Claude Skills and Plugins in 2026: What They Are and Which Ones Matter for Work

The short version: a skill is a folder of instructions Claude loads when a task needs it, changing how Claude works rather than what it can reach. Four ship with Claude and handle most office work, active by default on paid claude.ai plans:

• PowerPoint for decks
• Excel for spreadsheets and analysis
• Word for documents
• PDF for reports and fillable forms

Beyond those, custom skills are straightforward to write, since the only required file is a single SKILL.md, and plugins are bundles that package skills and connectors together for Claude Code and Cowork. The catch worth knowing before you plan a rollout: a custom skill uploaded to claude.ai belongs to the person who uploaded it, with no org-wide sharing, and it does not sync to the API.

This page is written for people doing business work rather than building software, which is the gap in most coverage of skills. The leaderboards ranking thousands of community skills by install count are genuinely useful if you write code; they are close to useless if you want to know whether any of this changes your Tuesday.

Skills, plugins, connectors: which is which

One paragraph of vocabulary, because these three get used interchangeably and mean different things:

• A skill is packaged know-how: instructions, examples, and sometimes scripts that Claude follows. It reaches nothing outside itself.
• A connector is the opposite: a live link to a tool such as Gmail, Notion, or a shared inbox, so Claude can read and act there.
• A plugin is a bundle: skills, connectors, and tools packaged together so a team installs a whole capability at once, mainly in Claude Code and Cowork.

The short test: if it changes how Claude behaves, it is a skill. If it gives Claude access to something, it is a connector. If it is several of those installed together, it is a plugin. Our full vocabulary guide covers the edge cases, including what the same words mean on the ChatGPT side.

How skills actually work, in plain English

This is the part that explains why skills are cheap to have and worth writing, and almost nobody covers it.

Claude loads a skill in three stages:

1. Metadata, always loaded. Just the name and description, around 100 tokens per skill. This is what Claude matches against your request to decide whether the skill is relevant.
2. Instructions, loaded when triggered. The body of SKILL.md, kept under roughly 5,000 tokens. Claude only reads this once it decides the skill applies.
3. Resources and scripts, loaded only if used. Extra files and code cost nothing until the instructions actually call for them, and when a script runs, only its output enters the conversation, not the code.

Two practical consequences. Having a dozen skills installed costs almost nothing until you use one. And the description field is the most important thing you will write, because it is the only part Claude sees when deciding whether to reach for your skill at all. A vague description means a skill that never triggers.

The four skills that ship with Claude

These are Anthropic's own, active by default on claude.ai for paid plans, and available through the API and on AWS and Microsoft Foundry:

For most business users this is the whole story. If your work involves producing documents, these four cover it, they need no installation, and you have probably already used one without knowing it had a name.

Anthropic also publishes open-source skills at its own repository, including a Claude API skill that carries current API reference material for eight programming languages, which is bundled with Claude Code.

Where skills work, and where they do not

Availability differs by surface more than you would expect, and this table is the thing to check before promising anyone a rollout:

The limits worth knowing before you build a workflow

Stated plainly, because the enthusiastic coverage tends to skip them, and all four come from Anthropic's own documentation as of August 2026:

• No org-wide sharing on claude.ai. A custom skill uploaded in the app is scoped to that individual. There is no central admin management, so "everyone in support uses our reply-style skill" is not yet a thing you can switch on from the app.
• No syncing between surfaces. A skill on claude.ai does not appear in the API, and the reverse is also true. Teams that build one for both maintain it twice.
• No network access on the API. Skills running there cannot call out to the internet or install packages at runtime, so anything needing live data belongs in a connector instead.
• Skill definitions are retained under standard data policy and are not eligible for zero data retention, which matters if your instructions would contain anything sensitive. Keep credentials and customer data out of skill files.

Plugins: skills and connectors bundled together

Where a skill is one capability, a plugin is a package of them, aimed at teams who want to install a whole way of working rather than assemble it. In Claude Code, plugins are how Anthropic distributes its own open-source skills, and plugin marketplaces let organisations publish internal bundles for their people. Cowork uses the same idea.

For a business team the practical read is: you probably do not need to think about plugins until you have written a skill worth sharing, at which point a plugin is how you hand it round.

Writing your own skill is genuinely approachable

The barrier is far lower than it looks, and it is worth knowing because the highest-value skills are the ones only your company can write: your tone of voice, your escalation rules, your report format, your review checklist.

A skill is a folder with one required file. That file, SKILL.md, needs YAML frontmatter with a name and a description, then plain markdown instructions underneath. No code is required unless you want it. The steps are in the how-to at the top of this page, and Anthropic ships a skill-creator skill that walks you through it interactively, which is the fastest route if you would rather be guided than read a spec.

The one piece of advice that matters most: spend your effort on the description. It has up to 1,024 characters, it must say what the skill does and when to use it, and it is the only thing Claude reads when deciding whether your skill is relevant.

Skills plus connectors: where this gets genuinely useful

A skill on its own changes how Claude writes. A connector on its own gives Claude something real to work with. Together they are the setup worth building toward, and the example that makes it concrete is customer email.

A connector gives Claude your actual queue: the threads waiting, who they are from, what was already said. A skill teaches it your support voice, your refund policy, and when a conversation should go to a human instead. Neither half is much use alone. The reply that sounds like your team and references the right thread needs both.

Disclosure, since it is our own product: Drag is a shared inbox for Gmail teams and publishes an MCP server, so it is the connector half of that pairing rather than a skill. The connector guides cover that side, and Drag is free for 7 days if you want to try the combination on a real queue.

Frequently asked questions

What is a Claude skill?

A folder of instructions, examples, and optional scripts that Claude loads when a task matches its description. It packages expertise, so instead of explaining your report format every time, you write it once and Claude follows it. Skills change how Claude works; they do not connect it to other software, which is what connectors do.

Do I need to pay for Claude skills?

The four pre-built document skills are included and active by default on paid claude.ai plans. Uploading custom skills to claude.ai requires a Pro, Max, Team, or Enterprise plan with code execution enabled. Claude Code discovers custom skills from your filesystem, and the API supports both with the relevant beta headers.

Can my whole team share a Claude skill?

Not from claude.ai as of August 2026. Custom skills uploaded in the app are scoped to the individual who uploaded them, with no organisation-wide management. Teams that need shared skills use the API, where skills are workspace-wide, or distribute them through a repository for Claude Code.

What is the difference between a Claude skill and a connector?

A skill is instructions Claude follows, with no access to anything outside itself. A connector is a live link to a tool such as Gmail, a CRM, or a shared inbox, letting Claude read and act there. Skills shape behaviour, connectors provide reach, and most useful setups combine one of each.

How do I write a Claude skill without coding?

Create a folder, add a SKILL.md file with a name and description in YAML frontmatter, then write your instructions in plain markdown below. No code is required. Anthropic's skill-creator skill guides you through the process interactively, and the description field deserves most of your attention because it determines whether Claude ever triggers the skill.

Related guides: vocabulary, connectors, and the ChatGPT side

The vocabulary in full, including what these words mean on the other assistant, is in MCP server vs connector vs plugin vs skill. For the connector half of the pairing: the best Claude MCP servers audits the business options and the no-code connector guide covers setup and first prompts. On the ChatGPT side, the plugins and connectors guide maps that ecosystem's equivalent pieces.