Prompt or skill: which should you use?

Use a prompt for a single request. Consider a reusable skill when the same task needs a repeatable sequence, examples, or supporting files. Packaging can make a method easier to reuse, but it also creates something you need to maintain and test.

Start with the repeated part

Write the request you keep sending. Then identify what stays the same: the process, output format, evidence rules, or code checks. Keep case-specific inputs out of the reusable instructions. A skill that includes one customer’s private information is a poor reusable asset.

A fictional progression

The boundary is practical, not a rule that a prompt can contain only a few lines. Use the simplest form that supports the task reliably in your chosen host.

Extract the reusable instructions

Review these repeated prompts.
Identify the process and constraints that remain the same.
Separate them from inputs that change each time.
Propose a reusable instruction set and a small example.
Do not add tools or permissions that the task does not require.
Prompts: [redacted examples]
Host capabilities: [known tools and limits]

Test before packaging more

Try the instructions with a complete input, a missing input, and a conflicting input. If the method fails to ask for an important fact, fix that before adding extra references or scripts. More files do not repair an unclear task boundary.

When the process stabilizes, package the supporting materials where the host can find them. Keep a version and a change note. If your host cannot activate the package natively, you may be able to use the instructions manually, but label that fallback accurately. It is a different interaction and should not be described as a verified native installation.