Simon Says – Musings of a Technology CTO

As AI becomes more embedded in day-to-day work, the real advantage is no longer just asking better questions. It is teaching the system to handle repeatable work in a more consistent way. That is where Skills come in.

A Skill is a reusable workflow that teaches ChatGPT how to do a specific task more reliably. A skill can include instructions, examples, and even code so that the same job gets done with more consistency each time. OpenAI describes skills as reusable, shareable workflows that can be installed and then used automatically when helpful in a conversation.

That makes skills different from a GPT.

A GPT is a custom version of ChatGPT that can be tailored with instructions, knowledge, and capabilities for a broader use case or persona. In practical terms, I think of a GPT as the overall assistant, while a skill is a focused operating procedure the assistant can invoke when the situation calls for it. GPTs shape the experience at a higher level; skills sharpen execution for a specific task. OpenAI’s documentation describes GPTs as custom versions of ChatGPT, while skills are reusable workflows that can also be used inside ChatGPT.

That distinction matters. If you want a broad assistant for a function, team, or domain, build a GPT. If you want a repeatable way to perform one concrete job, such as summarizing customer interviews, writing board updates, reviewing contracts, or rewriting emails in a specific voice, build a skill. That is usually the cleaner and more scalable pattern. This is also consistent with OpenAI’s positioning: GPTs combine instructions, knowledge, and capabilities for a custom assistant, while skills package a repeatable workflow that can be shared and reused.

How to create a skill

OpenAI currently supports creating skills directly in ChatGPT conversations, through the Skills page, or in the Skills editor. You can also upload a skill from your computer. Skills are available in beta for ChatGPT Business, Enterprise, Edu, Teachers, and Healthcare plans, and OpenAI notes that they do not yet sync across products even though they follow the Agent Skills open standard.

The simplest path is the conversational one. You can tell ChatGPT what you want the skill to do, and it will guide you through the build. The best skills usually start with a few basics:

• the input you will give it
• the output you want back
• any tools, files, or connectors it should use
• the tone, structure, or constraints it should consistently follow

This is where teams often get the most value. A good skill does not try to be everything. It captures one repeatable workflow clearly enough that the model can execute it the same way every time.

How to use a skill in a chat session

Once a skill is created and installed, ChatGPT can automatically use it when it is relevant. OpenAI explicitly says ChatGPT can use one or multiple skills when helpful in a conversation. That means you do not always need to remember the exact mechanics each time. You can simply ask for the outcome you want, and if the skill matches the task, ChatGPT can apply it.

In practice, I have found there are two useful ways to work with skills.

The first is implicit use: just ask for the task. For example, “Rewrite this note into a clear executive email,” or “Summarize these interview notes into product insights.” If the skill is well defined, ChatGPT can detect the fit and will automatically use it – you don’t need to explicitly ask!

The second is explicit use: reference the skill directly when you want to be certain you are invoking the right workflow. That is especially helpful when you have multiple writing or analysis skills that could apply. Simply type / and the available skill list will be shown

A practical example: an executive email voice skill

One of the best starter examples is a content rewriting skill tuned to your preferred executive voice – and YES I used it to help me draft this very blog.

This kind of skill is valuable because content, like email, is deceptively repetitive. The job is rarely just editing words. It is clarifying the ask, tightening the message, setting the right tone, and preserving authority without sounding stiff or theatrical. That is exactly the kind of structured, repeatable judgment that a skill can help standardize.

If I were creating that skill, I would define it around a few clear rules:

• rewrite rough drafts into a calm, credible executive tone
• keep the language direct, collaborative, and business-oriented
• make asks, ownership, and timing explicit
• remove fluff, clichés, and over-formatting
• adapt slightly depending on whether the email is internal or external

Then, in a live chat, the usage becomes simple. You paste in a rough draft and ask ChatGPT to rewrite it using your executive voice skill. The result is not just cleaner prose. It is more consistent communication.

That is the broader point with skills. They are not just about automation. They are about encoding judgment, standards, and working style into something reusable. Done well, a skill turns individual habits into operational leverage.

And for leaders, that is the real opportunity. You are not just saving time. You are creating repeatable quality.

Here’s my executive-voice skill to get you started – note I didn’t bother defining what my style is – I asked AI to interview me and work it out. Sometimes, the less you tell AI up front, and the more you ask it to discover, the better the result.

Create a skill called "yourname-executivevoice" that rewrites emails in my executive voice. It should handle rough drafts, notes, and incomplete emails, and turn them into polished internal or external messages. Use narrative form with minimal formatting, but bold important asks when useful. 

If the draft is vague, ask clarifying questions instead of guessing before rewriting and giving one final best result. 

Run a 5-round interactive tone calibration exercise with sample inputs and 3 variants per round, capture my style, directness, use of fluff words, emoticons etc, and how diplomatic or executive I sound through this calibration. 

Capture a different style for internal and external emails. 

Once 5 rounds have been completed give me some example emails based on these findings for confirmation. 

Finally summarize the final voice and package the skill as skill.zip.

And if your company has a messaging document that defines your corporate brand, you add it to a chat and use this prompt to create a corporate branding skill:

Create a skill called "Corporate Voice" - it's purpose is to rewrite content and validate that content conforms to corporate marketing messaging style and content. I attached a file explaining the style. the overall rules though are: 

Do:
Speak in benefit-driven, concise language
Use data with reliable sources to back up bold claims
Center the customer and their outcomes
Stay consistent in terminology and tone

Don't:
Rely on superlatives with no substantiation
Use vague phrases like "next-gen" 
Push content live without brand review

Some useful links:

Skills according to OpenAI – https://developers.openai.com/cookbook/examples/skills_in_api
Pre-defined OpenAI Skills – https://github.com/openai/skills

(This post was created with support from an AI writing Skill)