I've been writing code for 27 years. I picked up Claude Code the day it launched and haven't looked back. Last weekend I built a plugin that does something I've wanted for a while: brand positioning, messaging, voice, and visual identity using real expert frameworks, not "generate a brand strategy" vibes.
Here's what I shipped and what surprised me along the way.
The problem
Every time I've used AI for branding work, the output is the same. "Innovative solutions that empower teams." Stuff that could belong to literally any company on earth. The issue isn't that AI is bad at branding. It's that nobody tells it what good branding actually looks like.
So I packaged the actual methodologies into Claude Code skills:
Positioning: April Dunford's 5-component framework from Obviously Awesome
Messaging: Donald Miller's StoryBrand BrandScript
Voice: NN/g's brand voice dimensions + Aaker personality model
Visual identity: Chris Do's stylescape method
Each skill follows the real framework. The positioning skill works through Dunford's five components in order because the order matters. It's not "AI interprets the concept of positioning." It's "AI follows a specific methodology and checks its own output."
The anti-slop system
This is the part I'm most proud of. Every skill runs its output through quality checks before showing it to you:
Swap test: Could a competitor claim this? If your value proposition is "we provide innovative solutions that empower teams," that fails. Every software company on earth could say that. The skill flags it and helps you find what's actually specific to you.
Specificity test: Catches banned generic words (innovative, cutting-edge, leverage, empower, seamless, revolutionary, synergy) and replaces them with the concrete thing they were hiding.
Business-type test: Would this positioning work for both a lawn care company and a SaaS? If yes, it's too generic.
How the pieces connect
All 10 skills share state through a single brand-brief.md file with YAML frontmatter. Positioning fills in fields. Messaging reads those fields and fills in more. Voice reads messaging. Visual identity reads all of it.
Each component gets a confidence score: validated (backed by real customer data), researched (desk research), or assumed (educated guess). Downstream skills flag when they're building on assumptions. They don't stop you, they just make sure you know what's solid and what's still a guess.
What surprised me
The plugin system is more powerful than I expected. Claude Code plugins are just directories with structured markdown. No compiled code, no API calls, no build step. A skill is a markdown file with instructions that Claude follows. You can package entire methodologies into installable tools that anyone can clone and use.
Anti-slop checks actually work. I ran the toolkit on my own brand and it caught two of four tagline candidates as too generic. The discipline of checking your own output against real criteria matters more than I thought.
Adaptive interactivity solves the "blank page" problem. The positioning skill detects how much you know and picks one of three modes: research-assisted (you know nothing), guided (you know some things), or fast validation (you know your market). People at different stages get different experiences from the same tool.
Try it
git clone https://github.com/jgerton/brand-toolkit.git
claude --plugin-dir ./brand-toolkit
Run /brand-toolkit:start and it figures out where you are.
MIT licensed. Full getting started guide here.
I'm building more tools like this and sharing everything in my community. If you're into Claude Code beyond the basics, or you want to build your own plugins, come hang out.