Claude Cowork: build your first skill
Turn a repeatable task into clear instructions. Build a skill draft and test when it should apply.
Teach a repeatable method.
Build a workshop-brief recipe, inspect the draft, and test where it belongs.
A precise trigger
Describe the input and the result. Keep unrelated requests outside the scope.
65 / 200 characters in this draftResolution-independent illustration based on Anthropic’s reference. Labels and layout may differ in your app.
Reference: Anthropic ↗A method you can explain.
Keep your draft, test cases, and one revision together. Review the starter before uploading it anywhere.
Prefer written instructions? Open the complete class notes
What you’ll be able to do
- Recognize a repeatable task that benefits from a skill.
- Write a clear trigger, steps, and output requirements.
- Test a skill with both matching and unrelated requests.
Your practice: design a workshop-brief skill
Use the visual builder above to define a reusable method for turning workshop notes into a brief. The builder is a teaching simulation, not Claude’s live skill editor. Nothing is installed or enabled in your account. You will leave with a draft you can review and test.
A skill is a reusable set of instructions and supporting material for a task. A connector supplies access to another system. A skill does not grant that access just because it mentions a service. A plugin may package skills and other capabilities together.
Choose a repeatable job
The best starting point is a task you can describe and review consistently. “Help with my work” is too broad for this exercise. “Turn community workshop planning notes into confirmed facts, unanswered questions, and assigned actions” describes one useful method.
Write a trigger that identifies when the method applies. Include the kind of input and the desired result. A trigger should also be narrow enough to leave unrelated questions alone. The interactive exercise includes a matching request and an unrelated one so you can practice this distinction.
Define the output
Our workshop brief has three sections: confirmed facts, questions, and actions. Each section has a purpose. Confirmed facts preserve the source. Questions expose gaps. Actions identify work to do without inventing an owner or deadline.
The central rule is that missing information stays missing until someone confirms it. A skill should not turn a tentative idea into an established arrangement. In the exercise, select the output sections and this rule, then inspect the draft instructions.
A simple skill file
Anthropic documents a skill folder containing a skill.md file with name and description metadata, followed by instructions. The description explains what the skill does and when it should apply. The downloadable starter ZIP in this lesson contains an original UAI example with no executable scripts.
Treat that example as a draft to inspect. You do not need to upload it to finish this class. If you later use it in Claude, consult the current official creation and upload guides linked below. Review the contents before enabling any skill.
Find skills in Claude
The documented entry point is Customize > Skills. Capabilities, account settings, and organization policy can affect whether a skill is available. The official guide describes enabling individual skills and uploading custom skill packages. Labels can vary as the interface changes.
The screenshot in this class comes from Anthropic’s Help Center and shows an example of the add-skill menu. Use it as a reference rather than assuming your current app has exactly the same menu.
Write a test request
Use the workshop-brief method on these fictional notes.
Saturday library workshop. Free. Laptop optional.
Maya facilitates. Jordan checks the room.
Time, capacity, and registration are not confirmed.
Return confirmed facts, questions, and actions.
Do not invent owners, dates, promises, or deadlines.Test beyond one polished output
Run a matching request and compare the output with your source notes. Then try an unrelated request, such as explaining a word. The method should not force every response into a workshop brief. Finally, test incomplete notes and contradictory notes. Decide what a careful result should flag.
The simulator’s preview demonstrates these expectations; it is not a live model evaluation. When testing your real skill, record the actual outputs and corrections rather than relying on the preview.
Your finished artifact
Your finished artifact: A reviewed skill draft, two test requests, their expected behavior, and a worksheet recording one revision. If you test in Claude, attach the actual results to your record.
A common mistake
Making the description so broad that a skill applies to almost everything. Improve the trigger and examples before adding more instructions. Change one part at a time so you can see which revision helped.
Official guides
Checked October 22, 2026. Use skills in Claude and create custom skills. Independent UAI teaching material; no certification is awarded.
Check your understanding.
These practice questions give you feedback. They are not graded or used to award a certificate. Try again as often as you like.
Now make it your own.
Complete the hands-on assignment above, review your finished artifact, and use “Mark complete” when you’re ready to move on.