Okay, Anthropic released an article earlier today called Improving Frontend Design through Skills, and we'll be going over it and how you can use it within Claude Code to come up with better designs for your websites. So firstly, I'll go through some of the main points of the article and then we'll start applying it. So they do acknowledge a problem that when you ask an LLM to build a landing page without much guidance, it will almost always conform to Inter fonts, purple gradients and white backgrounds, and minimal animations. And that's because of the training data, because safe design choices usually dominate the training data that Anthropic have trained their models on, meaning that if you do not give it much direction, Claude will sample from these kind of design choices at work everywhere. Now, they do say that you can get Claude Code to avoid some of that behavior by using Inter and Roboto fonts, for example. And of course, if you then give a different constraints and aesthetic preferences, then it will come up with a different design. Now, of course, telling it to avoid Inter and Roboto fonts will only remove the font element of Claude Code generating a new design. The rest of it may still use purple backgrounds and so forth. And to have good frontend design, you need to have typography principles, color theory, animation patterns, and background treatment. And you would also need to specify which defaults to avoid in each of these areas and which alternatives to choose from. But now of course the problem does become that if you do add this into your claude. md file, for example, right at the bottom, then that will be included in every single request that goes to Claude Code and your claude. md file will get pretty long. And then you've used a more context window as well. And of course we all know that when you have a smaller context window with Claude Code, it performs significantly better than when you have a mostly full context window. So they say the question then becomes, how do you provide Claude with domain specific guidance exactly when needed? Without the permanent context overhead for unrelated tasks. Because of course, when you're getting Claude Code to debug Python or analyze data, then of course you don't need any information about how the frontend design should work. And as I talked about three weeks ago in this video about Claude Code's new skills feature, any information is progressively added to a context window instead of all being added at once. And that basically means that you can give Claude Code specialized context on demand without adding any permanent overhead. And because Claude Code is able to autonomously identify and load relevant skills based on the task at hand, if you have a frontend design skill, for example, and you get it to make changes to a landing page, then it will know to load that particular skill to help it make those changes. So essentially the way that they describe skills is that they are prompts and contextual resources that activate on demand when this specific task arises. And the way that you can imagine this is that before skills, you had your context window, which was kind of like this. So about 200,000 tokens maximum and you had any design related guidance that would be in your claude. md file, like maybe taking 10% for a pretty big project. And then after skills, you can then split this up into smaller, much smaller design related skills. And then in your skill file, you can have a file for fonts, colors, animations, logos, which could tell Claude Code to get logos from this particular website, for example. And then you would have any themes that you're using throughout your project. So they have some examples over here that we can take inspiration from. So they have this thing that you can add to your skill, which is using interesting fonts. And before adding that to their skills, this is a like very cloud looking design that I came up with. And then after adding it, It made it much better by using more interesting fonts over here. So you can see that this is probably Inter that it used, whereas here it used more code looking fonts instead. So of course the fonts here do look much nicer, probably not in this like input box, but yeah. And they did something very similar when it comes to themes. So you would add this to your Claude skill file and it would be like always use an RPG theme. And this is a design that it would come up with. And of course you can see that this design is very different. It's not as AI looking as other designs. And the same principle here does apply for other dimensions of design because you can have a very similar like prompt that can be used for animations in particular. And they say that when this skill is activated, Claude's output improves across several types of frontend designs. So for example, this is before using the skill. It's a very standard AI generated layout, nothing special. And then this is after using the skill. So you can see it uses more gradients, different colors, different fonts. I probably would not use this font myself, but it can be good for inspiration. This is before and then this is after. I think honestly this looks so much nicer. This is also before and then this is also after. Now let's go for a small