What Are Claude Skills? How They Work and When to Use Them
Claude skills (Anthropic calls them Agent Skills) are folders of instructions, and optionally reference files and scripts, that Claude loads on its own when a task calls for them. Each skill has a SKILL.md file with a name and a short description at the top. Claude reads only those descriptions until your request matches one, then pulls in the full instructions. The result: Claude follows your process for a specific job, every time, without you pasting the same long prompt into every chat.
What is inside a Claude skill?
A skill is a plain folder. At minimum it holds one file:
SKILL.md: a short header with the skill’snameanddescription, then the instructions in Markdown. The description is the trigger: it tells Claude what the skill does and when to use it.- Reference files (optional): templates, examples, checklists or longer guides the instructions point to.
- Scripts (optional): code Claude can run for steps that should be exact rather than improvised, such as formatting a file.
Because it is just files, you can read a skill, edit it and share it like any document.
How does Claude decide when to use a skill?
Skills load in stages, which keeps them cheap:
- At the start of a conversation, Claude sees each installed skill’s name and description, nothing more.
- When your request matches a description, Claude reads that skill’s
SKILL.md. - If the instructions point to a reference file or script, Claude opens it only when the step needs it.
This is why the description matters more than anything else in a skill. A vague description means the skill never triggers; a precise one means it triggers when you’d expect. You don’t have to name the skill in your message, though you can.
How are skills different from prompts, Projects and MCP?
They solve different problems, and they work together.
| Tool | What it holds | When it applies | Best for |
|---|---|---|---|
| A prompt | Instructions for one request | Only the message you paste it into | One-off tasks |
| Custom instructions | Your general preferences | Every conversation | Tone, format, how you like answers |
| A Project | Files and instructions for one body of work | Chats inside that Project | Ongoing work on one topic |
| A skill | A repeatable procedure plus its references | Whenever a request matches, in any chat | Jobs you do the same way again and again |
| MCP connector | Access to an outside tool or data source | When Claude needs that tool | Reading or acting in other apps |
The short version: a Project gives Claude knowledge about a body of work, a connector gives Claude access to a tool, and a skill teaches Claude how to do a specific job. A skill can even tell Claude how to use a connector well.
Where can you use Claude skills?
Skills work in Claude on the web and desktop apps (with code execution turned on), in Claude Code in the terminal, and through the Claude API for developers. The format is the same everywhere, so one skill folder works across all of them.
Installing them is a separate topic with its own steps for each place. Our guide on how to install Agent Skills in Claude walks through Claude.ai and Claude Code, checking a skill loads, and fixing the usual problems.
When is a skill worth it?
A skill earns its place when three things are true:
- You do the job repeatedly. Writing a weekly newsletter, drafting carousels, auditing a landing page.
- There is a right way to do it. A process, a checklist, a house style you’d otherwise explain again each time.
- Quality drifts without it. If Claude’s output on the task varies from chat to chat, a skill pins it down.
For a one-off question, a skill is overkill. Just ask. For context that is about a project rather than a procedure, a Project is usually the better home.
What do skills look like for a content brand?
For a solo operator running a brand, skills turn content work into repeatable steps. Some common examples:
- A brand interview skill that writes a brand file other skills read, so every output keeps the same brand voice.
- A planning skill that sets your content pillars for social media and fills a calendar from them.
- A repurposing skill that turns one idea into posts for each platform, the method behind content repurposing.
- Channel skills that know each platform’s formats, from Threads post ideas to scripts for faceless YouTube channel ideas.
If you work in ChatGPT instead, the same procedures can live as copy-paste prompts. Our list of ChatGPT prompts for Instagram shows the prompt-based version of that approach. For how these pieces fit a one-person operation, see how to run a faceless brand without a content team.
What are the limits of skills?
- They guide; they don’t guarantee. Claude still uses judgment, so a badly written skill gives inconsistent results.
- They need the right plan and settings in the Claude apps, including code execution.
- They don’t act on their own. A skill runs when you ask for something. It won’t post, schedule or send anything unless you connect tools that do.
- Only install skills you trust. A skill can include scripts, so treat one from an unknown source like any other software.
FAQ
Are Claude skills the same as Agent Skills?
Yes. Agent Skills is Anthropic’s name for the format, and “Claude skills” is how most people search for it.
Do I need to know how to code to use skills?
No. Most skills are written instructions in Markdown, and installing them takes a few clicks in Claude.ai.
Can I write my own skill?
Yes. Create a folder with a SKILL.md that has a clear name, a precise description of when to use it, and step-by-step instructions.
Do skills work in ChatGPT?
No, skills are a Claude feature. You can still reuse the instructions inside a skill as a prompt in ChatGPT, without the automatic loading.
How many skills can I install?
You can install many, because Claude reads only their short descriptions until one is needed. Keep descriptions distinct so Claude picks the right one.
Get a content system built for running a faceless brand. Nexin Core runs a brand interview and writes a brand file that every other Nexin module reads, so its tools for Instagram, YouTube, Threads and the rest write in one voice, in Claude or ChatGPT through one connector. Nexin writes; it doesn’t post or schedule, and it isn’t for you if you don’t use Claude or ChatGPT. See the Core module or see pricing.