The short answer
Leave what you want today in the prompt: the facts, the files, the goal. Move how it should be done into a Skill: steps, format, examples, checks. Your prompts get shorter and your results stop drifting.
Skills vs prompts side by side
A prompt lives in one chat and carries this run’s details. A Skill lives in a folder and carries the method, so every run and every teammate starts from the same place.
The method
- Reused across runs and teammates
- Versioned like any other folder
- Can carry scripts, templates, and examples
- Loads only when it’s relevant
This run
- Written once, for one task
- Lives in a single chat
- Carries your facts and files
- Gone when the session ends
Here’s how the split looks for three common kinds of output:
| Output | In the prompt | In the Skill |
|---|---|---|
| Slides | “Launch deck for our note app, 10 slides” | Story arc, words-per-slide limit, final checks |
| Web page | Your copy, logo, and pricing | Layout system, sections, checks at phone width |
| Video | Script and length | Scene timing, caption style, export settings |
The paste test: is it time for a Skill?
Not every prompt deserves a Skill. If any of these sounds familiar, though, the prompt has grown into a method and belongs in a folder.
Make it a Skill if…
- You’ve pasted the same instructions three or more times.
- A teammate has asked you for “that prompt”.
- Results drift between runs even though the prompt hasn’t changed.
- The prompt only works with a template file or script sitting next to it.
How to turn a prompt into a Skill
- 1
Split the prompt in two. The reusable method goes into the Skill; this run’s facts stay out.
- 2
Save the method as
SKILL.mdinside a folder with the same name as the Skill.launch-deck/SKILL.md --- name: launch-deck description: Build a 10-slide product launch deck from a brief. Use when asked for a launch or announcement deck. --- 1. Ask for the product, audience, and three proof points. 2. One idea per slide, at most 25 words each. 3. Check every number against the brief. - 3
Move the folder into your agent’s Skills folder, then call it by name along with today’s facts.
You should see: the agent asking step 1’s questions before it builds anything.
You don’t have to start from a blank file: $skill-creator drafts one in Codex, and /create-skill does it in Cursor. Claude Code’s docs suggest keeping SKILL.md under 500 lines and moving detail into separate files.
Common questions
Is a Skill just a saved prompt?
Partly. It also has a description the agent reads to decide when to load it, and it can bundle scripts and reference files. In Claude Code, command files saved in .claude/commands/ still work, and a Skill with the same name takes over.
Do I still need prompts if I use Skills?
Yes. The prompt brings this run’s goal and facts; the Skill brings the method.
Sources
Did this guide help?