Claude Skill Guide: New Features and How Skills Work

Claude Skill Guide: New Features and How Skills Work

Claude Skill is becoming one of the most useful parts of Anthropic's ecosystem, although the name makes it sound smaller than it is. A Skill is not a single shortcut or a clever saved prompt. It is a reusable package that can teach Claude how to perform a particular kind of work, complete with instructions, reference material, templates, and scripts.

That distinction matters. A prompt tells Claude what to do once. A Skill gives Claude a repeatable method it can recognize and use again, often without you having to name it. The result is less time pasting the same brand rules, report structure, or internal process into every new conversation.

Anthropic has also expanded Skills well beyond their original developer-focused use. They are now available across Claude plans, the Claude API, Claude Code, and managed team environments. Recent additions include easier sharing, organization-wide provisioning, an open Agent Skills standard, and a broader set of built-in document capabilities.

Here is what Claude Skills actually do, what is new, and where they fit alongside Projects, connectors, MCP, and Claude Code.

What Is a Claude Skill?

A Claude Skill is a folder built around a file called `SKILL.md`. That file explains what the Skill does, when Claude should use it, and how the task should be completed. The folder can also contain scripts, examples, templates, style guides, or other reference files.

Imagine a company that creates a weekly sales report. The finished report always uses the same spreadsheet tabs, formulas, charts, terminology, and executive-summary format. Without a Skill, an employee has to explain those requirements each week. With a Skill, the company can package the method once. Claude loads it when a request matches the job and follows the established process.

This makes Skills useful for much more than writing. A custom Skill might:

• format presentations using a company's fonts and slide layouts
• analyze customer interviews with a fixed research framework
• turn meeting notes into a standard project brief
• check code against a team's engineering conventions
• create Jira or Linear issues with the required fields
• generate a financial report from an approved spreadsheet template

Simple Skills can be written entirely in Markdown. More advanced ones can include executable code for calculations, validation, file conversion, or other deterministic steps.

How Claude Skills Work

Claude does not load the full contents of every installed Skill into every conversation. Instead, it uses a system Anthropic calls progressive disclosure.

First, Claude sees lightweight metadata, mainly the Skill's name and description. If your request appears relevant, it opens the main instructions in `SKILL.md`. It reads additional reference files or runs bundled scripts only when the task requires them.

This staged approach solves a practical problem. A user or organization may have dozens of Skills, and loading all of them at once would fill the context window with instructions that have nothing to do with the current job. Progressive disclosure keeps Skills discoverable without making every conversation heavier.

It also means the description is unusually important. A Skill can contain excellent instructions and still be ignored if its description does not clearly state what it does and when it should be used.

What Is New in Claude Skills?

Skills Are Now Available on Every Claude Plan

Anthropic now lists Skills for Free, Pro, Max, Team, and Enterprise users. Code execution and file creation must be enabled because Skills operate inside Claude's execution environment.

Individual users can find them under Customize > Skills. Team and Enterprise organizations have additional controls, while developers can use Skills through Claude Code or the Claude API.

This broader availability changes the audience. Skills are no longer only for people building agents through an API. A marketer can create a campaign-brief Skill, an analyst can package a reporting method, and a founder can turn a recurring operating process into a reusable tool without building a separate app.

Team Sharing and Organization-Wide Provisioning

One of the most important recent additions is centralized distribution. Team and Enterprise owners can provision approved Skills across the organization, so employees do not have to upload the same package one by one.

Organizations can also allow people to share a Skill with selected colleagues or publish it to an internal directory. Shared Skills are view-only for recipients, and updates flow through automatically. That gives teams a workable middle ground between a personal prompt library and a fully maintained internal application.

There is a governance benefit too. An approved legal-review or brand-writing Skill can give people a consistent starting point while the owner remains responsible for the source version. It does not remove the need for review, but it reduces the quiet drift that happens when everyone keeps a slightly different copy of the instructions.

Styles Are Moving Into Skills

Anthropic is folding Claude Styles into the Skills system. Styles were mainly about how Claude wrote: concise, explanatory, formal, or matched to a user's preferred tone. Skills can preserve that behavior while also carrying examples, templates, and task-specific rules.

The move reflects a broader idea: tone is rarely separate from workflow. A customer-support response may need a particular voice, but it may also require an escalation rule, a refund policy, and a fixed structure. A Skill can keep those pieces together instead of treating style as a cosmetic setting.

Agent Skills Is an Open Standard

Anthropic has published the Agent Skills format as an open standard at agentskills.io. A Skill is therefore not meant to be a proprietary Claude-only macro. Other tools can adopt the same folder structure and specification, and a reference SDK is available for developers implementing support.

Portability will depend on the runtime. A Skill that contains general Markdown instructions may transfer easily, while one that depends on Claude-specific tools or a particular execution environment may need adjustment. Even so, an open format makes it safer for teams to invest in documenting workflows this way.

Built-In Skills Create Better Office Files

Anthropic provides built-in Skills for PowerPoint, Excel, Word, and PDF work. Claude can invoke them automatically when a request involves creating or editing those file types.

These are easy to overlook because they often work behind the scenes. Ask Claude to build a presentation from a report or create a spreadsheet with formulas and charts, and the relevant Skill supplies a more disciplined creation process than a generic text prompt would.

The practical improvement is not merely that Claude can produce a file. The Skill can guide structure, formatting, validation, and rendering, bringing the result closer to something a person can actually use.

