Loading video player...
Skills in Claude Code are powerful, but they don't learn from your corrections. Every new session, you're teaching Claude the same things again. In this video, I show you how to build a self-improving skill system. Using a custom "autoskill" I built, Claude automatically extracts learnings from your corrections and updates the relevant skills - so it never makes the same mistake twice. I demo this with shadcn's brand new project scaffolding (released December 2025) - something Claude doesn't know about yet. After one correction, the skill updates itself and gets it right every time after. š Autoskill on GitHub: https://github.com/AI-Unleashed/Claude-Skills/blob/main/autoskill/SKILL.md š Anthropic's frontend-design skill: https://github.com/anthropics/claude-code/blob/main/plugins/frontend-design/skills/frontend-design/SKILL.md š shadcn/create: https://ui.shadcn.com/create Chapters: 0:00 The problem with static skills 0:28 Demo: Claude misses the Lyra style 1:56 Why this happens (knowledge cutoff) 2:17 Correcting Claude once 3:05 How the autoskill works 4:24 Running autoskill & setting up hooks