How I Use Claude Code to Make INSANE Instagram Carousels
A step-by-step demo of turning a Pinterest screenshot into a branded, reusable carousel system using a design system MD file, Higgsfield MCP, and a slash-command skill.
May 28thA 14-minute practical walkthrough on what Claude Code Skills actually are, how to install a team's full skill library in one command, and how to build a custom skill from scratch using a live design system demo.
A Claude Code Skill encodes your design system's repeatable patterns once and lets the AI reach for them automatically based on context, so your whole team stops re-teaching AI the same rules on every session.
Most design work has repeatable patterns — tokens, component conventions, handoff rules — and AI users typically re-paste this context on every session. A Claude Code Skill solves this: it is a skill.md file with a description specific enough that the AI auto-detects when to load it, so no manual invocation is needed. The presenter team publishes a free public repo of design-focused skills installable in one npx command. The video demos extracting design tokens from a live site and building a custom component-token-audit skill from scratch — both triggered by natural-language prompts with no explicit skill name invocation.
Sign in and you get 23 free chat messages on us — ask for the hook, quote a framework, find the exact transcript moment, generate a markdown action plan. Bring your own key when you want unlimited.
Create a free account →
Cold open states the core problem: teams keep re-teaching AI the same design patterns on every session.

Defines a skill vs. a .md file: auto-triggers from description, bundles skill.md plus templates and company context. Shows the design.md skill anatomy.

Explains why design systems are the ideal skill use case: component conventions, tokens, and handoff rules are all repeatable, high-stakes patterns.

Demos npx skills add command, spacebar-select menu, and the choice between project-level and global installation.

Walks through the public skill library: Figma Bridge, FigmaGen, Extract DS, AI Components, Caveman, Prompt Evaluator, Market Psychology. Shown via an 8-bit video-game UI.

Shows the Carrot open-source product, prompts naturally to extract its design system, and Extract DS auto-fires. Output is a JSON of colors, fonts, and spacing values.

Audits Carrot components to find hard-coded values, runs token extraction, then asks Claude to create a skill for this repeatable task. The skill-creation guidance skill writes the new skill.md automatically.

Summarizes the skill-vs-markdown distinction and encourages viewers to use skills in their design systems. Subscribe CTA.
The real cost of not using skills is not one bad output — it is every future output where the AI reverts to generic behavior because context was forgotten.
“So most of us keep teaching AI the same patterns over and over. A skill ends that.”
“A skill is something that your AI reaches for on its own.”
“We did not prompt it. We gave it the right context so it knew exactly which skill to pull.”
See every word as it's spoken — crank it to 2× and still catch all of it. The same dual-channel trick behind Amazon's Kindle + Audible.
Every time you start a new Claude Code session, you paste the same context: here are my tokens, here are my component rules, here is how handoff works. A skill ends that loop. You build the instruction set once, give it a description precise enough that the AI recognizes when it applies, and from that point forward the AI reaches for it on its own.
Every skill can carry up to three layers of files; the skill.md is required, templates and context are optional but powerful for design system work.
Markdown files require explicit invocation; skills auto-trigger from description matching. The difference is whether the AI reaches for the context or waits to be handed it.
Seven pre-built skills covering the full design-to-code loop, available free via the public agent-skills repo.
“We have a couple slots for workshops where I come in and I assess the situation, what's working, what's not working.”
Soft mid-video pitch framed as a collaborative team assessment rather than a sales pitch. Not interruptive — placed while a demo is running. Second CTA is standard subscribe at end.
00:00
00:16
00:28
00:38
00:49
01:00
01:11
01:21
01:32
01:43
01:54
02:05
02:16
02:30
02:38
02:49
03:00
03:11
03:22
03:38
03:43
03:59
04:00
04:16
04:27
04:38
04:49
05:00
05:11
05:22
05:27
05:44
05:55
06:05
06:16
06:27
06:38
06:49
07:00
07:11
07:22
07:33
07:44
07:58
08:01
08:17
08:28
08:38
08:49
09:00
09:11
09:22
09:33
09:44
09:55
10:06
10:20
10:30
10:39
10:50
11:00
11:11
11:22
11:33
11:44
11:55
12:06
12:17
12:33
12:39
12:50
13:01
13:12
13:22
13:33
13:48
13:55
14:06
14:17
14:28A step-by-step demo of turning a Pinterest screenshot into a branded, reusable carousel system using a design system MD file, Higgsfield MCP, and a slash-command skill.
May 28thCraig Hewitt reverse-engineers Corey Haynes' open-source marketing skills pack — showing what makes a skill structurally powerful and demoing email sequences and CRO audits live.
January 28thA 14-minute proof that the AI slop problem is a documentation problem, and that Fable only beats Opus when you let it make decisions from scratch.
June 11thA 25-minute zero-edit pipeline tutorial: one creator, one AI model, and a $2-per-video production stack built entirely inside Claude Code.
June 11thA 20-minute breakdown of the 10/80/10 system and loop engineering — the cost-efficient way to run the most expensive AI model on the market.
June 11thA 15-minute live build where Claude Code, a Higgs Field MCP, and four Vercel commands produce a SpaceX-inspired site with a rotating 3D Mars planet and scroll-linked zoom effects.
June 9th