📖 The style guide the agent reads by itself
New here? One skill (already introduced in Track 3) is a set of instructions saved in a file that the agent consults automatically before doing a particular type of work — like a procedure sheet an employee pulls from a drawer before starting a task that has an established "right way to do it" at the company. A design skill it’s that same mechanism, specialized in front end: it stores your color palette, typography, spacing, visual tone, and the principles that define what “beautiful” means to you—not to the entire internet.
Think of a real architecture studio: there’s a physical folder (or a PDF) called “visual identity guidelines”—exact colors in hexadecimal code, approved fonts, minimum margins, examples of what is and isn’t allowed. Every new intern reads it before their first project. Nobody re-explains every color in a meeting each time. The design skill is exactly that guide, except the “intern” is the agent, and it reads the guide on its own at the start of every front-end task— without you needing to type anything again.
💡 Main Concept
The design skill addresses exactly cause 1 from module 5.1 — the lack of reference that made the model default to the training data’s statistical average.
- •Without a skill, the agent fills the gap with what it sees most often—blue, white, centered.
- •With a skill: the agent fills the blank with what YOU defined—your reference, not everyone’s average.
⚖️ "Make a beautiful website" vs. a skill-guided request
"Build a beautiful website" is a vague request because "beautiful" has no fixed definition — everyone on the planet whose work was used to train that model has a different idea of "beautiful," and the model averages all of them, which is exactly the "AI feel" from the previous module. But a request backed by a design skill doesn’t need to say "beautiful" even once: the skill already defines in advance what "beautiful" means to you — what palette, typography, spacing, and tone to use (serious? fun? luxurious? handmade?). The agent doesn’t invent anything visual anymore: it applies the rules that are already there.
Caption: above, without any guide, the vague request falls straight to the statistical average. Below, the agent has a habit of opening the design skill BEFORE writing any front end — even a short request comes out with the right palette, typography, and tone because the decision was already made in the guide.
🔍 Inside
- "Beautiful" without a definition: a word that each training run interprets as the average of everything it has seen so far.
- "Beautiful" with a skill: a decision you make once and apply every time — it stops being the model’s opinion and becomes a project rule.
🧩 The anatomy of a design skill
Every skill (recalling Learning Path 3) has a description/trigger — the short sentence that tells the agent WHEN to consult that file, without needing to read the entire file to decide. In a design skill, the trigger might be something like “use before creating, redesigning, or reviewing any page, component, or interface.” That’s what makes the agent open the manual automatically, even when you only asked “create a contact form” without mentioning “design” or “skill” at all.
After the trigger come the style rules themselves: color palette (with exact hexadecimal codes, not vague "blue" or "green"), the typography (which fonts to use, at what size, for headings and body text—typography is just the technical term for “the choice and style of letters”), spacing (how much breathing room between elements, square or rounded corners), and the overall visual tone (serious and corporate? fun and colorful? minimalist and sophisticated?). Finally, a good skill includes examples — code snippets or concrete descriptions of "this is right" and "this is wrong" — because concrete examples are a better guide than abstract rules.
✓ What a good design skill provides
- ✓Clear trigger: when to invoke (before ANY front end work)
- ✓Palette with exact hex codes, not vague color names
- ✓Defined typography (heading font, body font, sizes)
- ✓Examples of “right” and “wrong” side by side
✗ What makes a weak skill
- ✗"Use pleasant, modern colors" (just as vague as the original vague request)
- ✗No trigger — the skill exists, but the agent never knows when to open it
- ✗Rules that change from project to project without ever being reviewed
💡 Practical Tip
A quick way to tell if your skill is good: read the color rule out loud. If you can recite the hex code from memory, it’s concrete enough. If the phrase is something like “colors that convey trust,” it’s still too vague—rewrite it with the exact hex codes.
🌐 Install globally once vs. rewrite every conversation
New here? "Global vs. project" It’s about WHERE the skill is stored and WHICH projects can see it (covered in depth in module 3.5, but central here again). A skill of the project lives inside that specific project's folder and only applies there. A skill global gets installed once, in a place on your computer that every Claude Code project can see automatically — it works for any folder you open, without needing to copy the file again for each new project.
