The xlsx skill treats a spreadsheet as a file to be edited, not a table to be regenerated. It is one of the four Anthropic document skills, and its SKILL.md is unusually blunt about which libraries do what, because the failure modes are silent: a wrong approach still produces a file that opens.
What it actually finishes
- Build a sheet from nothing — cells, formulas, formatting, charts, through
openpyxl. - Edit a sheet you did not make — add columns, rewrite formulas, fix formatting, clean rows that arrived malformed.
- Move bulk data in or out —
pandas.read_excelandto_excelfor volume, not for styling. - Glance at a file —
markitdown file.xlsxprints one## SheetNameblock per sheet and reads.xlsmtoo. - Read a model with both formulas and values — two passes:
data_only=Falsefor the formulas,data_only=Truefor the cached results.
When it fires, and when to skip it
The trigger in the skill's own description is broad on purpose: it fires any time a spreadsheet is the primary input or output, including the casual case where someone just calls it "the xlsx in my downloads." Cleaning messy tabular data into a proper sheet counts.
It explicitly does not fire when the deliverable is something else — a Word document, an HTML report, a standalone Python script, a database pipeline, or a Google Sheets API integration — even if tabular data is involved along the way. The test is the file you end up handing over, not the data you touched.
How to check it worked
- Run the skill's own recalculation script on the output:
python scripts/recalc.py output.xlsx. It drives LibreOffice to recalculate and writes the results back into the cached values. - Open the result with
data_only=Trueand confirm the formula cells hold numbers rather thanNone. markitdown output.xlsxfor a fast human look at what landed in each sheet — but not for planning edits, because it gives no cell coordinates.
If step 2 still shows empty cells, the recalculation did not run or did not cover that sheet.
What goes wrong
Formulas written by openpyxl carry no cached value. Excel calculates them when a person opens the file, which is exactly why this looks fine locally and fails downstream: pandas and most previewers read the cache, find nothing, and return empty cells. This is the single most common way an xlsx task appears to succeed and is wrong. The fix is the recalculation script, run after writing.
Saving after data_only=True deletes every formula. That mode keeps only the values, so writing the workbook back out turns the model into a static table. Read with data_only=True or write, never both in one pass.
Six functions the runtime cannot compute. XLOOKUP, XMATCH, SORT, FILTER, UNIQUE and SEQUENCE are not supported by the LibreOffice build behind the recalculation step, so a sheet that only works in recent Excel recalcs to #NAME?. Use INDEX and MATCH instead. This is the one place where writing worse Excel is the correct choice.
Merged cells only accept writes at the top-left anchor. A value written to any other cell in the merged range is dropped or errors.
Macros disappear from .xlsm unless it is opened with keep_vba=True. A macro-enabled workbook silently becomes a plain one.
Financial models follow a colour convention worth keeping: blue text for hard-coded inputs, black for formulas, green for references to other sheets. If a cell is a number someone typed, it is blue — that convention is how an analyst reads your file from across the room.
How it compares
- docx — words and layout. Choose it when the deliverable is read as prose.
- xlsx — numbers, formulas and cross-sheet logic. The one that survives someone else editing it later.
- pptx — slides. Same numbers, different job: somebody presents them.
- pdf — fixed layout for the finished thing you send out. Get the numbers into xlsx first, then print to pdf.
None of them converts into another's format, and the skill says so: pick by the artefact you must hand over.
Try a sample task
This walkthrough gives you a prompt and an acceptance check to run in your own client.
Create a workbook and verify formulas
Use a small CSV with item names, quantities and unit prices whose totals you can calculate by hand.
Use the XLSX skill to put the rows in a formatted workbook, calculate line totals with formulas, recalculate, then change one quantity and verify the grand total.
What to verify Formula cells remain editable, values recalculate correctly, and the workbook opens with no formula errors.
Best for Spreadsheet creation, formula updates, financial models and workbook cleanup.
How to use xlsx
Keep raw data on its own sheet and use formulas for figures that should update when inputs change.
Acceptance check Recalculate the workbook, change one input, and confirm the summary and chart update without formula errors.
Install for your agent
Run the command from your project directory. After installation, reload the client if it has already started. Tool access and execution features depend on the client.
Claude Code
npx skills add anthropics/skills --skill xlsx -a claude-codeCodex
npx skills add anthropics/skills --skill xlsx -a codexCursor
npx skills add anthropics/skills --skill xlsx -a cursorGitHub Copilot
npx skills add anthropics/skills --skill xlsx -a github-copilotAsk your agent to install the skill
Install xlsx from anthropics/skills in this project. Show me the destination folder and report any dependencies the source lists.
For manual installation, copy the entire skill folder, including references and scripts, into a skill directory supported by your client. See the installation guide.
What the upstream guide covers
The upstream spreadsheet guide pairs workbook construction and formatting with verification guidance. Its recalculation instructions matter when cells contain formulas.
HavenSkills workflow check
November 17, 2026 · Programmatic check with openpyxl on a small Sales workbook.
- What we checked
- Saved and reopened a workbook with inputs 125 and 75. The cells survived and the saved formula remained =SUM(A2:A3).
- Scope
- This check covers workbook and formula persistence. A spreadsheet application is still needed to verify recalculated values.
Requirements, permissions and limits
- Runtime
- Python and openpyxl; LibreOffice or another compatible spreadsheet engine for recalculation
- File and tool access
- Use a copy of the workbook. Checking calculated formula results can require a spreadsheet engine such as LibreOffice.
- Limits to check
- Writing formulas alone may leave cached values stale. Spreadsheet applications can differ on which functions they support.
Frequently asked questions
Can it read a CSV?
Yes — pandas handles .csv and .tsv directly. If the deliverable has to be a real workbook, write it back out to .xlsx.
Will it keep my existing formulas when it edits the file?
Yes, as long as the workbook is opened and saved with data_only=False. The one way to lose them is saving after a values-only read.
Why are the formula cells empty when I open the result in pandas?
There is no cached value. Run the recalculation script, then read again.
Does it support the old .xls format?
The formats named in the skill are .xlsx, .xlsm, .xltx, .csv and .tsv. Convert an old .xls to .xlsx first.
Is it safe for a file that matters?
The repository says these skills are for demonstration and teaching, and to test them thoroughly before relying on them for critical work. Read the source before you point it at a workbook you cannot rebuild.