← All posts

The Best Document Skills for Claude Code (Tested: Excel, Word, PowerPoint, PDF)

2026-09-22

Documents are where agent skills got their start — the official Excel, Word, and PowerPoint skills remain the benchmark everything else is measured against. I evaluated the full document category on six dimensions. Here are the seven worth your context directory.


Document work is the highest-demand category in the skill ecosystem, and it's where the gap between a great skill and a lazy prompt-wrapper is most expensive. A bad spreadsheet skill corrupts your formulas silently. A bad PowerPoint skill hands you a file that won't open on the conference machine.

After running every document skill through the six-dimension rubric — trigger quality, structure, workflow design, content, engineering, and security — here's what earned a spot, and the one thing to know about each.

1. xlsx (official) — 9.8/10

The highest-scoring skill in the entire directory, and it's not close.

Why: It verifies its own output. After every write, it recalculates the workbook and checks the result before claiming success. Most skills say "done" and hope. This one proves it.

The one thing: Its verification catches formula errors, not intent errors. If you ask it to "fix the numbers," open the file afterward and confirm the right thing got computed.

2. docx (official) — 9.7/10

Word documents: tracked changes, comments, find-replace, styled output, tables of contents.

Why: A task-routing table sends each job type to a proven recipe, and a render-to-image check catches layout breakage that text-only verification misses. When it builds a TOC, it looks at the rendered pages — not just the XML.

The one thing: Accept tracked changes in Word's review pane, not by reading the diff in chat. The visual check is the whole point of the skill's design.

3. pptx (official) — 9.7/10

PowerPoint creation and editing, template-aware.

Why: It knows the corruption patterns that naive XML editing causes — and routes around every one of them. Like xlsx, it verifies after building.

The one thing: Feed it your simplest working template, not the 40-master-slide monster your design team made in 2019. Complex templates are where PowerPoint itself breaks.

4. baoyu-slide-deck — 9.3/10

Generates slide decks as styled images — 17 style presets with a consistent visual identity across slides.

Why: A genuinely different philosophy from pptx: instead of wrestling PowerPoint XML, it renders each slide as a designed image. For "make this look good fast," it wins outright.

The one thing: The output isn't editable PowerPoint. If your boss needs to tweak slide 12 after the meeting, use pptx instead. Pick your use case first.

5. prompt-master — 9.3/10

Turns a vague request into a production-quality prompt via 9-dimension intent extraction.

Why: The best trigger design in the category — it knows when NOT to fire, which almost no skill gets right.

The one thing: Don't use it to write prompts for other skills. Meta-prompting a skill that will prompt an agent that will... just write the prompt yourself.

6. codex-ppt — 9.2/10

Builds presentation decks as unified image sequences, with per-slide subagents enforcing a style contract.

Why: A 30-slide deck keeps visual consistency where single-context generation drifts. Approval gates mean nothing ships without your sign-off.

The one thing: It needs an image-generation backend. Spend the five minutes on setup or the output will disappoint.

7. pdf (official) — 8.6/10

PDF reading, merging, splitting, form-filling, extraction.

Why: A gated form-filling workflow verifies field mappings before writing anything.

The one thing: Don't expect it to beat PDF itself. Scanned documents and odd encodings still fall over — that's the file format's fault, not the skill's.


The pattern across the category

The official skills dominate because of verification: they check their own output instead of claiming success. The third-party challengers win on alternative philosophies — image-based decks instead of XML wrestling. What loses is everything in the middle: prompt-wrappers that call themselves document skills but don't verify anything.

Every skill above was scored on the same public rubric, and each one's full scorecard — with written rationale and a security scan of every network call — is on the document collection page.