Guide · Excel

Claude skills for Excel: how to choose by stage

By SkillPicker Editorial · Published September 28, 2026 · 4 min read

Excel work is a stack: get the numbers in cleanly, calculate, summarize, chart, and only then automate the parts you repeat. The mistake most people make is reaching for an automation skill before the data underneath it is trustworthy. This guide orders the work, then explains how to pick a skill for each stage and how to tell the two kinds of Excel skills apart.

Every candidate here comes from the Excel topic, where the full demand-ranked list lives. The goal is to choose deliberately, not to install everything.

Quick scan: which stage needs a skill

StageWhat to look forWatch out for
Clean dataHandles messy headers, duplicates, and typesSilent type coercion that changes values
FormulasWrites and repairs lookups and logicHardcoded ranges that break when rows move
SummarizeBuilds pivot tables and aggregationsPivots that cannot refresh from the source
ChartCreates charts that survive a data refreshFormatting that hides the axis or misleads
Data modelRelates multiple tables for analysisA model only one person understands
AutomateRepeats a procedure without the clicksAutomation over unverified data

Before you pick: file-writer or live-workbook tool

This is the decision that causes the most disappointment. Excel skills fall into two families, and they fail in different ways.

FamilyHow it worksBest when
File-basedReads and writes .xlsx/.csv with a library such as openpyxl or pandasHeadless work, batch jobs, no Excel window needed
Live workbookDrives a running Excel session, often with xlwingsYou need formulas, charts, or a workbook to stay open and recalculate

openpyxl and pandas operate on files; xlwings controls Excel itself. A skill that says it does formulas or charts usually needs the live family, because a formula written as a static string is not the same as a formula Excel recalculates. Check which family a skill uses before you judge it.

Choose by stage

Clean data first

Cleaning skills handle the unglamorous work: inconsistent headers, duplicates, mixed types, stray whitespace. A good one shows you what it changed. Prefer a skill that reports the rows it touched over one that silently rewrites the sheet, and keep a copy of the original.

Formulas and repair

Formula skills write lookups, logic, and aggregations, and some audit existing formulas for errors. The valuable ones use structured references or named ranges rather than hardcoded ranges, so a later insert does not silently shift the calculation. An audit skill is worth pairing with any skill that writes formulas.

Summarize and chart

Summarizing is pivot-table work; charting is turning a range into something a reader can interpret without you in the room. Look for skills that keep charts tied to the source range so they refresh. If the output is a dashboard someone else will maintain, the Dashboard design topic is the better anchor.

Data models

When one sheet is not enough, the work moves to a data model: related tables, Power Query for shaping, and DAX for measures. This is the stage where a skill saves the most time, because the setup is fiddly and easy to get subtly wrong. Check that the skill understands your stack — Excel's Power Query and DAX, not a generic SQL model.

Automation last

Automation skills repeat a procedure: refresh, recalculate, export, email. They are only as good as the procedure underneath. Automate after you trust the data and the formulas, not before, or you will reproduce bad numbers at speed.

How to judge a candidate

  • Family: file-based or live workbook? Match it to whether you need Excel running.
  • Transparency: does it tell you what it changed, or only that it finished?
  • Range hygiene: structured references and named ranges over hardcoded ones.
  • Dependencies: openpyxl and pandas are light; a hosted API adds a network and a cost.
  • Source: read the SKILL.md before pointing it at a workbook you cannot lose.

Where to start

A representative candidate per stage, to compare rather than to rank:

Representative Excel candidates by stage
StageCandidateApproach
Create/editxlsx (Anthropic)File-based; creation, formulas, charting
Clean dataclean-data-xlsFormulas and dedup on messy sheets
Auditaudit-xlsChecks formula and model integrity
Analyzeexcel analysispandas/openpyxl; reads, pivots, charts
Data modelexcel-mcpPower Query, DAX, PivotTables
Automateexcel-automationxlwings; drives a live workbook

For adjacent work, Data analysis covers analysis beyond the spreadsheet and PowerPoint covers turning the result into slides. If your real output is a table for a web page, the HTML topic is the more direct route.

Before you install

Frequently asked questions

Can a Claude skill edit an .xlsx file directly?

Yes. Some skills write files with a library such as openpyxl or pandas; others drive a live Excel session with xlwings. The file-based ones work headless; the live ones need Excel running.

Should I use one Excel skill or several?

Use several if your work spans stages. Cleaning, formula work, and automation are different jobs; a single skill that claims all of them is usually thin in the places that matter.

Is it safe to let a skill edit my spreadsheet?

Treat it like any script that writes your files. Work on a copy, read the skill's instructions first, and check the result against a known total or a sample of rows.

Reviewed, not endorsed.These guides explain how to compare skills. They are not a recommendation or a security guarantee. Read each skill's own SKILL.md before installing it. How this index works