How to Install the xlsx Skill in Claude Code (2026)

You have a 4,000-row sales export and your boss wants a clean summary tab with a regional pivot, a chart, and conditional formatting on anything below target, all by lunch. You could spend an hour fighting openpyxl, or you could install one skill and describe the result. The xlsx skill turns Claude Code into something that reads and writes real Excel files the way a competent analyst would, formulas and charts included. Here's how to set it up in about five minutes, plus the exact commands that earn their keep on a real workday.

What the xlsx skill actually does

Out of the box, a coding agent can write Python that touches spreadsheets. What it can't do reliably is handle real Excel: preserve formulas, write to a specific sheet, build a chart, keep the formatting your finance team expects. The xlsx skill packages that knowledge. Once it's installed, "add a pivot of revenue by month on a new tab and bold the totals" becomes one instruction instead of a debugging session.

It covers reading (pulling data and structure out of a workbook), writing (creating new files from scratch), and editing (changing an existing file without nuking the rest of it). That last one matters most. Most tools that "support Excel" actually rebuild the file and lose your formatting.

Step 1: Confirm your setup

You need Claude Code installed and working. Open a terminal and check:

claude --version

If that returns a version, you're set. The xlsx skill also relies on Python being available, since it runs spreadsheet libraries under the hood. Most setups already have it; confirm with:

python3 --version

Step 2: Install the skill

Skills live in your Claude Code skills directory, usually ~/.claude/skills/. Installing is a matter of putting the skill folder there. If you're pulling from a public repo:

cd ~/.claude/skills
git clone <the-xlsx-skill-repo-url> xlsx

If you use a skills marketplace, the install is typically a single command from inside Claude Code instead. Either way, the result is a folder containing a SKILL.md file and supporting scripts.

Before you go further, read the SKILL.md. This isn't boilerplate advice. A skill is code that runs on your machine, and the SKILL.md tells you exactly what it does. Open it:

cat ~/.claude/skills/xlsx/SKILL.md

Two minutes of reading tells you what the skill touches and whether it bundles any scripts you'd rather review. Never install a skill you haven't opened.

Step 3: Restart and verify

Restart Claude Code so it picks up the new skill. Skills load based on their description matching your request, so you don't "turn it on." You just ask for something spreadsheet-shaped and the agent reaches for it. Test with something small:

> "Create a new Excel file called test.xlsx with a sheet named Sales, columns for Date, Region, and Amount, and five rows of sample data."

If you get a real test.xlsx with those columns, the skill is live.

Step 4: The commands that actually save time

Now the useful part. A few prompts that replace real work:

Reshape an export. "Open quarterly.xlsx, drop the two total rows at the bottom, split the Name column into First and Last, and save it as quarterly_clean.xlsx." The agent reads the structure, makes the edits, preserves the rest.

Build a summary tab. "Add a new tab to sales.xlsx called Summary with a pivot of total Amount by Region and by Month, and bold the grand total." This is the one that saves the most time. Manual pivots plus formatting is 20 minutes you get back.

Add a chart. "On the Summary tab, add a bar chart of revenue by region below the pivot." It writes the chart into the actual workbook, not a separate image.

Apply business rules. "Add conditional formatting to the Amount column: red fill for anything under 5,000." The kind of finishing touch that makes a sheet usable and that nobody enjoys clicking through by hand.

The pattern across all of these: describe the outcome, name the file and sheet, and let the agent write and run the code. The more specific you are about where things go, the less back-and-forth.

Step 5: Build a repeatable script, not a one-off

The single best habit with this skill: when a task will recur, ask for a script instead of a one-time edit. "Write a Python script that takes any monthly export in this format and produces the clean summary tab." Now next month is a 10-second run, not a re-prompt. You're turning a chore into a tiny pipeline. This is where the xlsx skill quietly becomes spreadsheet automation rather than a faster way to do manual work.

A real workflow: month-end in one prompt chain

Here's how this looks on a real task instead of a toy one. Finance hands you transactions_june.xlsx: 6,000 rows, one flat tab, no summary. They want the usual month-end view by Friday.

You walk it in four instructions. First: "Open transactions_june.xlsx and tell me the columns and any data issues." The agent flags that Category has 14 inconsistent spellings and three rows have blank amounts. Second: "Standardize the Category spellings to this list, and flag the blank-amount rows on a new tab called Review rather than deleting them," because a blank amount is a question for finance, not something you silently drop. Third: "Add a Summary tab with total spend by Category and by week, with the grand total bolded." Fourth: "Add a bar chart of spend by category below the pivot, and red conditional formatting on any category over budget."

What was a two-hour afternoon of pivot tables and manual formatting is now a five-minute conversation, and the Review tab means you caught the data problem instead of burying it. Save those four prompts as a script and next month is one command.

The security notes you need

The skill runs locally and reads or writes files on your machine, so the file operations themselves are low risk, and you can point it at a copy if you're nervous. The real consideration is what reaches the model. If you ask the agent to "summarize what's in budget.xlsx," the contents go into the conversation with the model. For sensitive financials, prefer the script route: ask it to write code that processes the file locally and run that, so the raw numbers never leave your machine. Same principle as the broader skill-security rule: the skill is an open folder, so read it, and scope what you hand it.

When the xlsx skill isn't the answer

It's the wrong tool for a few jobs. For PDFs, use the pdf skill instead. For Word documents, the docx skill. And for a genuine database with millions of rows, don't route through Excel at all. That's a SQL job. The xlsx skill is for spreadsheet-shaped work, which is a huge slice of real analytics, but not all of it. If you find yourself wishing the skill did something it doesn't, the skill-creator skill lets you build your own.

The bottom line

Installing the xlsx skill takes five minutes: drop the folder in, read the SKILL.md, restart, test. What you get back is every "can you just clean up this spreadsheet" task collapsing from an hour to a sentence. Build scripts for the recurring ones, keep sensitive data local, and the worst part of spreadsheet work mostly disappears. Start with one real file this week. The next month-end will feel different, and you'll wonder why you spent years clicking through pivot dialogs by hand.

FAQ

What does the xlsx skill do in Claude Code?

It teaches the agent to read, create, and edit real Excel files, including formulas, multiple sheets, formatting, and charts, without you writing openpyxl or pandas code by hand. You describe what you want in plain language and the agent builds or edits the workbook.

How do I install the xlsx skill?

Clone the skill folder into your Claude Code skills directory (commonly ~/.claude/skills/), or install it from a skills marketplace if you use one. Restart Claude Code, and it loads the skill automatically when your request involves a spreadsheet. Always read the SKILL.md before installing.

Is the xlsx skill safe to use with my financial data?

The skill itself runs locally and reads/writes files on your machine, which is low risk. The thing to watch is what the agent sends to the model — if you ask it to summarize the file contents, that text goes to the model. For sensitive numbers, ask it to write a script that processes the file locally rather than pasting the data into the conversation.

Do I need to know Python to use the xlsx skill?

No. The point of the skill is that you describe the outcome ("add a pivot of sales by region on a new tab") and the agent writes and runs the code. Knowing a little Python helps you review what it generates, but it is not required to get useful results.