Claude Skills Guide: 19 Official Skills Measured, 5 Run for Real

Claude Skills are folders of instructions that Claude opens on its own when a request matches their description. We measured all 19 skills in Anthropic's public anthropics/skills repository with a script, and we have installed and run five popular skills on real tasks in Claude Code. This page puts both together: what a skill is, what the official ones look like on the inside, which ones changed our results, and why a skill sometimes does not load.

Measured on 23 Oct 2026 against commit dbd4588 of the repository (dated 9 Oct 2026). The five hands-on tests were run between 8 and 13 Oct 2026 with Claude Code 2.1.293–2.1.296 on Linux; each one has its own write-up linked below.

At a glance

Key results

• 19 official skills, one template. The skills/ folder of anthropics/skills holds 19 skills plus a template folder. A typical SKILL.md is 148 lines (median, about 1,200 words); the shortest is 33 lines (internal-comms), the longest 617 (claude-api).
• 11 of the 19 ship scripts. docx, pptx and xlsx carry 12–15 script files each; eight skills are instructions only.
• Descriptions decide loading, and they vary a lot. The median description is 324 characters. Read one by one, 16 of 19 say when to use the skill ("Use this skill when…", "Trigger whenever…"); 3 (frontend-design, theme-factory, webapp-testing) only say what the skill is.
• One official skill is over the documented limits. claude-api has a 1,068-character description and a 617-line SKILL.md. Anthropic's docs cap the description at 1,024 characters and advise keeping the body under 500 lines. The same docs list "claude" as a reserved word that a skill name cannot contain.
• In our tests, the skill with the vaguest description was the one that did not load. frontend-design (204 characters, one of the three with no "when to use" line) stayed unused until we named it in the prompt. pptx, xlsx and our own skill-creator skill, whose descriptions list trigger words, loaded on their own. That is five tests, not a benchmark.

Hands-on test

Setup

• Repository: github.com/anthropics/skills, branch main, commit dbd4588f9e10, downloaded as a zip on 23 Oct 2026.
• Script: Python 3.13 with PyYAML. For every folder with a SKILL.md it parsed the YAML frontmatter and counted lines, words, description length and the other files in the folder. Whether a description "says when to use" the skill was judged by reading all 19 by hand.
• Hands-on tests: Claude Code 2.1.293–2.1.296, Opus 5.5 at high effort, on Linux, signed in with a Claude Max plan, auto mode on. Prompts and sample data were in Vietnamese; all data was made up.
• Not done: we did not run every official skill, did not test claude.ai or the API, and did not repeat runs enough to give trigger rates.

What are Claude Skills?

A skill is a folder with one required file, SKILL.md. Its top block (frontmatter) holds a name and a description; the rest is the instructions, and the folder may add scripts, templates and reference files. Claude keeps only the name and description of each installed skill in view. When a request matches a description, Claude reads the full SKILL.md, and only then the extra files it needs. That is why you can install many skills without filling the context window, and also why the description matters so much: it is the only part Claude sees before deciding.

Skills work in Claude Code, on claude.ai and through the Claude API. Custom skills do not sync between these: Anthropic's overview says a skill added on one surface has to be added again on each of the others. In Claude Code a skill is just a folder, either in ~/.claude/skills/ (all your projects) or .claude/skills/ in a repository, or it arrives inside a plugin. Our step-by-step install from GitHub, with a PowerPoint test: What are Claude Skills? How to install them.

Inside the 19 official skills

Three things stand out. First, the document skills (docx, pptx, xlsx, pdf) are small in text but heavy in code: they lean on scripts to build and check files, which matches what we saw when the pptx skill validated its own deck. Second, the longest descriptions are the ones that list exact triggers and exclusions ("Do NOT use for PDFs…"). Third, Anthropic's own claude-api skill breaks two rules from Anthropic's authoring docs: its description runs past the 1,024-character maximum, and its name contains "claude", which the docs list as reserved. We did not test whether either matters when you install it; we only report what the file says.

The five skills we ran, and what happened

Our ranking of the five, with the reasoning for each place: Best Claude Skills: 5 tested, ranked.

Why a Claude skill does not load

Put the two tables side by side and a pattern shows up, with the caveat that it rests on five tests. The three skills that loaded without being named (pptx, xlsx, our skill-creator output) all have descriptions that list concrete trigger words: file extensions, "deck", "slides", "spreadsheet". frontend-design describes itself as "guidance for distinctive, intentional visual design" and never says when to use it. A request for a café landing page did not match that wording closely enough, so Claude built the page without it. The repository's frontend-design/SKILL.md was last changed on 3 Sep 2026, before our test on 10 Oct, so the description we measured is the one that was installed.

What to do about it:

1. Name the skill when it matters. Typing "use the frontend-design skill" (or /skill-name for skills that offer a command) made it load in our test.
2. For your own skills, write the description as a trigger list. Say what the skill does and when to use it, with the words a user would actually type. Anthropic's docs recommend both parts.
3. Keep the body short. Move long reference material into separate files the skill points to, so Claude reads it only when needed.
4. Read what skill-creator writes. In our test its draft contained a calendar mistake: an example date labeled Friday that was actually a Saturday.

Which skills to install first

• Making documents (Word, PowerPoint, Excel, PDF): check before installing. In our Excel test the xlsx skill was already available in Claude Code; the document-skills plugin added nothing to the numbers. In a separate test, PDF invoices to Excel with Claude Code, Claude Code also loaded the xlsx skill on its own.
• Planning anything with real constraints: grill-me. It changed the outcome more than any other skill we ran.
• Repeating the same task every week: build your own skill with skill-creator, then test that it loads from a request that does not name it.
• Web design: frontend-design helps, but plan to name it in the prompt.

Skills can run scripts on your machine. Anthropic's docs say to use skills only from trusted sources, meaning ones you wrote yourself or got from Anthropic; for anything else, read the SKILL.md and its scripts before installing.

What worked

What worked

• Measuring the repository with a script gave numbers nobody has to take on trust: anyone can rerun it on the same commit.
• The trigger-word pattern lined up with all five of our hands-on results.

Where it fell short

What didn't work

• Our first pass used a keyword filter to spot "when to use" lines. It missed 4 of the 16 descriptions that have one ("Use it when", "Use for…", "TRIGGER"), so we dropped it and read all 19 by hand.
• Five tests, each run once or twice, are enough to spot a pattern, not to give a trigger rate. We did not measure how often each skill loads.

Our verdict

Claude Skills are worth using, and the official ones are well built, but whether Claude picks one up depends mostly on its description. Skills whose descriptions spell out when to use them loaded on their own in our tests; the one that only described itself did not. Install the few skills that match your work, name a skill in the prompt when you need it for sure, and write your own descriptions as a list of triggers.

Frequently asked questions

How many official Claude Skills are there?

On 23 Oct 2026 the skills/ folder of Anthropic's anthropics/skills repository had 19 skills plus a template. Separately, Claude Code signed in with a Claude account already offered an xlsx skill in our test, with nothing installed.

Why is my Claude skill not triggering?

Usually the description does not match the way you asked. Name the skill in the prompt to confirm it works, then rewrite the description so it says when to use the skill with the words you actually type.

How long can a skill description be?

Anthropic's docs set a maximum of 1,024 characters for the description and 64 for the name, and say the description should cover both what the skill does and when to use it. The official skills we measured range from 204 to 1,068 characters, median 324.

Do I need to install the Excel or PowerPoint skill?

Not always. In our test Claude Code used the xlsx skill with nothing installed. Check what loads on a small task before adding plugins.