Skills and commands: save what works, reuse it in one word

What it is

Many people write a good prompt once, get a great result, then lose it. Saving what works fixes this.

AI assistants such as Claude, ChatGPT, Microsoft Copilot and Gemini offer several ways to save instructions. The names differ by product, but the ideas are the same:

• Saved prompts: a piece of text you keep and paste or select when needed.
• Custom instructions: standing preferences applied to every chat.
• Custom assistants or projects: a named assistant with its own instructions and files for one job.
• Commands: in some tools you type a short word, often starting with a slash (for example /summary), to run a saved instruction.
• Skills: a folder containing written instructions, a few examples and, optionally, small scripts (programs) that the assistant loads when the task matches.

Think of a recipe card: you do not reinvent the dish each time, and anyone in the kitchen can follow it.

Why it matters

• You get more consistent quality, not just on a good day.
• You save typing and thinking time on tasks you repeat.
• New colleagues can do a task well on day one by using the saved version.
• Improvements build up: fix the card once and everyone benefits.

How to do it

1. Pick a task you do at least weekly where the AI already gives you a good result.
2. Write the trigger: one sentence saying when to use it ("Use when I paste raw meeting notes").
3. List the steps the assistant should follow, in order.
4. Define the output format: headings, length, tone, and what to leave out.
5. Add checks: what the assistant must flag or refuse, and what you will check yourself.
6. Include one short example of good output, using made-up content.
7. Save it in whatever your approved tool offers (saved prompt, custom assistant, project or skill) with a clear name and a version number.
8. Share it with your team once it has worked well for you several times.

Example

Before: each week you type "summarise these notes" and get a different shape every time.

After: you save this instruction set as "Meeting notes v1".

Name: Meeting notes v1
Use when: I paste raw notes from an internal meeting.

Steps:
1. Read the notes in full.
2. List decisions made, one line each.
3. List actions as: owner, task, due date. If any is missing, write [missing].
4. List open questions.

Output format:
- Headings: Decisions, Actions, Open questions.
- Plain British English, no more than 250 words.

Checks:
- Do not invent names, dates or decisions that are not in the notes.
- If the notes mention customer or personal data, say so at the top.

Example of an action line:
- [Sam], send revised budget to finance, [Friday]

Now you paste the notes and run "Meeting notes v1". You still read the result before sending it, but it arrives in the same shape.

Common mistakes

• Saving an untested prompt. Use it on three different inputs first.
• One giant instruction set. Keep one task per skill; small ones are easier to fix.
• No version number. Update the name (v1, v2) and note what changed.
• Hiding the human check. Write what the reader must verify before use.
• Adding skills from unknown sources. Only add skills from approved, trusted sources; a skill with scripts can run code.
• Sharing before agreeing on data rules. Make sure a shared assistant does not contain confidential reference files that some colleagues should not see.

Try it this week

1. Find a prompt in your chat history that gave you a result you liked (5 minutes).
2. Rewrite it using the five parts above: trigger, steps, output format, checks, example (5 minutes).
3. Save it in your approved assistant with a name and "v1" (2 minutes).
4. Run it on one new, made-up input and note one improvement for v2 (3 minutes).

Going further

• Write skills in Markdown. Skill instructions are usually a Markdown file: plain text with headings for trigger, steps and output (see guide 06).
• A skill can include reference files (a style guide, a template) and small scripts, so the house format or a calculation is the same every time.
• In tools that load skills when relevant, the one-line description matters most: write it so the assistant can tell exactly when the skill applies.
• Test a changed version on the same three sample inputs before replacing the old one, and give each shared skill an owner.

Key words

• Saved prompt: a prompt you keep and reuse instead of retyping.
• Custom instructions: standing preferences an assistant applies to every conversation.
• Custom assistant: a named assistant set up with its own instructions and files for one kind of job.
• Slash command: a short word starting with "/" that runs a saved instruction.
• Skill: a folder of instructions, examples and optional scripts that an assistant loads when a task matches.

Examples and figures in this guide are illustrative. AI tools change often: check what your organisation's approved tools offer today. Use approved tools only, and a person checks and decides.

Prefer to print or share it? Download this guide as a PDF (250 KB), or the complete AI Field Guide (1.7 MB).