Notice how often you tell ChatGPT the same thing? "Use our house style. Headline, then three bullets. British spelling. Sign off as the team." Every time, from scratch. Custom instructions (lesson 2.5) help with the general picture, but they're one block of text for everything. A skill is how you teach ChatGPT one specific job. Teach it once, and it does that job your way every time.
Which plans: skills are on Business, Enterprise, Edu, and Healthcare. Plugins are open more widely; on Plus you can install and create them, but without the skills part.
What a skill is
A skill is a small package for one task: the steps to follow, and whatever the job needs alongside them, such as an example of a good result, a template to fill, or a bit of code to run.
The best way to picture it: an onboarding note for a new hire, for a single job. Not "everything about the company," just "here is exactly how we turn meeting notes into a client update." Hand it over once and you never explain it again.
And a plugin?
You'll see plugins right next to skills, in the Plugin Directory in the sidebar. A plugin is the box a workflow ships in. It can hold one or more skills, one or more apps (connections to outside services like Slack or Google Drive), or both. A sales plugin might bundle a CRM connection with a "prepare for a client call" skill; a writing plugin might be skills only.
Some plugins are built by OpenAI and partners, some carry an OpenAI Verified badge, and some are made by people in your own workspace. A badge is a quality check, not a security sign-off: your company's own review still applies. Apps get their own lesson in 4.5; here we focus on the skill inside.
If you used custom GPTs: OpenAI is replacing them with plugins, and an existing GPT can be migrated into one.
You don't always have to call it
Here's the part people find surprising. Once a skill is installed, ChatGPT can reach for it on its own. You describe the task in plain words, ChatGPT compares your request with each skill's description, and uses the one that fits. You can also call it explicitly with an @mention, or from the + menu, when you want to be sure.
That puts the weight on the description. It has to say clearly what the skill does and when to use it. "Helps with documents" never fires at the right moment. "Turns rough meeting notes into our standard client update; use when I paste notes and ask for an update" fires exactly when it should. OpenAI's own advice: after creating or migrating a skill, test it both ways, by describing the task and by mentioning it.
Where skills earn their keep, and where they don't
Reach for a skill when a task is repeatable and has a right way to do it. Formatting to your brand. A weekly report that always takes the same shape. A review checklist you run over every contract. These are jobs where consistency is the point.
Skip it for one-offs and open exploration. Brainstorming, a single question, a first rough draft: a plain prompt is faster.
A quick way to place it. A prompt is for right now. Custom instructions are broad background for every chat. Project instructions cover one stream of work. A skill is a specific job, packaged, that ChatGPT can pick up in any chat when it fits, and that can carry its own examples and steps.
Making your own
You don't need to be technical. There are three ways in:
- Create with chat. Ask ChatGPT to make a skill. Eligible accounts come with a built-in skill-creator skill that asks you questions and drafts it for you.
- Create with the editor. Build and adjust it in the Skills editor, under Plugins, Skills.
- Upload a skill file from your computer, or install one a colleague shared.
Whichever way, the ingredients are the same: a clear name, a sharp description (what plus when), the steps in order, and an example or template if the job needs one.
Sharing is where skills beat a personal prompt file. From the skill's menu you can share it with specific colleagues or groups, or publish it to the whole workspace, if your admin allows it. A team that shares one "client update" skill writes client updates the same way.
One safety note
A skill can contain instructions and code, which is what makes it powerful, and what makes a bad one risky. ChatGPT scans uploaded skills: most become available straight away, some are marked Needs Review, and risky ones are Blocked. OpenAI is clear that the scan doesn't replace your own judgment. So keep the habit from lesson 1.4: use skills you built or got from a trusted source, and read anything else first.
Where this goes next
A skill is the how: a job done your way, on demand. Next comes the where. Projects (lesson 4.3) keep your files, instructions, and related chats together, so ongoing work has a home.
Try it
Write down three tasks you keep re-explaining to ChatGPT. Pick the one with the clearest "right way."
Sketch its skill in five lines: a name, a one-line description (what it does and when to use it), and three or four steps. If you're on Business, ask ChatGPT to "create a skill from this," let skill-creator ask its questions, install the result, then test it in a new chat by just describing the task.
(General exercise. Role and industry versions come once accounts are in.)