How to Install and Use Skill Files in Claude and ChatGPT

If you use Claude or ChatGPT for work, you already know the drill. You open a new chat, and before you can get anything useful out of it, you spend five minutes re-pasting your brand voice guidelines, your report format, your pricing structure, or your “never use these words” list. Skill files fix this by letting you package instructions, examples, and reference files into a reusable folder the AI loads automatically whenever it’s relevant, so you stop re-explaining your business from scratch in every single conversation.

The copy-paste problem every AI user hits

Here’s a scenario you’ve probably lived through. You’ve got an 800-word prompt saved somewhere, maybe in a Google Doc, maybe in a Notes app, that covers your brand voice, your formatting preferences, and three examples of “good” output. Every time you start a new chat, you paste it in. Every time, you hope you didn’t leave out a detail. Every time, the AI treats it like the first conversation you’ve ever had, because it is.

This isn’t a you problem. It’s how chat-based AI tools have worked until recently. Each conversation starts at zero. The model has no memory of your standard operating procedures, your client’s preferences, or the format your boss likes reports in, unless you tell it again, every time.

Skill files change that. They’re a relatively new feature in both Claude and ChatGPT that let you save a set of instructions once and have the AI pull them in automatically when the situation calls for it. No re-pasting. No hoping you remembered everything. You write it once, and it’s there whenever you need it, on any device, in any new chat.

What is a skill file, exactly?

At its simplest, a skill is a single file called SKILL.md. That’s it. It’s a plain text file written in Markdown (a lightweight formatting language that mostly just looks like normal writing with a few symbols for bold text, headers, and lists) with a short block at the top, written in a format called YAML, that gives the skill a name and a description.

That description line matters more than anything else in the file. It’s what the AI actually reads to decide whether this skill is relevant to what you just typed. If your description is vague, like “helps with writing,” the AI has almost nothing to go on and the skill will sit there unused. If your description names specific situations in specific language, the AI recognizes the match and loads the skill automatically. We’ll come back to this because it’s the single most common reason a skill “doesn’t work.”

A skill doesn’t have to be just one file, though. It can be a folder that contains SKILL.md plus whatever supporting material you want to hand over: a reference document, a spreadsheet template, your brand style guide, even a small script. The AI only opens the additional files when the task actually needs them.

A useful analogy

Think about training a new contractor on your crew. You wouldn’t re-explain your entire process from memory every morning. You’d hand them a standard operating procedure binder and say “check the section that applies.” They flip to the relevant page, follow it, and move on. They don’t read the whole binder cover to cover every single time.

That’s exactly what a skill file does for Claude or ChatGPT. It’s a binder. The AI checks whether a page applies to what you’re asking, and if it does, it opens just that page and uses it. If it doesn’t apply, the skill stays closed and doesn’t clutter up the conversation.

What a skill is not

It’s worth being clear about what skill files aren’t, because the terminology in AI tools gets muddy fast. A skill is not a chatbot. It’s not a custom persona like a Custom GPT, where you’re essentially creating a branded assistant with a name and a personality. It’s not a plugin or an integration that connects to other apps or pulls in live data. A skill is instructions plus optional supporting files. That’s the whole thing. It’s closer to a very smart template than it is to a piece of software.

Why you’d actually want one

This only matters if it solves something you actually deal with. So here are five real situations small business owners and marketers run into constantly, and how a skill changes them.

Brand voice consistency

Before: every time you ask the AI to draft a social post, an email, or a blog intro, you re-explain that you’re casual but not sloppy, that you never say “synergy,” and that you like short punchy sentences. Half the time you forget a rule and the draft comes back sounding like a different business entirely.

After: a skill holding your tone rules, your banned words list, and two or three sample paragraphs that nail your voice. Every draft that comes out already sounds like you, without you typing a single instruction beyond the actual request.

Repeatable deliverables

Before: you write a monthly client report and manually rebuild the structure every time, hoping you didn’t forget a metric or a section your client expects.

After: a skill containing the exact section order, the metrics that matter for that type of client, and a filled-in example report. You ask for “the July report” and get something that’s structurally identical to last month’s, with new numbers dropped in.

Proposals and quotes

Before: you retype your pricing tiers and scope language every time you send a quote, which means every quote has a slightly different phrasing and, occasionally, an outdated price.

After: a skill holding your current pricing tiers, what’s in scope versus out of scope, and your standard terms. Every quote pulls from the same source, so nothing drifts.

Onboarding

Before: a new hire asks how you handle new client intake, and the answer lives in your head, or in a document nobody’s opened in eight months.

After: a skill that documents your actual intake process step by step. Anyone on the team who uses the AI for that task gets the same output, whether it’s their first week or their third year.

Social repurposing

Before: you take one piece of content and manually rewrite it for Instagram, LinkedIn, and Facebook, remembering character limits and hashtag habits as you go.

