Using the prebuilt skills: documents, slides, spreadsheets, pdfs

Enabling the skills

The document skills depend on code execution, so that has to be available first.

In the Claude apps (web, desktop, mobile):

1. Open Settings, then look for the Capabilities or Features area where code execution / file creation lives.

2. Enable file creation and analysis. On consumic and Team plans this is the toggle that lets Claude build and return files.

3. Skills surface automatically when relevant, but you can confirm and manage which are active in the Skills section of settings.

You do not "install" the four Anthropic document skills the way you would a custom one. They are first-party and available once file creation is on. Custom skills are a separate workflow, which you will see in the next lesson.

For developers via the API, code execution and Skills are exposed as tools you attach to a request. Anthropic documents the current setup in the Skills docs at docs.claude.com. The APIAPIApplication Programming Interface: a standardised interface that lets applications communicate and exchange data without knowing each other's internal workings.View full definition → path is how you would wire deck generation into your own product rather than driving it by hand in the chat.

The three-slide deck, start to finish

Here is the concrete run. Open a chat with file creation enabled and prompt:

> Create a 3-slide PowerPoint pitching a weekly internal AI office-hours program. Slide 1: title slide with the program name "AI Office Hours" and a one-line tagline. Slide 2: three bullets on why it helps the team. Slide 3: how to join, with a time and a sign-up link placeholder. Keep the design clean and corporate.

What happens behind the scenes:

1. Claude recognizes this as a slide task and loads the pptx skill.

2. It writes a Python script using python-pptx.

3. The script runs in the sandbox and writes ai-office-hours.pptx.

4. Claude returns a download link (and usually a short summary of what it built).

You click the file, it downloads, you open it in PowerPoint, Google Slides, or Keynote. Done.

What Claude is running under the hood

You do not write this code, but seeing it demystifies the skill. This is the shape of what the pptx skill executes:

Notice the file is a real Presentation object. Every slide is a structured layout, every bullet is a paragraph with a font size. That is why the result opens cleanly and stays editable.

Retrieving the output file

In the apps, the file appears as a downloadable attachment in Claude's reply. Click it. On desktop it lands in your normal Downloads folder; on mobile you get a share or save sheet.

Via the APIAPIApplication Programming Interface: a standardised interface that lets applications communicate and exchange data without knowing each other's internal workings.View full definition →, the model returns a reference to a file produced in the code execution environment, and you fetch it with a follow-up files request. The pattern is: run the request, read the file ID from the response, download the bytes. The exact field names live in the current APIAPIApplication Programming Interface: a standardised interface that lets applications communicate and exchange data without knowing each other's internal workings.View full definition → reference, so check the Messages API docs rather than trusting a snippet that may drift.

Iterating without starting over

The real power is the follow-up turn. Because the deck exists as a file Claude built, you can refine it conversationally:

> Add a fourth slide listing three example questions someone might bring.

> Make slide 2's bullets shorter, max six words each.

> Re-export with our brand color #1A73E8 on the title slides.

Claude reruns the script with changes and hands back a new file. You are editing through language, then leaving with a normal .pptx you can polish by hand. This loop is the entire reason the skill beats copy-pasting bullets: the artifact is durable and structured.

The other three skills, briefly

The same mechanics apply across the family. What changes is the underlying library and what "good output" means.

docx (Word documents)

Best for structured prose: reports, memos, contracts, formatted letters. It preserves heading hierarchy, lists, and styles. Prompt it to "produce a two-page project brief with an executive summary, three sections, and a risks table" and you get a real Word file with a real table object, not Markdown pretending to be one.

xlsx (spreadsheets)

This is the one people underestimate. It writes live formulas, not just values. Ask for "a 12-month cash-flow model with revenue, costs, and a running balance," and the balance column contains =SUM(...) formulas that recalculate when you change inputs. Combine it with uploaded data: feed Claude a messy CSV and ask for a cleaned, formatted, formula-driven workbook back.

pdf

The pdf skill leans two directions. It can generate polished PDFs (think a one-page summary with layout), and it pairs with Claude's strong document reading to fill forms or extract structured data from PDFs you upload. A common workflow: upload an invoice PDF, ask Claude to pull the line items into an xlsx. That is two skills cooperating in one task.

When skills fire, and when to be explicit

Skills are loaded by relevance, which is usually invisible and correct. But two situations trip people up.

Ambiguous output format. "Summarize this quarterly data" might give you a chat reply, an Artifact, or a file. If you want a downloadable spreadsheet, say so: "return it as an xlsx file." Naming the format nudges the right skill.

Skill availability. If file creation is off, Claude will write you a nicely formatted answer in the chat and no file appears. If you expected a download and got text, check that capability first. This is the single most common "it didn't work" cause.

A clean habit: state the artifact and the format up front. "Build a deck (pptx) ..." or "Give me a workbook (xlsx) with ..." removes the guesswork.

Where this fits in a real workflow

These skills shine when the file is the deliverable. A few concrete patterns worth stealing:

• Recurring reports. Drop raw numbers into a Project, ask for the same xlsx layout each week. The Project's context keeps the format consistent.
• First-draft decks. Generate the pptx skeleton in seconds, then spend your time on the two slides that actually matter instead of building structure from scratch.
• Format conversion with judgment. "Turn this messy meeting transcript into a clean docx action-items memo." The skill handles structure; Claude handles the editorial call about what matters.
• Pipeline via API. When you need this on a schedule or inside an app, the same skills run through the Messages APIAPIApplication Programming Interface: a standardised interface that lets applications communicate and exchange data without knowing each other's internal workings.View full definition → with code execution, so a button in your tool can produce a branded report on demand.

The mental model: the chat is the editor, the skill is the printer, and the file that drops out is yours to keep and edit forever.

Key Takeaways

• Turn on file creation first. The four document skills (xlsx, docx, pptx, pdf) only work when code execution / file creation is enabled in your Claude settings. No toggle, no download.
• Name the format in your prompt. Say "return as a pptx" or "give me an xlsx workbook" to reliably trigger the right skill instead of getting chat text.
• The output is real and editable. Spreadsheets carry live formulas, docs keep true styles, decks have real slide objects. Treat the result as a starting file, not a screenshot.
• Iterate by conversation. Refine slides, shorten bullets, or recolor through follow-up messages; Claude reruns the script and hands back a fresh file.
• For production, go through the API. The same skills run via the Messages API with code execution, so you can generate files on a schedule or from inside your own product.