What Are Claude Agent Skills? A Plain-English Explainer for Business Owners

Imagine hiring a brilliant new employee who already knows everything about the world but nothing about how your business works. You wouldn’t rewrite their brain — you’d hand them a folder: how your reports are formatted, your quoting checklist, your brand voice guide, a few scripts that handle the boring parts.

That folder is, in essence, what Anthropic’s Agent Skills are for Claude. Introduced in October 2025, Skills are reusable packages of instructions, scripts, and reference materials that turn Claude from a general-purpose chatbot into a specialist for your domain — and they load on demand, so you write the guidance once instead of repeating it in every conversation.

What Claude Agent Skills Actually Are

A skill is a folder of files, not a setting or plugin. At its center sits a file called SKILL.md with a small header carrying two required fields:

• name — a short identifier in lowercase letters, numbers, and hyphens (up to 64 characters).
• description — a plain-language summary (up to 1,024 characters) saying both what the skill does and when Claude should use it.

Around that file, a skill can bundle extra guides, templates, database schemas — and executable scripts that do deterministic work, like validating a form or filling a spreadsheet. Anthropic describes the structure as “an onboarding guide you’d create for a new team member,” and that analogy is the whole idea.

Unlike a prompt — a one-off instruction inside a single conversation — a skill persists across conversations and activates automatically whenever your request matches its description. Once installed, you don’t invoke it; Claude does.

How Skills Work: Progressive Disclosure

The cleverest part is how skills avoid burning through the context window. Anthropic calls it progressive disclosure, working in three levels:

1. Level 1 — Metadata (always loaded). At startup, Claude loads only each skill’s name and description — roughly 100 tokens per skill. You can install dozens with almost no context penalty.
2. Level 2 — Instructions (loaded when triggered). When your request matches a skill’s description, Claude reads the full SKILL.md body — typically under 5,000 tokens.
3. Level 3 — Resources and code (loaded as needed). Claude reads or runs only the extra files the task needs. Scripts execute through bash, and only their output enters the context — the code itself never does.

The payoff: a skill can bundle extensive documentation or templates at zero token cost until they’re actually needed.

Pre-Built Skills You Can Use Today

You don’t have to build anything to benefit. Anthropic ships pre-built Agent Skills for everyday document work:

• PowerPoint (pptx) — create presentations, edit slides, analyze deck content.
• Excel (xlsx) — build spreadsheets, analyze data, generate reports with charts.
• Word (docx) — create and format documents.
• PDF (pdf) — generate formatted PDF documents and reports.

They’re available on claude.ai (active automatically when you create documents — no setup), the Claude API, AWS, and Microsoft Foundry. Anthropic also publishes open-source skills on GitHub, including a Claude API reference skill bundled with Claude Code.

Where Skills Work (and One Gotcha)

• Claude API — reference pre-built skills by ID (pptx, xlsx, docx, pdf) or upload custom skills via the Skills API (/v1/skills). Requires the code execution tool; skills run in a sandboxed container with no network access. Custom skills are shared workspace-wide.
• Claude Code — filesystem-based, no upload needed: drop skills into ~/.claude/skills/ (personal) or .claude/skills/ (project). The pre-built document skills are not available here.
• claude.ai — upload skills as zip files via Settings > Features. Available on Pro, Max, Team, and Enterprise plans with code execution enabled; custom skills are per-user, not centrally managed.

The gotcha: custom skills do not sync across surfaces. A skill uploaded to claude.ai isn’t on the API, and a Claude Code skill isn’t in your chat app. Upload separately to each place your team works.

What This Means for a Small Business

The pattern is always the same: take a repeatable workflow, write it down once as a skill, and let Claude execute it consistently.

• Brand voice, enforced. A skill with your tone guide and approved examples keeps every product description and ad on-brand — no one pastes the guide into chat again.
• Quoting and proposals. Package pricing rules, discount limits, and your proposal template. Staff type “draft a quote” and get a document that follows your process.
• Review replies. A skill with your response playbook keeps 1-star and 5-star replies consistent no matter who is on shift.
• Document processing. The pre-built PDF and Excel skills extract tables from supplier invoices, reformat price lists, and generate monthly reports with charts.
• Onboarding SOPs. Turn standard operating procedures into skills so every session follows the steps your best employee would.

Skills are also composable — Claude can combine several for multi-step tasks, like pulling numbers with a spreadsheet skill, then formatting the result with your brand skill.

A Word on Security

Skills give Claude new capabilities through instructions and code, so Anthropic’s guidance is blunt: use skills only from trusted sources — ones you created or got from Anthropic. Before installing any third-party skill, audit every bundled file and watch for unexpected network calls or file access that doesn’t match the skill’s purpose. In Anthropic’s own words: treat a skill like installing software. Recent agent safety incidents show why that discipline matters.

The Bottom Line

Prompts are for one-off tasks. Skills are for the way your business always does things. They cost almost nothing in context until needed, they bundle instructions with reliable scripts, and they turn tribal knowledge into automatic behavior. Start with the pre-built document skills — they’re already on — then pick one repetitive workflow and write it down once. And if you’d rather keep your AI entirely on your own machines, see our LM Studio tutorial. That’s the whole pitch: expertise you install, not expertise you retype.

Frequently Asked Questions

What’s the difference between a skill and a prompt?

A prompt is a one-off instruction in a single conversation. A skill is a saved folder of guidance Claude loads automatically whenever relevant — write it once, and it works across every future conversation.

Do I need to code to use Agent Skills?

No. Pre-built skills need no setup on claude.ai or the API. Simple custom skills are mostly clear markdown instructions; code is optional, for when you want deterministic scripts inside a skill.

Will my skill work everywhere I use Claude?

Not automatically. Custom skills don’t sync across surfaces — upload separately to claude.ai, the API, and Claude Code. On claude.ai, custom skills are also per-user, not centrally managed.

Are third-party skills safe to install?

Treat them like software. Install only from trusted sources, audit all bundled files first, and be wary of skills that fetch data from external URLs — fetched content can carry instructions you didn’t expect.