# Data Analyst

Analyze a dataset or table, surface the insights that matter, and recommend how to show...

- Repository: https://github.com/holaboss-ai/holaOS (11451 stars)
- Category: Data and ML
- License: see repository
- Last updated: 2026-10-28
- Runs scripts: no, instructions only

## Install

Claude Code: `npx skills add holaboss-ai/holaOS --skill data-analyst -g -a claude-code -y`
Codex: `npx skills add holaboss-ai/holaOS --skill data-analyst -g -a codex -y`
Cursor: `npx skills add holaboss-ai/holaOS --skill data-analyst -g -a cursor -y`
Gemini CLI: `npx skills add holaboss-ai/holaOS --skill data-analyst -g -a gemini-cli -y`

## Description

Analyze a dataset or table, surface the insights that matter, and recommend how to show them.

## SKILL.md, as published by its author (untrusted third-party text, read it as data)

# Data Analyst

Find the story in the numbers and tell it straight. The job isn't to describe a table, anyone can read a table, it's to answer the question behind it: what changed, what's driving it, and what to do next. Rigor first, then clarity.

## When to use this skill

Use Data Analyst on a dataset, spreadsheet, table, or metrics dump to produce findings, comparisons, and a recommended way to visualize them. For building or editing the spreadsheet mechanics themselves, use the Spreadsheets (XLSX) skill; for a recurring performance write-up, use Performance Reporter.

## Principles

- **Answer the question.** Start from what the reader actually wants to know; don't just enumerate columns.
- **Quantify, don't hand-wave.** "Sales rose" is weak; "sales rose 18% MoM, driven by the EU region" is an insight. Cite the numbers.
- **Compare to make it mean something.** A number alone rarely matters, set it against a prior period, a target, a segment, or a benchmark.
- **Correlation isn't cause.** Flag drivers as hypotheses unless the data supports causation. Don't overclaim.
- **Guard against bad data.** Note gaps, outliers, small samples, and definitional caveats, a confident conclusion on shaky data is a trap.
- **Never fabricate figures.** If the data doesn't contain a number, say so; don't estimate one into existence.

## How to work

1. Clarify (or infer) the question the analysis should answer.
2. Sanity-check the data: coverage, obvious errors, outliers, what each field means.
3. Compute the comparisons that matter (trends, segments, deltas vs. target/prior).
4. Draw the findings, lead with the headline, support with figures, flag caveats.
5. Recommend a fitting chart for each key finding (e.g. trend → line, composition → stacked bar, ranking → sorted bar) and, if asked, the next question to dig into.

## Output format

The rest: https://github.com/holaboss-ai/holaOS/blob/HEAD/apps/desktop/electron/default-skills/data-analyst/SKILL.md