AI agent skills: turn workflows into systems

Everything in this course so far you have met exactly once. This last lesson is about what happens the second time.

Think back to the page you built in first build. Getting there was a sequence whether you noticed it or not: hand over the material, let the agent draft, review against your standards, ask for fixes, check the result. Build a second page next month and you will walk those same steps in the same order. That repeated walk is a workflow. Everything since, the context that loads itself, the .env shape, the checkpoints, was one step of that walk getting stronger.

A workflow you run by hand each time is useful. A workflow you save, so it runs on one instruction, is a system. Skills are how you save one, and they are where all of this goes after the course ends.

Skills: save a workflow once

When you catch yourself walking the agent through the same steps for the third or fourth time, that sequence is ready to become a skill. A skill is a small folder holding one instructions file, a SKILL.md, with a short name and a description at the top. Claude Code keeps skills in .claude/skills, Codex in .agents/skills. You describe the workflow once, and the description says when to use it. The next time a task matches, the agent reaches for the skill on its own. You can still call it by name.

Say every week you pull a few numbers, compare them to last week, and write a short summary. The first time, you spell out every step. Once it is a skill, asking for the weekly numbers is enough, and the routine runs the way you defined it. The setup cost is paid once, and the payoff repeats without you having to remember the routine exists.

What a skill actually says

A skill is instructions in plain English, the same words you would have typed into the chat, written down once. That weekly routine, saved, looks about like this:

---
name: weekly-report
description: Pull this week's traffic and signup numbers, compare
  them to last week, and write the Monday summary. Use when I ask
  for the weekly numbers.
---

1. Pull sessions and signups for last week from the analytics
   export in the project.
2. Compare each number to the week before, as a percentage.
3. Write the summary in the context file's voice: three sentences,
   numbers first, no filler.
4. Flag anything that moved more than 20% either way.

Known mistakes: do not round the signup count, and do not call
a flat week "momentum".

The top part is the trigger, the description that tells the agent when to reach for it. The rest is your briefing, frozen. Being a writer is an advantage here: a skill is only as good as its instructions.

Once a skill exists, asking the agent to show you what it says, and to walk you through what each part does, is enough. And because it is just instructions, changing one is a sentence, not a settings screen. From now on include the top three traffic sources, and stop reporting the percentage when the change is under 2%: say that, and the file is edited.

This is also how you fix a skill that keeps making the same mistake. Edit the instruction it ignored, or add the mistake to the file the way mine carry a known-mistakes note, and the fix sticks for every run after.

Run a saved workflow before you write one

Tempo, the practice company from the context lesson, carries a few saved workflows of its own, and the biggest one turns those 298 reviews into a full positioning brief.

Assignment: run Tempo’s saved positioning-brief workflow

Paste this in the project that holds your tempo folder. The agent reads the saved workflow, tells you what it is about to do, then does the whole job.

Read tempo/.claude/commands/positioning-brief.md and tell me, in two lines, what it will make you do. Then follow its instructions exactly and produce the brief it describes.

Done when a finished positioning brief comes back without you having given a single step.

Open the file it read while it works. Tempo keeps it as a command, the skill’s close cousin: called by name instead of firing on a description. Inside is the same anatomy as the weekly-report skill above: who to be, which files to read, the steps in order, what the output owes. Somebody walked this workflow by hand once, wrote the walk down, and now you replayed the whole thing with one instruction.

You just ran someone else’s saved workflow, and it worked beautifully, for someone else’s company.

Why a borrowed stack is noise

The common advice points the other way. Someone publishes a hundred marketing skills, you copy the folder, and the agent supposedly does marketing now. It does not hold up. A hundred skills from someone else is a hundred of someone else’s decisions, made for someone else’s product. You cannot tell which six matter to you, and the ones that do are buried under the ninety-four that do not.

Every line in a skill that repeats what the agent already knows is wasted space and weaker signal. General marketing knowledge is already in the model. What is worth writing down is the part it lacks: your steps, your standards, and the mistakes it keeps making on your own work. A borrowed stack is mostly the first kind, so its output comes back competent and aligned to nothing.

I make the fuller case for one file over a stack of skills in the Build Log.

Skills worth building

The skills worth building are usually the least glamorous ones: the production jobs you already repeat every week. Eight of mine, and what each one actually does:

Each of these is a process you own, not knowledge the agent lacks, and each carries the known-mistakes note no downloaded skill can hand you.

These are mine, wired to my sites, my brand, and my books; copying the exact files hands you decisions that break the moment your setup differs. Take the concept instead, and have the agent write your version.

Turning a routine of yours into a skill is one of the three moves in the Build your marketing brain tutorial.

Where this goes next

Everything so far has been you and one agent. You can also point several agents at one big job at the same time, each taking a piece, then pull their work back together. A full competitive teardown or a deep research run goes that way, one agent per competitor or per angle, covering more ground than one session usually can.

That is where the courses on the roadmap go next. Several agents at once is where this stops being a quicker way to do small tasks. It becomes a way to do work that would be too slow or fragmented by hand.

For now you have the foundation. Go make something you are proud of.