After: a skill with your per-platform rules baked in, character counts, hashtag policy, CTA style. One request produces properly formatted versions for each platform.

The pattern across all five is the same. You do the thinking once, put it in a file, and it compounds. It works for you today, it works for your team tomorrow, and it works in every future chat without you lifting a finger. If you’re already trying to run your AI tools like an actual system instead of a pile of disconnected chats, skill files are one of the more practical ways to do that. We wrote about the bigger version of this idea in Stop Collecting AI Tools. Start Running an AI Team., and skills are one of the concrete mechanisms that makes it real instead of aspirational.

How to install a skill in Claude

Skills are available across Claude’s Free, Pro, Max, Team, and Enterprise plans, though how you turn them on differs slightly depending on your account type.

Turn on the capability first

Before you can install anything, you need a setting enabled. If you’re an individual user, go to Settings, then Capabilities, and turn on “Code execution and file creation.” Without this switched on, skills won’t work at all, so this is the step people miss.

If you’re on a Team or Enterprise plan, this is managed at the organization level instead. An admin needs to go to Organization settings, then Skills, and enable it there for the whole workspace.

Installing your skill

1. Go to Customize, then Skills.
2. Click the “+” button.
3. Choose “+ Create skill” if you’re building one from scratch inside Claude, or choose “Upload a skill” if you already have a skill folder ready to go.
4. If uploading, select the ZIP file containing your skill folder.
5. Once it uploads, the skill appears in your Skills list, where you can toggle it on or off whenever you want.

A couple of things worth knowing. Skills you upload as an individual user are private to your account by default. They don’t show up for anyone else, even coworkers on the same plan, unless you’re on Team or Enterprise and you specifically share the skill with your organization.

The “missing required skill.md file” error

This is the most common snag people hit, and it’s almost always the same mistake. When you zip your files, you need to zip the folder itself, not just the loose files inside it. If you select SKILL.md and your other files individually and zip those, Claude can’t find the expected structure and throws a “missing required skill.md file” error, even though the file is technically in there.

The fix: put everything inside one folder, then right-click that folder and compress it. The ZIP should contain one folder, and that folder should contain SKILL.md plus anything else. Not the other way around.

Claude also offers skills in beta for Claude Code, and API users can access skills through the code execution tool, if you’re working in a more developer-facing setup. Most small business users won’t need either of those, but it’s worth knowing they exist if your team has someone technical involved.

How to install a skill in ChatGPT

ChatGPT uses the same underlying idea and the same SKILL.md format, which is genuinely good news since it means what you build isn’t locked to one tool. The installation path looks like this:

1. Open ChatGPT and select your profile icon.
2. Select Skills.
3. Choose “Create a new one” to build a skill inside ChatGPT, or select “Upload from your computer” to bring in a skill package you already built.
4. Once installed, your skill lives under the Installed section, where you can manage it.

If you’re on ChatGPT Work, there’s an additional option: admins and users can install shared skills from the Plugins directory. Select Plugins, browse or search for what you need, hit the plus button to add it, and then start a new task for the skill to actually take effect.

A necessary honesty check on availability

Here’s where you need to manage your own expectations a bit. Skills access in ChatGPT is genuinely fragmented right now, and it’s still rolling out. Personal Skills are generally available for Business, Enterprise, Healthcare, and Edu plans, and available on paid plans excluding Free and Go. Whether you can see the feature at all can depend on your specific plan, whether you’re working in regular Chat versus ChatGPT Work versus Codex, what device you’re on, your region, and whether an admin at your organization has actually turned Skills on.

Rather than assume you have access based on something you read, the practical move is to check your own account. Look for Skills under your profile icon. If it’s not there, it’s worth asking whoever manages your organization’s ChatGPT plan whether it’s been enabled.

A full walkthrough: building a “Client Monthly Report” skill

Reading about skills only gets you so far. Here’s one built start to finish, using a scenario a lot of small marketing agencies will recognize: a recurring monthly report for a client.

The SKILL.md file

---
name: client-monthly-report
description: Use when the user asks to draft, write, or create a monthly client report, marketing report, or performance update for a specific client, including phrases like "draft the report for [client]" or "write this month's update."
---

# Client Monthly Report

## When to use this
Trigger this skill any time a user asks for a monthly report,
performance update, or recap for a named client. This applies
whether they say "report," "recap," "update," or "summary."

## Report structure
1. Executive summary (2-3 sentences, plain language, no jargon)
2. Traffic overview (sessions, top sources, month-over-month change)
3. Lead/conversion summary
4. Wins this month (2-3 specific items)
5. Recommendations for next month (2-3 items)

## Tone rules
- Write like you're briefing a business owner, not a marketer.
- Avoid acronyms unless defined on first use.
- Keep sentences short. No filler phrases like "as we can see."

