If you’ve paying any sort of attention to the AI ecosystem, you’ve probably heard about MCPs a whole bunch (essentially AI’s version of APIs). But more recently, there’s also this new open standard by Anthropic around Agent Skills. In a nutshell, skills essentially train LLMs to know how to think in a specific way, giving a more consistent and repeatable outcome. I found this video very helpful as an explainer.

As stated in the account on GitHub, it says “Skills are folders of instructions, scripts, and resources that agents can discover and use to perform better at specific tasks. Write once, use everywhere.”  Gee, that sounds familiar. Quite a bit like design tokens, right? One set of design variables to drive experience across multiple types of platforms and experiences…

Where design tokens explain how something looks, Agent Skills tells LLMs how to behave when the system is executed via vibe. This goes far beyond code repositories or visual styles, and solves root level tribal knowledge issues for new team members or clients trying to adapt to a design system.

For example, you may decide that you have two buttons in your UI. Both want to be in the furthest right corner. How do we decide? There’s likely a tree of logic behind the decision, a set of stakeholders, or just an internal rule that most people in your UX group are aware of – but it’s not public. Write the answer as part of your Skill’s knowledge.

Skills can be the reasoning behind design decisions made public. It presents the ‘why’ in a practical way.

If your team is struggling to provide UX rules for how a system should be constructed, take a look at Agent Skills. Find ways that you can use them to refine your designs and your design process. Try vibing with those Skills. Better yet, record talking about those Skills (I use Superwhisper), convert them to plain text, and ask an LLM to make a Skill for you. Make those Skills public to the groups via Github or whatever and find a way to provide collective governance & contribution on that set of Skills.

While it’s still early-ish (Anthropic posted first in Dec ’25, and Open AI in Feb ’26), we are entering a new era of design systems, this might be your next best leg up on advancing yours in a hyper-scalable way.