From Prompts to Skills: The Shift That Makes AI Execute For You

There's a scene in The Matrix where Trinity needs to fly a helicopter. She's never flown one. She calls Tank, he uploads a B-212 pilot program directly into her mind, and she opens her eyes: "Let's go." Instant expertise. Immediately applied.

That's what a skill does for AI. Not a vague prompt. Not a conversation. A complete package of expertise, loaded on demand, executed with precision, consistent every time.

I wanted Nuro, my AI co-creator to write a weekly AI briefing. Not help me write it. Write it. Curated stories, tool picks, the one thing nobody's talking about yet. A Sunday companion to the Thursday dispatch I write myself.

But to make that happen, I had to answer questions a few questions. What should it cover? Where does the data come from? What voice should it use — and how is that voice different from mine? What structure? What makes a good edition vs. a forgettable one?

Defining all of that for Nuro is what created the skill. The newsletter and the skill were born at the same time — because you can't hand someone a job without describing the process. And that description, it turns out, is the whole thing.

There's a lot of jargon flying around right now. Agents, tools, MCPs, orchestration. If you're not an engineer, it's easy to feel like this stuff isn't for you.

So let me strip it down.

That's it. It's not code. It's not a product. It's a set of instructions — the same kind you'd write for a sharp new hire who needs to do your job while you're on vacation. What to do first, where to find things, what the output should look like, what "good" means.

The difference between a skill and a prompt is scope and persistence. A prompt is something you type once. A skill is something you build once and invoke whenever you need it. Prompts are conversations. Skills are capabilities.

And unlike system instructions — which load into every session whether they're relevant or not — skills only load when the task matches. Your AI sees a short description of each skill, and pulls in the full instructions only when it needs them. That's what makes the system scale. You can have dozens of skills without bloating every conversation.

Vercel's explanation puts it well: "Skills solve problems that emerge when teams rely on agents for complex workflows. Prompts drift. Two people can ask for the same thing and get different results because they used different words. Without consistent instructions, agent behavior varies even when performing the same task."

Skills fix that. They make your process repeatable, portable, and invokable with a command.

This isn't just an Anthropic thing anymore. The timeline is worth knowing because it tells you where this is heading.

On October 16, 2025, Anthropic shipped skills as a feature in Claude Code — a way to package reusable instructions into folders that AI could discover and load on demand. Simple format: a SKILL.md file with instructions, optional scripts and references. Markdown, not code.

On December 18, 2025, Anthropic released it as an open standard. Not proprietary. Not locked to Claude. An open spec that any AI tool could adopt.

Six days later, OpenAI adopted it for Codex. Their announcement explicitly referenced Anthropic's open specification.

By the end of December, Microsoft had brought it into VS Code and GitHub Copilot. Cursor adopted it. So did Goose (from Block), Amp, OpenCode, and Google's Gemini CLI. Enterprise partners — Atlassian, Figma, Notion, Cloudflare, Stripe — started building skills for their platforms.

As of today, 16+ major AI tools support the same standard. A skill you write for Claude Code works in Codex, Cursor, Copilot, and Gemini CLI without modification. Write once, use everywhere.

Anthropic researcher Barry Zhang captured the shift at an industry conference: "We used to think agents in different domains will look very different. The agent underneath is actually more universal than we thought." The implication: instead of building specialized agents for every use case, you equip one agent with a library of skills. The agent stays general. The skills make it specific.

That's the trajectory. Skills went from an Anthropic feature to an industry primitive in about two months. The format is simple enough that non-technical people can write them — and that's the point.

The clearest way I've found to explain how skills fit into the bigger picture comes from Anthropic's skills guide:

You can have the best kitchen in the world. Without a recipe, you're improvising every meal. Some nights you get lucky. Most nights you don't.

A skill is how you stop improvising and start producing consistent results.

Let me walk you through a real example. Every Sunday, Nuro (my AI co-creator) publishes a briefing called the Sunday Edition — curated AI news, tool picks, and one underreported trend. Here's how it gets made:

The outcome: A weekly newsletter with 3 curated stories, 3 tool picks, 1 quiet signal, and a cover image. Written in Nuro's voice. 600-900 words. Published to subscribers.

The pipeline:

Each step has a specific data source, a specific job, and a specific output. The skill file describes all seven steps, including the voice rules, the structure template, the API details, and the quality checklist.

When I invoke /sunday-edition, the whole pipeline runs. One command, one session, one newsletter.

Here's what I want you to notice about that pipeline: none of those steps required me to write code. I described what happens at each stage in plain English. The skill file is a markdown document — the same format you'd use to write notes.

The real work wasn't technical. It was thinking clearly about my own process. That's the hard part, and it's also the part that only you can do. AI can't tell you what your newsletter should sound like. It can't decide which stories matter. It can't define what "quiet signal" means for your audience.

But once you've made those decisions and described them clearly, AI can execute them with remarkable consistency.

Here's the process I use for turning any repeatable outcome into a skill. No coding required.

Steps 1-3 are yours — only you know the outcome, the sources, and what good looks like. Steps 4-6 are where AI handles the implementation. The first version won't be perfect. That's fine. Skills are living documents — they get sharper every time you use them.

There's a mental shift buried in this process that matters more than the technical details.

When you build a skill, you stop being the person who does the work and start being the person who defines the work. You're not writing the newsletter — you're designing the system that writes it. You're not the chef — you're the head chef who writes the recipes, trains the team, and tastes the result.

As one writer put it: "Value migrates away from manual execution toward reflection, architecture, and governance. The premium is no longer simply doing the work, but orchestrating it."

That's not a future prediction. That's what I did last month with a newsletter.

The pattern works for anything repeatable. Weekly reports. Client briefs. Meeting summaries. Research digests. Onboarding docs. If you do it more than once and it follows roughly the same steps, it can be a skill.

You don't need to build seven skills next week. Pick one thing you do repeatedly. Something that takes 30 minutes to an hour. Something you could describe to a colleague.

Walk through the framework:

The first one takes the longest — maybe 30-45 minutes of thinking and conversation with AI. The second one takes half that. By the third, you'll start seeing your entire workflow as a collection of skills waiting to be built.

That's the orchestrator mindset. And once you have it, you don't lose it.

This dispatch was written using a skill — the same kind it teaches you to build. The Rift Dispatch skill defines the voice, the structure, the research process, and the quality bar. I describe the topic, have a conversation on the angle and the skill handles the rest.

Thanks for reading through.
I’d love to know how you felt about today’s newsletter. This will help me make the newsletter better.