Anthropic skills, officially called Agent Skills, are folders of plain instructions that Claude loads on demand when a task matches the skill's description. A skill is not a plugin, an app, or a chatbot. It is a text file (SKILL.md) that tells the model how to do one job well, optionally backed by scripts and reference files. This article explains how the format works, how it compares with ChatGPT custom GPTs, Gemini Gems and Copilot agents, and where you can get ready-made skills, including the KissMySkills agent skills marketplace.
What an Anthropic skill actually is
Anthropic's Agent Skills format is deliberately simple. A skill is a folder. Inside the folder there is one required file, SKILL.md, and any number of optional supporting files. That is the whole specification at a practical level. There is no compiler, no app store approval and no proprietary binary format.
SKILL.md has two parts:
- Frontmatter at the top of the file, containing at minimum a name and a description. The description is the most important line in the whole skill, because it is what the model reads to decide whether the skill is relevant to the current task.
- Instructions below the frontmatter, written in Markdown. This is where you describe the procedure, the standards to follow, the output format, the edge cases and anything else the model needs to do the job the way you want.
Optional extras live alongside SKILL.md: scripts the model can run (for example a Python file that formats a spreadsheet), reference documents it can read when needed (a style guide, a template, a checklist), and example outputs. The instructions can point to these files so the model only opens them when they are relevant.
How Claude decides to use a skill
Skills work through progressive disclosure. At the start of a session the model only sees the name and description of each installed skill, which keeps context usage low. When you ask for something that matches a description, the model loads the full SKILL.md and follows it. If the instructions reference a script or a supporting document, it reads or runs that too. If nothing you ask matches, the skill stays dormant and costs almost nothing.
This is why a vague description is the most common reason a skill "does not work". If the description says "helps with business", the model has no way to know when to reach for it. If it says "drafts weekly investor updates from a bullet list of metrics, in a fixed five-section format", the trigger is obvious.
Where skills run
Anthropic supports the same format across its surfaces: the Claude apps (web and desktop), Claude Code in the terminal, and the Claude Developer Platform (API), where skills can be attached to agents you build yourself. The same folder works in all three, which is the main practical benefit of a file-based format. You write it once and carry it with you.
For a slower, more foundational walkthrough see What are Claude skills. If you are unsure whether you need a skill or just a good prompt, read the difference between a prompt and a skill.
Agent skills vs custom GPTs vs Gems vs Copilot agents
All four are ways to give an AI model a persistent set of instructions so you do not retype them every time. The differences are in where the instructions live, who owns them and how far they travel.
| Aspect | Claude skill (Agent Skills) | ChatGPT custom GPT | Gemini Gem | Copilot agent |
|---|---|---|---|---|
| Where instructions live | A SKILL.md file in a folder on your machine or in your project | Inside OpenAI's GPT builder, stored on their platform | Inside Google's Gem manager, stored on their platform | Inside Microsoft's Copilot Studio or the agent builder, stored in your Microsoft tenant |
| Portability | High. Plain Markdown you own; copy it anywhere, version it in git, paste the instructions into other models | Low. Locked to ChatGPT; you can copy the text out by hand | Low. Locked to Gemini; same manual copy-out | Low. Tied to the Microsoft 365 environment |
| Supporting files | Scripts, templates and reference docs inside the folder | Knowledge file uploads inside the GPT | File uploads attached to the Gem | Connected data sources and SharePoint content |
| Sharing | Send the folder, publish a repo, or sell the file on a marketplace | Share link or the GPT Store | Share link within Google accounts | Share within the organisation |
| Cost model | Included with your Claude plan or API usage; skills themselves are free or one-time purchases | Requires a paid ChatGPT plan to build; usage included in the plan | Available on Gemini plans; usage included | Requires Microsoft 365 Copilot licensing |
| Best for | Repeatable professional workflows you want to own and reuse across tools | Consumer-facing chat assistants inside ChatGPT | Personal assistants inside Google Workspace | Internal business agents on company data |
The practical takeaway: a GPT, a Gem and a Copilot agent are configurations of someone else's product. A skill is a document you own. That distinction matters most when you switch tools, when a platform changes its rules, or when you want to sell or share what you built.
