AI Guides › Claude Mastery

The Difference Between A Skill And A One-Off Prompt

By Nigel Guy · 2 min read

Something works well once, so you do it again the same way next week by half-remembering the prompt, getting it slightly wrong, and patching it live until it's close enough. This is fine for a genuinely one-off task. It's a slow leak when it's actually the fifth time you've reconstructed the same request from memory, each version a little different from the last, none of them quite as good as the one that worked.

The mechanism

1. Notice the repeat, don't plan for it in advance. You rarely know a task will recur the first time you do it. The signal is retrospective — you catch yourself typing something close to what you typed last week.
2. Write down what actually worked, not what you meant to write. The version worth keeping is the prompt that produced the output you kept, including whatever specific phrasing or constraints made it work — not a tidied-up version of your intent.
3. Separate the part that's fixed from the part that changes. A reusable prompt has a stable structure (format, tone, the questions to answer) and variable inputs (this week's numbers, this candidate's resume). Naming that split is most of the work of turning a prompt into a skill.
4. Test it on the next real instance, not a hypothetical one. A formalised prompt that only gets tested in the abstract tends to be too rigid or too vague. Run it for real and see where it breaks.
5. Only invest in a fuller skill — steps, checks, a defined structure — once the simple reusable prompt has proven itself a few times. Building an elaborate system around a task you've done twice is solving a problem you don't have yet.

What to skip

Skip formalising a task you've done once, no matter how well it went — one success tells you the prompt worked that time, not that it will work as a general pattern. And skip building an elaborate multi-step skill for something a two-line reusable prompt already handles fine; complexity should track the actual difficulty of the task, not your enthusiasm for the idea of having a system.

Guardrails

• A reusable prompt still needs to be checked each time you use it, particularly on the variable parts — formalising the request doesn't formalise your obligation to review the output.
• Requests that depend heavily on unstated context you're carrying in your head don't formalise well; if a prompt only works because you're feeding it detail you already know, write that context down as part of the prompt, don't assume it'll be re-supplied by memory.
• What counts as "worth formalising" is genuinely personal — a task that's rare for you but constant for someone else won't be worth the same investment, and that's fine.

All 751 AI guides · JulieMango plans from £17/mo