ChatGPT Skills: A Practical Guide

ChatGPT skills are reusable instructions, with optional supporting files, for one specific task. ChatGPT picks a skill automatically when your request matches it, or you can choose one by typing @ and its name. You can build a skill with ChatGPT's skill creator or upload one, and on workplace plans you can share it with colleagues.

OpenAI describes skills as reusable instructions and supporting resources for a specific task or workflow (OpenAI: Skills and plugins). Think of a skill as a job card for a capable new starter: the steps, the format and the checks for one job, so the work comes out the same way whoever asks.

ChatGPT skills build on the open Agent Skills standard (OpenAI: Build skills). A skill is a folder with a SKILL.md file containing a name, a description and the instructions. It can also hold scripts, reference documents and templates.

Plugins are a separate thing. A plugin is an installable package that combines skills with connections to services such as GitHub, Google Drive or Slack (OpenAI: Skills and plugins). The skill describes the work; the plugin is one way of delivering it with the connections it needs.

Using a skill. ChatGPT reads each skill's description and chooses a relevant skill when your request matches. To pick one yourself, type @ followed by the skill's name (OpenAI: Build skills).

Creating a skill. Ask ChatGPT's built-in skill creator, @skill-creator, to help. It works through the skill's purpose, when it should trigger and whether it needs scripts. You can also write the folder and SKILL.md file yourself.

Adding a skill you already have. OpenAI's help centre describes uploading a skill from your computer through the Skills area. ChatGPT scans uploaded skills before they become available (OpenAI Help: Skills in ChatGPT).

Plans and admin settings. The same help article lists skills for workplace plans, including Business, Enterprise, Healthcare and Edu. Workspace admins decide who can create skills, upload them, share them and publish them to the whole workspace. If you can't see skills, check your plan and ask your admin.

OpenAI's own guidance matches what works elsewhere: keep each skill focused on one job, and write a description with clear scope and the words people use to ask for the task.

Tidewater Engineering, a fictional firm, sends clients a weekly project update. Everyone on the team uses ChatGPT, but each update looks different and some state proposed decisions as agreed.

A first skill for this job:

• Description: Turns this week's project notes into a client update with progress, decisions, risks and next actions. Use for weekly updates, status reports and client progress notes.
• Inputs: this week's notes, last week's update and who the update is for.
• Decisions: mark an action with no owner as "owner needed". Only call a decision agreed if the notes say so. Leave out risks that are only implied, and list them as questions instead.
• Output: four short sections, under 400 words.
• Check before finishing: every date, decision and owner appears in the notes.

The skill holds the method. It doesn't give ChatGPT access to the project notes. If the notes live in SharePoint or Google Drive, ChatGPT needs a working connection to that storage, and you should confirm it can open a test document. Better wording in the skill won't fix missing access. Skills vs MCP explains why.

• Choose one job your team repeats weekly, with a result you can check.
• In ChatGPT, type @skill-creator and describe the job, its inputs and the output you want.
• Answer its questions, then read the draft. Add the decisions for missing information if they're not there.
• Tighten the description so it names the task and the words people actually use.
• Save the skill, then run it by typing @ and its name.
• Run the three tests below before you share it.
• Share it with one colleague first and ask them to try their own real example.

• A normal request. Use a full set of notes. Judge the update against the notes, not its tone. A polished update with an invented commitment is still wrong.
• Incomplete input. Remove an owner. The update should flag the gap, not invent a name.
• An unrelated request. Ask for a meeting agenda. The update skill shouldn't take over. If it does, narrow the description.

Keep the three examples and run them again after each edit.

ChatGPT shares skills well inside a ChatGPT workspace. Most teams also use other AI tools, and each one has its own copy to maintain. A skill tells your AI what to do. Your knowledge base shows it how, in detail. Your company context tells it who it's for. The Contexta harness brings them together, in every AI.

Contexta holds one set of skills and your company context, and delivers both to ChatGPT, Claude and Cursor. Each tool connects once. Company context covers your terminology, naming and formats, set for the company and adjusted for each team, so the update skill doesn't need to explain who Tidewater is. Your knowledge base stays in your own SharePoint or Google Drive. Company, team and personal spaces route each piece of work to the right home and rules.

When you publish a change, connected tools fetch the new version the next time they use the skill. Version history lets you compare and restore. A downloaded file, by contrast, is a separate copy. ChatGPT is tested with Contexta and connects without code; check Integrations for the setup and How It Works for the full picture.