Did Anthropic Just Solve Prompt Spaghetti with Claude Skills?

Claude just shipped something interesting: Agent Skills. They’re like “prompt plugins,” but done in a very dev-friendly way.

A skill is a tiny folder: instructions, examples, maybe a script. Claude only loads it when relevant, so you don’t have to shovel context every time. And since you can include real code, the output becomes predictable, not “vibes-based.”

I’ve been testing them for scaffolding projects, enforcing team conventions, generating boilerplate and cleaning data. It’s shockingly effective. You can learn more about my testing here: https://www.shuttle.dev/blog/2025/12/02/claude-skills-complete-guide

Feels like a missing primitive for AI-assisted dev work.

Anyone else playing with these yet?