What are Claude Skills, and how does a team use them?
Example
Sandra Vogel, team leader of order processing at Haldenstein Verpackungen (example; company name fictitious), explains the structure and approval process to Claude anew with every order confirmation. Her colleague, however, comes up with a different result. Is there a way for her to document this information so that the outcome is consistent across the entire team?
Claude Skills are folders containing instructions that provide Claude with domain-specific knowledge and fixed workflows: a structure, your terminology, and your approval rules. Claude automatically applies a Skill when its description matches the task at hand. You create it once; on Team and Enterprise plans, it can be shared (according to Anthropic, as of November 2026).
Example
Product Information: Anthropic, as of November 2026. Author: Nico Röpnack. He has a background in engineering: over 20 years in the industry, ranging from startup planning to leading global engineering teams. Over 300 participants in webinars and training sessions.
Instructions
A skill works like a restaurant chain’s recipe book: The recipe is set in stone, every kitchen prepares the dish the same way, and any improvements are fed back to the source that maintains the recipe. Sandra Vogel uses these examples in her order processing work. Product specifications according to Anthropic, as of November 2026.
A task that repeats itself and always follows the same structure. For Sandra Vogel, this is the order confirmation: the same structure, the same approval rule, for every order. Information that always applies—such as a price list—should be saved as a file in a project. A workflow that Claude should follow on his own when the task calls for it should be defined in a skill.
A skill is a folder containing a file named SKILL.md. At the top are the name and description—that is, what the skill does and when it applies—followed by the instructions: for Sandra, the structure and sharing rules for the order confirmation. According to Anthropic, you should first enable code execution and file creation under Settings > Capabilities. Then upload the folder as a ZIP file: Customize > Skills > Create skill > Upload a skill.
Perform the same task twice—once without the skill and once with it—and compare the results side by side. Sandra checks her order confirmation to see if the structure and approval rules are correct. In the course, the difference looks like this: Without the skill, the result is decent but generic; with the skill, Claude follows the predefined structure. Test again after each change.
Pre-built skills are a starting point; they become effective when you customize them to fit your template, terminology, and approvals. According to Anthropic, sharing is available on Team and Enterprise plans. The owner (the person who manages the Team account) enables sharing, after which you can share via the three-dot menu > Share. Shared skills are read-only; recipients can enable them and receive updates. Sandra’s colleague works with the same skill in this way.
Common Mistakes
Three reasons why a skill might not work as expected.
The finished skill is not familiar with your template. A finished skill produces a neat but generic result: a different structure, different terms, and no approval rule. It only knows what’s in its folder. It doesn’t know when an order confirmation needs to be approved at Haldenstein.
The description is too vague. According to Anthropic, the description determines when a skill is applied. A vague description gives Claude little indication of whether the skill is appropriate for the task.
The skill is available in only one location. According to Anthropic, user-created skills do not transfer between the Claude app, the API, and the Claude code. Users who access Claude from multiple locations must upload the skill separately for each interface (as of November 2026).
Here's how to avoid it
Define the structure, terms, and approval rules in the skill. Sandra's skill outlines the structure of the order confirmation and indicates when approval is required. A skill controls Claude's behavior; it does not technically enforce anything.
More on thisIn the description, explain what the skill does and when it takes effect. Anthropic also recommends including examples of input and output, as well as running a test after every change.
More on thisUpload the skill to the platform where your team works, and test it there. Passwords and keys do not belong in a skill; test third-party skills before activating them.
More on this
Frequently asked questions
Answers based on the manufacturer's specifications as of November 2026.
Documents
The Lighthouse Academy has shown us how to use AI tools in a truly meaningful way in our day-to-day work. No theoretical buzzword bingo—just practical, immediately applicable knowledge.
.0 out of 5 stars from 37 Google reviews for Lighthouse Consultings. Read all reviews on Google. As of November 2026
Next step
Sandra sets up her order confirmation as a skill, and her colleague follows the same instructions. During the initial consultation, we’ll determine which processes in your team are suitable for this.
Select an appointment directly, free of charge and without obligation.
Get started for free with n8n, the open-source automation platform.
Briefly describe your situation. We will get back to you within 24 hours.