Loading video player...
Claude Code skills 2.0 are here and can now be tested, benchmarked, and auto-tuned with the new Skill Creator 2.0 plugin. I ran evals on my own skill live - the results are amazing. * You can grab the plugin for Skill Creator 2.0 by simply typing /plugins in your IDE/CLI and go to "manage plugins". 1. Click marketplaces tab and add the official repo: GitHub: anthropics/claude-plugins-official 2. Search for skill-creator and click "install" Timestamps: 00:00 Skill Creator 2.0 just dropped 00:24 What changed in version 2 01:02 Evals, benchmarks and blind A/B testing 02:24 Two types of Claude Code skills 03:40 Live eval: testing my meeting prep skill 05:04 Watching the 4-agent eval pipeline run 06:40 Reviewing the eval benchmark report 09:19 Skill vs no skill: the results 10:53 Front matter description optimization 12:24 Should-trigger and shouldn't-trigger testing 13:36 Optimization loop results 14:12 Final thoughts What to watch next: Build an AI OS - https://youtu.be/-GV1MRNB4hQ Claude Skills Deep Dive: https://youtu.be/7s9Fnorg3eI Claude + NotebookLM: https://youtu.be/Fj6iP_XWct8 -- Work with me: š Learn to build anything ā https://tinyurl.com/aitransformationacademy š¤ Ready to transform your business? Let's talk: https://bit.ly/3TinLo5 š” Add me on LinkedIn - https://www.linkedin.com/in/mansel-scheffel/ [ABOUT ME] If you're new here ā I'm Mansel. In 2010, I left South Africa with nothing but a laptop and a digital piano. I taught myself IT, broke into cybersecurity in London, and built two consulting businesses - one in cyber, one in cloud - scaling solo to $52K+/month. Today, I run atomicOps, an AI consultancy based in New York where we help mid-market companies turn AI chaos into scalable systems as a transformation partner. I also help aspiring AI builders deploy their own AI Workflows, Apps or AI Operating System in 30 days in my community