XLSX

The XLSX Claude skill builds and edits Excel files from a chat, with actual formulas, proper formatting, color-coded financial models, and zero #REF! errors. Hand it a CSV. Or describe the model.

The Problem

Spreadsheets do the world's actual work. Building them is somehow always a mess.

You ask an AI to build you a revenue model. You describe the inputs, the assumptions, the projection periods. Back comes a file. You open it in Excel and every cell is a hardcoded number. The AI calculated everything in Python and typed the answers into the cells. Change an input. Nothing updates. There isn't a single formula in the entire workbook.

Or you inherit a CSV from someone's CRM export. Headers on row 4. Junk rows at the top. Dates stored as strings. A column literally called amount_2 (final). You try to pivot. Excel offers you a single row of NaN and a polite suggestion to try again.

Or you build a 3-tab forecast at 11pm for an investor meeting. Inputs scattered everywhere, no color coding, years formatted as 2,025, negative numbers showing as -1234 instead of (1,234). Investor opens it tomorrow and immediately knows you don't build models for a living.

This skill solves that problem.

What this Claude skill does

• Writes real Excel formulas: SUM, AVERAGE, VLOOKUP, XLOOKUP, IF, NPV, IRR. Actual formulas in cells, not hardcoded results typed in by Python.
• Edits existing .xlsx, .xlsm, .csv, and .tsv files: add columns, change formulas, restructure sheets, fix broken references without touching the rest of the file.
• Cleans messy tabular data: misplaced headers, junk rows, mismatched columns, dates stuck as strings, all turned into a proper sheet you can pivot.
• Follows financial model conventions: blue text for inputs, black for formulas, green for cross-sheet links, red for external links, yellow fills for key assumptions.
• Applies the right number formats: $#,##0 for currency, parentheses for negatives, 0.0x for multiples, years as 2024 (not 2,024), zeros as -.
• Recalculates and error-checks every output: runs the file through LibreOffice, catches #REF!, #DIV/0!, #VALUE!, #NAME? before the file ever lands in your inbox.

Why it's amazing!

• Stop asking an AI for a spreadsheet and getting back a wall of hardcoded numbers.
• Ship a financial model an investor won't flag as "not built by a real analyst."
• Clean a messy CSV in seconds, not the rest of your afternoon.
• Never send a model with a #REF! lurking in cell BL47 again.

Who it's for

Anyone who lives in Excel. Founders building forecasts before a raise, analysts modeling deals, ops teams cleaning vendor exports, marketers reconciling campaign data, and consultants shipping client deliverables. Anyone who's stayed up past midnight fixing a model that worked yesterday.

How it works

Ask Claude something like:

• "Build me a 3-statement model from this 10-K, projected 5 years, with a sensitivity tab."
• "Clean this CSV: headers are on row 4, junk at the top, dates as strings."
• "Add a column to expenses.xlsx with month-over-month % change and conditional formatting."
• "Take this messy revenue-by-customer export and pivot it into a clean monthly cohort table."
• "Build a SaaS metrics dashboard: ARR, NDR, gross margin, CAC payback, all real formulas, blue inputs, the works."
• "Open Q3-budget.xlsx, find every #REF! error, and tell me what each one was supposed to point at."

You'll get back something like:

Built: revenue-model.xlsx
  - 3 tabs (Assumptions, Model, Sensitivity)
  - 187 formulas, 0 hardcoded calculations
  - Inputs in blue, formulas in black, cross-sheet links in green
  - Recalculated: 0 #REF! / #DIV/0! / #VALUE! errors

Cleaned: customers-export.xlsx
  - 4 junk rows removed from top
  - Headers promoted from row 4 to row 1
  - 1,247 dates parsed from string to date format
  - Dropped: 'amount_2 (final)'. Kept: 'amount'.

Edited: Q3-budget.xlsx
  - New column "MoM %" added (=B3/B2-1, filled down)
  - Conditional formatting: red below 0%, green above 5%
  - Existing sheet structure and styling untouched

Pivoted: monthly-cohorts.xlsx
  - 12 cohort rows × 12 period columns
  - Retention % as formulas, not hardcoded values
  - Headers frozen, currency format applied

Built: saas-metrics-dashboard.xlsx
  - 5 KPI tiles, 4 charts, 1 assumptions tab
  - ARR formula chain traces back to MRR inputs
  - Negatives in parentheses, years as text strings

Every file runs through LibreOffice for a full recalc before it's handed back, so when you open it in Excel, every formula resolves cleanly instead of showing #N/A across half the sheet.

FAQ

Does it cost anything to run?

The skill itself is free. It uses open-source Python libraries (pandas, openpyxl) and LibreOffice for formula recalculation, all installed once. No Excel license required to build the file, though you'll want Excel or Google Sheets to open it.

Does it actually write Excel formulas, or hardcode the numbers?

Real formulas. This is the whole point. When you ask for a sum, the cell gets =SUM(B2:B9), not the calculated total typed in. Change an input later and the spreadsheet recalculates the way Excel is supposed to.

What about complex models, like sensitivities, scenarios, dashboards?

Yes. Multi-tab models, cross-sheet links, named ranges, conditional formatting, charts, freeze panes, and data validation all work. The skill follows investment-banking formatting conventions (blue inputs, black formulas, green cross-sheet links) unless you tell it otherwise.

Can it work on macro-enabled .xlsm files?

It reads, edits, and saves the workbook content. Macros themselves (the VBA inside) are preserved on edit, but the skill won't write new VBA for you — use a real IDE for that.

Does it work on CSV and TSV files too?

Yes. CSV in, XLSX out. CSV in, cleaned-CSV out. Or take a busted XLSX and export a tidy CSV. The skill treats .xlsx, .xlsm, .csv, and .tsv as the same family.

How big a file can it handle?

Tens of thousands of rows, no problem. For very large files (hundreds of MB), openpyxl's read-only and write-only modes stream rather than loading everything into memory.

Does it preserve existing templates and formatting?

Yes, and it's strict about it. Hand it a model with its own conventions and it matches those (fonts, colors, number formats) rather than imposing its own. The existing template always wins over the default styling.

Does it check for formula errors before handing the file back?

Yes. Every file gets run through LibreOffice for a full recalc, then scanned for #REF!, #DIV/0!, #VALUE!, #NAME?, and #N/A. If anything's broken, the skill fixes it and recalculates again before saving.

Safe To Install

Please read before installing. Any skill you copy online is someone else's code running inside your Claude, and a lot of what's out there should give you pause. Most authors who give skills away don't give a second thought to quality or security. Most are vibe-coded slop pushed out untested, and some are the occasional bad actor hiding an exploit in plain sight. (Read about OpenClaws ClawHub security nightmare). Every skill we publish is vetted by a programmer with 20+ years of professional experience and checked line by line for anything that touches your files without reason. With Stim-Pack Studios Claude Skills, you get peace of mind knowing you're installing safe, vetted code, not a stranger's guess.

Add it to your Claude workflow

If you build, edit, or clean up spreadsheets at all (forecasts, models, vendor exports, campaign reports), this one earns its slot. It replaces the back-and-forth of asking an AI for a spreadsheet and then rebuilding half of it by hand.

Get the XLSX Claude skill and build Excel files that actually work like Excel files.