The method repeats
You keep giving ChatGPT the same sequence of steps or evaluation criteria.
Skills
Turn a repeated method into a Skill
A Skill preserves how work should be done—even when the inputs change—so people do not need to rebuild the process from a long prompt.
WorkflowReusableShareable
Good Skill signals
Inputs change
The source material varies, but the definition of a good result stays stable.
Others need it
Several people should use the workflow without learning the original prompt.
Pick the right container
Preserve…
Use
Think of it as
A repeatable method
Skill
How to do the work
Background and history
Project
Where the work lives
A separate assistant
Custom GPT
Who people open
Treat external Skills like software
A Skill can include code. Review its source and permissions rather than treating it as a harmless block of prompt text.
From repeated correction to a stable method
The best Skill candidates have a consistent process and definition of quality, even when the source material changes.
Proposal review
A Skill can apply the same evaluation criteria, evidence requirements and response structure to every proposal rather than relying on a reviewer to remember the full checklist.
Interview synthesis
A Skill can turn different interview notes into the same themes, supporting quotes, risks and next-step structure for every research cycle.
Weekly updates
A Skill can define the sections, tone, metrics and review checks for a recurring business update while allowing the underlying information to change each week.
Skills
Install, share and govern Skills carefully
Skills can be selected automatically, combined with other Skills and shared across a workspace. That power makes clear scope, source review and administrative controls important.
Automatic useInstallationGovernance
How ChatGPT uses a Skill
It responds to relevance
Once installed, a Skill can be used when ChatGPT decides that its method matches your request. You do not always need to remember and invoke the Skill by name.
Skills can work together
ChatGPT can use more than one installed Skill in the same request when several methods are relevant. Each Skill should therefore have a clear trigger and a focused responsibility.
The package can contain more than text
A Skill may include step-by-step instructions, examples, supporting resources and executable code. More structured Skills can preserve the same sequence across repeated work.
Ways to create or install one
Route
Use it when
What to expect
Create with chat
You know the workflow but do not want to edit the package structure yourself.
Describe the method conversationally and refine the generated Skill.
Skills editor
You want direct control over instructions, resources and configuration.
Edit the Skill structure explicitly and test realistic requests.
Upload from a computer
You already have a compatible Skill package.
Review the source and code before installing it.
Install a shared Skill
A colleague or workspace has published an approved workflow.
Use the common method without rebuilding it personally.
Sharing and workspace controls
Share one common process
Eligible workspace users can share a Skill with selected people or groups, or publish it to a workspace library so a team can use the same approved method.
Administrators can set boundaries
Workspace administrators can control who may create, upload, install, share, publish or use Skills. These controls matter when Skills contain code or access business workflows.
Keep each Skill focused
If two workflows have different triggers or different definitions of good output, separate them. Focused Skills are easier for people to understand and easier for ChatGPT to select correctly.
Review Skills as reusable operational assets
A Skill can influence repeated work across many requests, so errors or unsafe code can also repeat at scale.
Test realistic inputs
Use examples that represent normal, difficult and incomplete requests. Confirm that the Skill activates at the right time and produces the expected structure without hiding uncertainty.
Inspect external packages
OpenAI may scan uploaded Skills, but that scan does not replace your own review. Examine instructions, bundled resources and code before trusting an external Skill.
Maintain versions deliberately
When the process changes, update the Skill rather than correcting every result manually. Communicate material changes to people who rely on the shared workflow.