I live in the terminal. Like, really live there.

As a developer advocate, it’s where I spend most of my time — running commands, testing code, writing docs. Over the past few months, Claude Code became my primary interface for getting work done. Not just coding. Content creation, documentation, the constant stream of blog posts that come with DevRel.

That created a weird problem.

I got really good at using AI tools. But it was still a personal, one-off workflow. Every blog post started with me crafting the perfect prompt, tweaking the output, running linting checks, formatting everything manually. It worked. It didn’t scale. Our marketing team needed to produce more content, hit more keywords, maintain consistency… and they couldn’t all live in my head.

So I built a Claude Code skill to systematize it.

Codifying What I Was Already Doing

The blog skill started as a way to encode my own process. I’d open Claude Code, start a conversation about a blog topic, go back and forth on structure, check for banned words, run Vale linting, generate images, submit a PR. Efficient for me, completely opaque to anyone else.

Here’s what changed: I realized I could encode all those standards — brand voice, banned words, style requirements, SEO checklists — into a reusable workflow. Not documentation about how to write. An executable skill that enforces those standards automatically.

What the skill does:

• Research the topic – searches existing posts, docs, sources
• Create an outline – structured with SEO and LLM citation optimization
• Checkpoint 1 – show the outline, get approval before drafting
• Write the draft – following brand voice, with banned word checks
• Checkpoint 2 – review the draft before finalizing
• Vale linting – enforce style consistency, must pass with 0 errors
• Generate images – Python script creates header and thumbnail
• Submit PR – ready for review  

The skill is a single slash command: /blog “topic”. That’s it.

/blog “topic”Making Standards Executable

Here’s how we enforce brand voice — this is from the actual skill file:

Banned Words:
opt, dive, unlock, unleash, intricate, utilization, transformative,  alignment, proactive, scalable, delve, landscape, evolving, nuanced,  paradigm, comprehensive, holistic, synergy, pivotal, robust, empower,  enables, transforms, solution, optimized

Banned Phrases:
- "In today's world," "at the end of the day," "best practices"
- "Game-changer," "paradigm shift," "breakthrough technology"
- "Seamless integration," "effortless setup," "works like magic"
- "Enterprise-grade" (without defining), "industry-leading"

When the skill generates a draft, it checks against these rules. Before creating a PR, it runs Vale linting, and it must pass with 0 errors. If banned words slip through, Vale catches them. Technical terms need backticks for formatting? Vale flags them.

What used to require me reviewing every piece of content now happens automatically. The skill enforces consistency I couldn’t maintain manually across a team.

The Part I Didn’t Expect

We rolled this out to the marketing team a a week ago.

They adopted it fast. Really fast.

Giving this tool to the team essentially replaced one of my primary functions: being the bottleneck for content quality and technical accuracy. And that’s freeing me up to do what I actually want to do: developer outreach, building relationships, experimenting with new approaches to advocacy.

We’re covering more keywords, faster. Blog posts ship with consistent voice and style. SEO optimization, FAQ sections, internal links, quotable statistics — is baked into the workflow instead of being an afterthought. The team can focus on research and strategy rather than fighting with formatting and style checks.

We don’t have metrics yet (it’s been three weeks), but we’re already producing more content than we were before.

What’s Working, What Isn’t

AI tools are most powerful when they’re systematic. One-off prompts are useful. Codified workflows are different. The skill isn’t “use AI to write a blog post,” it’s “here’s our exact process, automated and enforceable.”

Quality controls matter. Without checkpoints and linting, AI output drifts. We have two approval gates (outline and draft) and automated linting before anything ships. This keeps quality high without manual review of every detail.

This doesn’t replace judgment. It can’t replace deep technical knowledge, understanding our audience, knowing when to break the rules. It handles structure, consistency, grunt work. Humans handle the hard stuff: knowing what to say, how to say it, when to bend the rules.

Where This Goes Next

I’m not afraid of AI replacing developer advocacy work. I’m more worried about being too slow to adapt.

The industry is still figuring out what AI-assisted DevRel looks like. We’re in the crystallization phase — experimenting, sharing what works, discarding what doesn’t. I’m already thinking about the next iteration: using skills to personalize developer onboarding, create custom documentation paths, generate context-aware code examples.

The terminal has always been where I do my best work.

Now it’s where I build tools that let the whole team do theirs.

If you’re in DevRel or developer marketing: start small, build workflows, share what you learn. We’re all defining what this looks like together.