The practical difference is huge. Without a global skill, every time you open a new project—the bakery’s website, the consulting dashboard, the course landing page—you explain everything from scratch: “use this color, this font, this spacing.” It’s tiring, it’s easy to forget a detail, and the result varies from project to project even when you wanted the SAME style across all of them. With the skill installed globally, you save the guide just once, and every new project—even six months from now, even one you can’t imagine today—starts with that same reference.
📊 Where each type lives and who can see it
- Project skill: inside the project folder—only someone who opens THAT project can see it. Good for a rule specific to ONE client or brand.
- Global skill: installed once on your account — EVERY project you open can see it. Good for “your default way of doing front-end,” which you reuse all the time.
✗ No global skill
- ✗You explain the color, font, and tone again in every new conversation
- ✗Easy to forget a detail — results vary from project to project
- ✗Each new project "starts" with no references at all, generic all over again
✓ With a global skill
- ✓You write the guide once, and every new project starts with it built in
- ✓Automatic consistency across different projects, with no extra effort
- ✓Works six months from now, in a project you can't even imagine today
You write the skill once
Colors, fonts, tone, examples — everything decided and written down at once, in the global location.
You open any project, today or a year from now
The folder doesn’t matter—the agent can see the same global skill without you reinstalling anything.
The agent checks first before coding the front end
Without you remembering to ask—it’s the habit built into the skill trigger, topic 5.
🎯 Invoke the skill before writing front-end code — the habit
The discipline that separates a useful skill from a forgotten one comes down to a single sentence: always invoke the design skill BEFORE writing any front end, never afterward. If you let the agent write the component first and “review it with the skill” afterward, it has already spent effort building on a generic guess—that means rework. The right habit is: a front-end request comes in → open the skill first → only then write the code using the right reference from the start.
Objective: test in your own environment whether a globally installed design skill changes the result of a simple component request — without you needing to mention color, font, or style even once in the request, because that should already come from the skill.
Before writing any HTML or CSS, check whether a design skill is installed
and use it as a style reference. Now create a product card for
with a title, price, and a buy button. Don't ask me about colors or fonts — decide based on the skill.
If you already have a design skill installed globally, the result should use its palette and typography, without a generic blue “Get Started.” If you don't have any skills installed yet, the agent will probably tell you it couldn't find any—which confirms the lesson: without a manual to consult, it falls back on guesswork. (Installing your own design skill is covered step by step in modules 5.6 and 5.7, along with the complete brand folder.)
✅ How to verify
- Check whether the agent mentioned at the start of its response that it consulted a design skill (or said it couldn't find any installed).
- Open the generated HTML: do the colors match a specific palette, or did it still come out in the default blue and white?
- Repeat the same request in a different project folder—if the skill is global, the result should look the same in both folders.
⚠️ Attention
Having the skill installed isn’t the same as the agent remembering to use it on its own every time. If the result comes out generic even with the skill installed, the problem is usually a weak trigger (the wording is too vague to prompt the skill to be read) — revise the skill description, not the request.
🗺️ What the skill can't solve on its own
The design skill addresses cause 1 from module 5.1 — lack of reference — but it doesn’t address cause 2: no one makes the agent actually LOOK at the result before handing it to you. A well-written skill greatly reduces the “AI smell” because the agent starts with good choices, but it doesn’t replace a critical eye on the final result — the CSS can be technically correct and still look misaligned, cramped, or low-contrast on a real screen. The next module addresses exactly this second half.
🎯 Main Concept
- •Design skill (5.2) = written reference, consulted before coding.
- •Screenshot loop (5.3) = the agent takes a screenshot, reviews the result, and makes corrections — after coding.
- •Together, the two complete the full cycle: the right reference BEFORE, a critical eye AFTER.
Quick check (optional): How is a “global” design skill different from rewriting style instructions in every conversation?