code | ~/.claude/skills/data-analysis-skills | Main host, T1–T14, full eval passing |
| Codex CLI | ~/.codex/skills/data-analysis-skills | Adapted: task list goes update_plan, access control questions are downgraded to text numbered multiple-choice questions; Interactive full-process and codex exec smoke acceptance pass |
| WorkBuddy | ~/.workbuddy/skills/data-analysis-skills | Adapted: The installation path has been tested and in place, the author has tested loading and running in WorkBuddy; The S1–S5 cross-host acceptance checklist item by item has not yet been backfilled |
| Other isomorphism tools | The tool’s own skills directory | Just read SKILL.md + frontmatter + subdirectory resources to load; not tested individually |
Cross-host invariants: Five-stage process, Stage 1 confirmation access control, statistical discipline, report six-paragraph structure—these do not change by host. Only the interaction appearance changes: Use a structured query tool if the host has one (e.g., AskUserQuestion); if not, downgrade to numbered text multiple-choice questions and wait for your response; Use task list tools (TodoWrite / update_plan); if not, print plain text stage lists. **Access control itself never degrades; only its presentation does. **
What it can do for you
- “This is our Q2 order data. Give me an analysis report.” → A complete narrative report organized from a retail perspective: GMV, average order value, repurchases, returns—each chart with interpretation, conclusions first
This page was translated automatically. Technical details may differ from the original.
Recommended Tools
Try a different keyword or remove a filter.
Install
npx skillfish add cabbage2000-lab/data-analysis-skills