A skill file is not Claude-only in practice
Because the instructions are plain Markdown, the same text works as the system instructions of a custom GPT, the instructions of a Gem or the description of a Copilot agent. You lose the automatic triggering and the bundled scripts, but you keep the expertise. This is why skills sold on KissMySkills ship with a README explaining how to install the same file in each platform.
Which KissMySkills products work where
| Product type | Claude apps | Claude Code | ChatGPT (custom GPT) | Gemini (Gem) | Copilot |
|---|---|---|---|---|---|
| Skills (.md skill file + README) | Yes, native | Yes, native | Yes, as GPT instructions | Yes, as Gem instructions | Yes, as agent instructions |
| Generators (structured skills that produce documents, plans or assets) | Yes, native | Yes, native | Yes | Yes | Yes |
| Prompt libraries | Any chat | Any chat | Any chat | Any chat | Any chat |
Every skill on the store is a one-time purchase with a 30-day money-back guarantee, no questions asked. You can browse the full range in the Claude skills collection.
Where to get skills
Anthropic's own examples
Anthropic publishes example skills that show the format in use, covering things like working with documents and common file types. They are the best place to see how a well-structured SKILL.md reads, and they are free. Treat them as reference material rather than finished professional tooling: they demonstrate the mechanics, not a specific job.
Open-source repositories
Developers share skill folders on GitHub and similar platforms. Quality varies widely. Some are excellent, many are a description and three bullet points. When you pull a skill from a public repo, read it fully before installing it, especially any scripts, because the model will run what the skill tells it to run.
Marketplaces such as KissMySkills
An agent skills marketplace fills the gap between free examples and building everything yourself. On the KissMySkills Claude skills marketplace each skill is written for a specific role or workflow, tested on real tasks, packaged as a .md file plus a README, and priced as a one-time purchase. You get a file you own, not a subscription to a chatbot. For a sense of what finished skills look like across different jobs, see these Claude skill examples.
What a good skill contains
Whether you download, buy or write a skill, these five things separate the useful ones from the noise.
- A precise description. It names the task, the input and the output in one or two sentences, so the model triggers at the right moment and never at the wrong one.
- A defined procedure. Numbered steps, decision points and what to do when information is missing. A good skill tells the model to ask two clarifying questions before drafting, not to guess.
- Quality standards. What "done" looks like: tone, length, format, what to always include and what to never include. This is where a real practitioner's judgement gets encoded.
- Worked examples. One or two examples of good output help the model far more than another paragraph of adjectives.
- Scoped resources. Templates, checklists or scripts kept in separate files and referenced only when needed, so the skill stays fast and does not flood the context window.
If a skill you are looking at fails on more than two of these, it will behave like a mediocre prompt. The format does not make the model better; the content does.
Frequently asked questions
Are Anthropic skills the same as Claude skills?
Yes. Anthropic is the company, Claude is the model, and Agent Skills is the official name of the format. People use all three terms for the same SKILL.md folder.
Do I need to code to use a skill?
No. Most skills are pure Markdown instructions. Scripts are optional and only appear in skills that need to process files or run repeatable computations. Installing a skill is a matter of adding the folder where your Claude client expects it, or pasting the instructions into your chosen tool.
Can I use a Claude skill in ChatGPT or Gemini?
You can use the instructions. Paste the body of SKILL.md into a custom GPT's instructions or a Gem's instructions and it will behave the same way. You lose automatic triggering and bundled scripts, which are Claude-specific features.
Is a skill a subscription?
Not on KissMySkills. Skills are one-time purchases. You download the file and it is yours. There is a 30-day money-back guarantee if it does not fit your workflow.
How is a skill different from a prompt?
A prompt is a single request you type each time. A skill is a persistent, structured set of instructions with a trigger description and optional supporting files, loaded automatically whenever the task matches. Skills encode a process; prompts encode a request.
The bottom line
Anthropic's Agent Skills format is a folder with a SKILL.md file: a description that tells the model when to act and instructions that tell it how. Unlike a custom GPT, a Gem or a Copilot agent, the skill is a plain file you own and can move between tools. Start with Anthropic's examples to learn the shape, then pick up finished, role-specific skills from a marketplace when you want something that does a real job on day one.
Browse all Claude skills at KissMySkills.