One of my biggest AI productivity unlocks this year is the extensive use of agent skills.

In this post, I share my insights after building around 75 skills over 5 months. Coding and non-coding. LinkedIn posts, cover images, carousels, presentations, proposals, marketing briefs, SEO audits, project setup, deployment planning, best-practice reviews.

A skill is a markdown file with your instructions inside. Saved once, reused every time. You run it with a /command like /linkedin-post or /seo-audit. It’s an open standard. A plain text file that works with any AI coding tool - Claude Code, Codex, Gemini CLI. If you switch tools tomorrow, your skills come with you.

—

3 ways to create a skill:

• Chat first, then convert Work with AI the normal way. Give feedback until the output is good. Then tell it to create a skill from the conversation. Your best session becomes reusable.

• Ask AI to research and create Tell AI to research best practices for a task and write the skill. It does the research and writes the instructions for you.

• Write your own process Type out your step-by-step process in plain English. Your expertise, your rules, your quality bar. Ask AI to format it as a skill file.

—

Building a skill is not a one-time thing.

Think of it like training an apprentice. You show them the job once, they get maybe 60% right. You correct, they improve. After 10 rounds, 85%. After months, good enough that you only review, not rewrite. But you never stop reviewing.

Skills work the same way. My /linkedin-post skill has gone through 50+ revision cycles. It still needs my edits every time. But the starting point gets closer to my voice each round.

—

5 steps to improve skills over time:

• Use the skill on real work Not a test. Real tasks, real stakes. That’s how you know if the skill actually works.

• Edit the output Don’t accept the first version. Edit to your liking. Keep the original draft so you have both versions for comparison.

• Compare and update Ask AI to diff its output against your final edit. “List every change I made. Update the skill so next time it gets closer.” This is where compounding happens.

• Feed it winners When something performs well, feed it back as a reference. “This post got the most engagement. Analyze why. Update the skill.”

• Add what to avoid When you spot bad patterns, add them to the skill. Telling AI what NOT to do is just as important as telling it what to do.

—

Repeat for every piece of work. Each cycle is a small improvement. After 50+ cycles, those improvements compound into something a new person would take months to replicate.

The skill is never done. Just like an apprentice never stops learning.

I wrote a deeper guide on skills and Claude Code for non-techies: https://lnkd.in/gHfpq8k5

#AI #AgentSkills #ClaudeCode