## What to never include
- Never guess at numbers. If data isn't provided, leave a
  placeholder labeled [NEEDS DATA].
- Never use the word "leverage."
- Never pad the report with generic industry commentary.

Look closely at that description line: “Use when the user asks to draft, write, or create a monthly client report, marketing report, or performance update for a specific client, including phrases like ‘draft the report for [client]’ or ‘write this month’s update.'” That’s a good description because it’s written in the words a person would actually type, not in job-title language. It doesn’t say “for marketing professionals producing client deliverables.” It predicts the actual sentence you’ll type into the chat box, which is exactly what the AI is trying to match against.

The folder structure

client-monthly-report/
├── SKILL.md
└── sample-report.md

The sample-report.md file is just a real, filled-in example of a finished report, following the structure above, with realistic (but fictional) numbers. Having an actual example alongside the instructions gives the AI something concrete to pattern-match against, rather than just abstract rules.

Zipping and uploading

Right-click the client-monthly-report folder and choose to compress it. You should end up with client-monthly-report.zip, and inside that ZIP should be the one folder containing both files. Upload that ZIP through Claude’s “Upload a skill” option, or ChatGPT’s “Upload from your computer” option, depending on which tool you’re using.

The payoff

Now, in a brand new chat, with no additional context, you type: “draft the July report for Northgate Dental.” The skill’s description matches, it loads automatically, and you get back a fully structured report following your exact section order, your tone rules, and your formatting, with placeholders anywhere real data is missing. You didn’t re-explain a single thing. That’s the entire point.

Tips for skills that actually fire

Building a skill and having it actually trigger reliably are two different skills (no pun intended). A few things that make the difference:

Write the description in trigger language, not job-title language. Describe the situations and phrases a real person would type, not an abstract category of work.

One skill, one job. Resist the urge to build one giant “marketing skill” that tries to cover reports, proposals, and social posts all at once. Split them up. A focused skill fires more reliably than a bloated one.

Keep SKILL.md short. Put long reference material, pricing sheets, or style guides in separate files inside the folder rather than cramming everything into the main file. The AI can pull those in as needed.

Test with a natural request. Don’t test a skill by saying “use my client report skill.” Type the request the way you actually would in real life: “draft the July report for Northgate Dental.” That’s the real test of whether your description is doing its job.

If it never fires, the description is almost always the problem. Before you rewrite the whole skill, go back and sharpen that one line first.

Treat it as living documentation. Update it as you learn what works. A skill isn’t a one-time setup task, it’s a process document that gets better the more you use it.

The portability point

Here’s the part that matters most for a small business thinking about this seriously rather than just poking at a new feature. Claude and ChatGPT both converged on the same basic convention: a SKILL.md file with a YAML header, optionally packaged inside a folder with supporting files. That means a skill you build isn’t locked to one vendor. If you write your client report skill for Claude today, that same folder structure works with only minor adjustments in ChatGPT.

That’s worth pausing on, especially if you’ve read our piece on the one AI question every small business should be asking, which is essentially: if this tool disappeared tomorrow, what would break? A skill file is one of the few AI artifacts where the answer is “not much.” Your documented process lives in a plain text file on your computer. You own it. It doesn’t evaporate if a vendor changes its pricing or shuts down a feature.

Start with one

You don’t need to build ten skills this week. Pick the single most repetitive thing you type into Claude or ChatGPT, the prompt you’ve pasted so many times you could recite it, and turn that one into a skill first. Test it. Refine the description. Then move to the next one.

If you’d rather have someone build your actual business processes into your AI tools properly, instead of assembling them piecemeal on evenings and weekends, that’s the kind of work we do at Lemon Head Design. We’ve been building websites and digital systems for small businesses, credit unions, and home service contractors since 2007, and increasingly that means helping clients turn their real operating procedures into tools their whole team can actually use. Reach out if you want help doing this right the first time.

FAQ

Do I need to know how to code to build a skill?

No. A skill file is plain written text with a small formatted header at the top. If you can write clear instructions and save a text file, you can build one. No programming knowledge required.

Will my skill work on both Claude and ChatGPT?

Largely, yes. Both platforms use the same SKILL.md convention, so a skill built for one generally transfers to the other with little to no modification, though it’s worth double-checking behavior after uploading since each platform’s interface differs slightly.

Is my skill visible to other people?

By default, no. Skills you upload as an individual user are private to your account. If you’re on a Team or Enterprise plan, you can choose to share a skill with your organization, but that’s an opt-in step, not the default.

What’s the difference between a skill and a custom GPT or persona?

A custom GPT (or a similarly branded assistant persona) is a standalone character built inside one specific platform, with its own name, personality, and setup. A skill is more modest and more portable: it’s an instruction set, optionally with supporting files, that any conversation on that platform can load when relevant. Skills aren’t personas, and they’re not locked to a single interface the way a custom GPT typically is.