Skills Work Across Claude's Agent Products

Custom Skills now span the main places people use Claude:

• Claude.ai: upload Skills and use them in ordinary conversations
• Claude Code: keep personal Skills in your home directory or project Skills inside a repository
• Claude API: upload custom Skills to a workspace and attach them to API requests
• Managed agents: attach built-in or custom Skills to longer-running agents
• Claude Platform on AWS and Microsoft Foundry: use supported Skills in enterprise deployments

The setup is not identical everywhere. Claude Code discovers Skills directly from the filesystem, while API integrations refer to uploaded Skills by ID. The runtime also differs: API Skills do not have open internet access and cannot install packages while running, whereas Claude Code inherits the access available on the user's computer.

Claude Skill vs. Project, Prompt, MCP, and Connector

These features overlap just enough to be confusing.

A prompt is best for a one-time instruction. If you only need a different tone for one email, creating a Skill would be unnecessary.

A Project keeps conversations and background knowledge together around an ongoing subject. It is a good home for product documents or research material. A Skill is more procedural: it describes how to carry out a repeatable task and can be reused across unrelated conversations.

A connector or MCP server gives Claude access to an external system such as Google Drive, Slack, GitHub, or a company database. A Skill tells Claude how to use available information and tools to complete a workflow. The two are often strongest together: the connector supplies access; the Skill supplies the method.

In Claude Code, Skills also overlap with slash commands. A command is explicitly invoked by the user. A Skill can be selected automatically when its description matches the task, and it can include a richer collection of resources. Commands remain useful when you want a predictable, named action on demand.

How to Create a Claude Skill

For a basic Skill, create a folder and add a `SKILL.md` file with YAML metadata at the top:

~~~markdown

---

name: customer-interview-summary

description: Analyzes customer interview transcripts and creates a product research summary. Use for interview notes, discovery calls, and user research transcripts.

---

Customer Interview Summary

Instructions

1. Separate direct evidence from interpretation.

2. Group findings by customer goal, pain point, and workaround.

3. Include short supporting quotations.

4. End with open questions, not unsupported recommendations.

~~~

You can then add reference material such as a report template, a glossary, or examples of good output. If the workflow needs a reliable calculation or transformation, place a script in the Skill folder and explain when Claude should run it.

On Claude.ai, package the folder as a ZIP file and upload it from the Skills settings. In Claude Code, place it in the appropriate Skills directory. API users upload it through the Skills API and reference the returned Skill ID in the container configuration.

What Makes a Good Claude Skill?

The best Skills are narrower than most first drafts. “Help with marketing” is too vague to trigger reliably and too broad to produce consistent work. “Create a launch brief from product notes using our six-section template” has a clear input, method, and output.

It also helps to separate guidance from reference material. Keep the main `SKILL.md` concise, then link to detailed policies or examples that Claude can open when needed. This preserves the benefit of progressive disclosure.

Test the Skill with real phrasing, including requests that should not trigger it. If users normally say “turn this call into research notes,” do not describe the Skill only as a “qualitative synthesis framework.” Claude needs language that connects the stored capability with the way people actually ask for it.

Finally, use scripts for work that should not depend on interpretation. Calculating totals, checking a file schema, or confirming required fields is better handled deterministically. Claude can decide when to run the check; the script can make the check repeatable.

Limits and Security Considerations

A Skill can contain executable code, so installing one deserves the same care as adding a small software package. Review unfamiliar scripts, check which files and tools they access, and avoid installing Skills from sources you do not trust.

Skills also inherit the limits of their environment. In the Claude API, they cannot browse the web or install a missing dependency at runtime. In Claude Code, they may have broader local and network access. A workflow that succeeds on one surface may therefore need changes on another.

Anthropic also notes that Agent Skills are not eligible for Zero Data Retention. Organizations with strict retention requirements should review the current policy and their account configuration before using the feature with sensitive material.

Most importantly, a Skill improves consistency; it does not guarantee correctness. A polished template can make a weak conclusion look authoritative. Important legal, financial, medical, security, or production decisions still need appropriate human review.

Are Claude Skills Worth Using?

If you repeat the same multi-step instructions more than a few times, probably. Skills are particularly valuable when the quality of the result depends on a method, a house style, or a set of reference files that Claude would not otherwise know.

They are less useful for casual requests and constantly changing processes. A short prompt is easier to maintain when the task is genuinely one-off.

The larger opportunity is organizational memory. Teams accumulate good ways of doing work, but those methods are usually scattered across documents, example files, and the heads of experienced employees. Claude Skills provide a relatively simple way to package that knowledge where an agent can discover and apply it.

That is the real new feature: not another button in the chat window, but a practical layer between a general AI model and the way a person or company actually works.

Frequently Asked Questions

What is a Claude Skill?

A Claude Skill is a reusable folder of instructions and optional scripts, templates, and reference files. Claude loads it when a request matches the task it was designed to handle.

Are Claude Skills free?

Anthropic lists Skills for Free, Pro, Max, Team, and Enterprise users. Code execution and file creation must be enabled. Some team management and sharing features require Team or Enterprise plans.

Do I need to code to create a Claude Skill?

No. A simple Skill only needs a Markdown file with a name, description, and clear instructions. Code is optional and is useful for workflows that require calculations, validation, or file processing.

Can Claude use a Skill automatically?

Yes. Claude checks the descriptions of available Skills and loads a relevant Skill when a request matches it. Users can also mention a Skill directly when they want to make the choice explicit.

Can I share a Claude Skill with my team?

Yes. Team and Enterprise accounts can support direct sharing, internal discovery, and organization-wide provisioning, depending on the controls enabled by the organization owner.

Is a Claude Skill the same as MCP?

No. MCP usually connects Claude to tools or external data. A Skill packages instructions and resources that explain how to perform a workflow. A Skill can use tools supplied through MCP.

Sources

Anthropic Help Center: What are skills?

Anthropic Help Center: Use skills in Claude

Anthropic Help Center: Provision and manage skills

Claude Platform Docs: Agent Skills overview

Claude Platform Docs: Skill authoring best practices