Thanks for your interest below is a simple guide + info to get you started on making your own skill the RIGHT way:
If you want to join My completely FREE community where we can chat, ask questions and learn from each other, here's the link:
https://www.skool.com/ai-automation-experts-5012/about?ref=a0e8de0b607240589239ff47d72650a8 Happy Building, Daniel
Claude Skills 2.0: A simple guide to building automations that actually work
Skills are saved recipes for Claude. You build one, and Claude follows it every time. No re-typing, no copy-pasting. It can search the web, connect to Notion, write files, even run code. You just trigger it and it goes.
Skills 2.0 adds built-in testing, which is the part that actually matters. Your first version will probably be wrong in some way you didn't expect. Now you can catch that automatically instead of finding out when it matters.
Getting started
On Claude desktop, the Skill Creator is already built in. In Claude Code, open a project folder, run /plugins in a terminal, search for "skill," install it, then restart. Takes two minutes.
What a skill is made of
Three parts: a .md file with step-by-step instructions, any reference files you want Claude to use (writing examples, your brand voice, whatever), and optional code or tool connections.
A real example: meeting notes This one's worth building. It reads a meeting transcript and writes a summary plus a list of who owes what. Open the Skill Creator (type skill, hit Tab), describe what you want: "Read my meeting transcript, write a short summary, list action items by person, save it as a markdown file." Claude will ask a couple of questions, then build and test the skill on its own. Once it's ready: drop your transcript in an "Inputs" folder, type your skill name, hit Tab, tag the file with @, run it. Done. The output saves locally, and you can ask Claude questions about old summaries later — it'll dig through them for you.
Testing: let Claude fix itself
The first version won't be perfect. That's fine. Skills 2.0 lets Claude spin up multiple test versions at once and compare them, instead of you running one test, checking it, tweaking, and repeating forever.
One rule: test one thing at a time. Trying to fix the writing style, speed, and formatting in one shot makes the results unreadable.
Example prompt: "Run 5 versions using [link]. Grade each one on: how closely it matches my writing style, whether it uses em-dashes, output length, and whether it pulls in stories from my background file." Review the report. Tell Claude what didn't work. It rewrites the skill.
A/B testing: which version actually wins
Once the skill works, you can run two versions against each other to see which is faster or better. Especially useful when a new Claude model drops and your old skill gets sluggish.
Example: "A/B test my YouTube-to-newsletter skill. Version A uses all 8 reference files. Version B skips the voice file. Test both on [link]." Keep whichever wins.