Build Your Own Brand Kit.

How this fits in the OS.

What changes for you.

The artifact is the seven folders. The bigger thing is what shifts internally the first time you ask Claude for a deck and it comes back already on-brand. Three changes, in order of how soon you'll feel them:

• You stop redoing visual decisions on every artifact. The colors are picked. The type stack is picked. The voice rules are written down. The agent reads them once, every surface inherits.
• You start treating your brand as code. A new accent color is a one-line change in tokens/colors_and_type.css, not a hunt through fifteen documents. The brand stops being a folder of files and starts being a system you can update in one place.
• You earn a deliverable you can sell. The same scaffold ships your personal brand and a paid client engagement. Once you've built it once for yourself, you can build it again for a client in an afternoon.

This is the moment the brand stops being something you maintain and starts being something your fleet maintains for you.

The two reasons to do this.

Most students land on this card for one of two reasons. Either reason is the right reason. The pattern is identical.

• Your own brand becomes consistent across every agent-generated surface. Your portfolio site, your one-pagers, your slide decks, your social posts, your Morning Brief dashboard, all of them inherit the same colors, typography, and voice. You stop being the bottleneck for "make this on-brand."
• You can sell this to clients as a deliverable. Small businesses and solo operators have visual identity that lives in their head or in a Canva file. Once you've shipped your own, you can scaffold one for a client in an afternoon. The deliverable is a private GitHub repo, a brand spec, a token file, and a SKILL.md their team and their agents can both read going forward.

Either way, the scaffold is the same. The difference is whose voice gets written into the BRAND-SPEC.

The seven-folder shape.

Every good brand kit follows the same skeleton. The shape below is the standard one, and your repo has it too. The contents are yours, the structure is the reusable part.

The order matters. The BRAND-SPEC is written first because every other folder is downstream of the decisions in it. Tokens are second because the components and slides reference them. SKILL.md is third because it needs the spec and the tokens to point at. Assets and slides come last because they consume everything above.

The AI-native unlock.

The difference between a brand kit and an agent-readable brand kit is the SKILL.md. Without it, your repo is a static asset library, useful when a human opens it. With it, every agent in your fleet can read the brand context in one shot and render in-brand by default.

A good SKILL.md is short and names the file-read order explicitly. Keep yours to the four sections in Step 04 below. The shape is small. The leverage is large.

Before you start.

The build.

Seven steps. The BRAND-SPEC comes before tokens because the tokens are downstream of the visual decisions in the spec. The SKILL.md comes after both because it points at them. Each step has a duration and a recovery line.

The agent does most of the typing. Your job is the decisions. Voice rules, palette picks, the no-em-dash equivalent for your brand. Claude handles the file creation, the directory structure, and the boilerplate.

When it works.

You'll know the brand kit is live when three things hold without you doing anything:

1. You ask Claude for a slide deck, a one-pager, a social post, an email signature, anything. The first draft comes back on-brand without you naming the rules in the prompt.
2. You change a token value and the next render picks it up. Swap the primary accent in tokens/colors_and_type.css, ask for the same deck again, every accent in the new deck moved.
3. A new agent in your fleet (a sub-agent, a research agent, a different CLI session) inherits the brand without you re-explaining it. The SKILL.md is doing the work. You stopped being the bottleneck.

Run the loop on three different prompts. The third clean render is when the pattern is yours. Now apply it to a client. The afternoon you spend on a client brand kit pays for the half day you spent on yours.

If you get stuck.

Common failure modes, ranked by how often they bite the first time through:

• The SKILL.md grew past a thousand words and Claude stopped using it well. A manifest that long stops being a manifest. Trim it back to the four short sections. Push the depth into BRAND-SPEC.md. The SKILL is the index, not the encyclopedia.
• The brand voice rules contradict each other. "Sentence case for headings" plus a phrase example with title case in it. "No emoji in headlines" plus a hero example with a rocket in it. Claude will follow one or the other, not both. Walk through the spec one more pass and resolve every contradiction. A rule with no exception is worth ten rules with carve-outs.
• Tokens don't propagate to a downstream file. The file isn't linking tokens/colors_and_type.css at the top. Have Claude grep for hardcoded hex values across the repo and replace them with token references. Add a line to the SKILL.md non-negotiables banning hardcoded hex.
• The skill doesn't appear in Claude Code. Either the install location is wrong, the frontmatter is malformed, or the session hasn't been restarted since install. Walk through Step 05 in order.
• Two clients' brands leak into each other's decks. You installed both SKILLs globally. Move one to per-project at .claude/skills/ inside the client repo. Global skills inherit everywhere, per-project skills don't.
• The kit feels like a static asset library, not an agent-readable system. The SKILL.md isn't doing enough. Read the AI Build Lab's SKILL.md side-by-side with yours. Theirs is short, names the file-read order explicitly, and lists the non-negotiables. If yours is longer and vaguer, you wrote documentation where a manifest belongs.

Still stuck? Post a screenshot, the prompt you ran, and what Claude returned in the workshop Slack. Someone on the team has hit the same thing.

Where to go next.

The brand kit is a layer of the OS. The natural next moves apply the layer to surfaces you already shipped, or to new ones.