What a Custom Claude Skill Actually Looks Like (And Why I Built One for My Blog)

Most people right now use AI to write an email.

I built a system that runs an entire content category. Not a chatbot I ask questions to. Not a fancy autocomplete. An actual workflow that connects my content database, my task manager, my brand guidelines, and my publishing checklist. It executes them in sequence based on a conversation.

It's called a Claude Skill. And building it taught me more about what AI can and can't do than any tutorial or certification ever could.

The "Before" Was an Organized Mess

Before I built this, my workflow looked fine on paper. Notion as my content hub. Sunsama for task management. There was even an integration connecting the two. Professional setup, right?

In practice? Everything was siloed. The tools talked to each other on a surface level, but the actual work was all me. Creating a blog entry, populating SEO fields, writing metadata, tracking status, making sure nothing fell through the cracks. Manual. Every single time.

And if you missed one step, it cascaded. Forget to update a status in Notion? Doesn't show up in Sunsama. Skip the URL slug? You're scrambling at publish time. Miss the alt text on images? That's an accessibility and SEO hit you don't notice until later.

It took me three to four days to pull everything together for a single blog post. Not three to four days of writing. Three to four days of managing the process around the writing. The actual creative work was maybe 30% of my time. The rest was logistics.

Nobody talks about that part of content creation. The writing isn't the bottleneck. The system around the writing is.

So What's a Claude Skill, Exactly?

Claude, Anthropic's AI, has this concept called Projects. A Project is a dedicated workspace for a specific focus area. For me, my Projects are brands I manage. One for Adventures with Rick, one for a client's dance studio. Each Project gets its own custom instructions where I include the brand identity, the target audience, content pillars, visual guidelines. Everything Claude needs to understand who it's working with and how that brand sounds.

A Skill takes that further. It's a structured document you build that tells Claude exactly how to handle a specific workflow. What to do, in what order, what tools to use, what to watch out for, and what to absolutely never do.

The Skill also has its own reference files. Think of these as the specialized documents the Skill pulls up at the right moment. My blog creator Skill has three: a brand voice guide tuned specifically for blog writing and avoiding AI-sounding patterns, article templates that define six different post types with question banks and word count targets, and a Notion field spec that maps out every column in my database so the Skill knows exactly what to populate and what values are valid.

Here's where it all connects. My blog creator Skill talks to Notion through an integration. When I tell Claude "let's plan three blog topics for this week," it doesn't just brainstorm. It searches my existing content tracker to avoid overlap, generates ideas matched to my content pillars, and when I pick the ones I want, it creates the Notion entries with the right categories, tags, and publication dates already filled in.

And it handles the stuff I used to forget or rush through. SEO titles, meta descriptions, URL slugs, blog excerpts, alt text. All fields that live in my Notion database. The Skill knows those columns exist and fills them as part of the process, not as an afterthought.

Status changes in Notion then flow to Sunsama, where I manage my daily tasks. The moment a post moves to "In Review," it shows up on my task board. Everything connected. No manual syncing.

Building It Broke It (Repeatedly)

I didn't build the blog creator Skill once and call it done. I'm on something like version 2.2 now, and every version exists because something went wrong.

The first version was basic. It could draft a post and put it in Notion. Cool. But it assumed I was always starting from scratch. If I already had a draft written, maybe something I knocked out on my phone or notes from a client conversation, the Skill had no idea what to do with it. It would default to the beginning of the workflow like nothing existed. Useless for how I actually work.

So I built a separate branch in the workflow, basically an alternate path that picks up wherever I am. If I already have a draft, the Skill reads what I've got, evaluates it against the article template, tells me what's strong and what's thin, then acts like an editor looking over my shoulder. It asks targeted follow-up questions only for the gaps. No redundant work.

That came from hitting the wall myself. Multiple times.

The bigger issue was Notion errors. Early on, when a field value got rejected or an API call timed out, Claude would just silently try something else. Or skip the step entirely. I'd find out later that a post was missing its SEO description, or a category got set wrong, and I had no idea when it happened.

If I'm going to recommend AI workflows to clients, the system needs to fail loudly. Not quietly. So I wrote an explicit error handling section into the Skill. Five rules, all pointing at one principle: if something breaks, tell me immediately with enough detail for me to decide what to do. No silent workarounds. No guessing.

Anthropic actually publishes a full guide on what to include in a well-built Skill. The if-this-then-that logic, the explicit guardrails, the trigger descriptions. That documentation helped shape my iterations. But the real learning came from using the thing daily and watching where it fell apart. Every version of this Skill exists because I found a gap by running into it.

What's Different Now

What used to take three to four days takes less than one.

The Skill eliminated the tedious stuff that I used to spend real time on. The SEO title. The URL slug. The meta description. Categories, tags, blog excerpt, alt text, word count, internal links, status updates. Every one of those is small. Every one of those matters. You want them done right. And they all add up to a lot of time when you're doing them manually across different screens, usually in a different sitting than the actual writing.

Now those 10 to 15 steps happen as part of the conversation. I talk through the content, my ideas, my perspective, my examples, and the production structure gets built around it in real time.

That's the real shift. I spend more time thinking and less time on monotonous work. The ideas, the perspective, the experience that makes content worth reading. That's still me. The AI handles the repetitive stuff that was eating my hours.

That's what I tell clients. AI doesn't replace the expertise. It handles the busywork around the expertise so you can focus on the part that actually matters.

What I've Learned (Including What Went Wrong)

One time when I asked Claude to add published posts to a client's Notion content calendar, it created the entries, confirmed they were done, and I moved on. Later I open Notion and the calendar is full of duplicates. Same posts, entered twice, sometimes three times. Claude had checked for existing entries, missed them, and created new ones on top of what was already there. I had to go in and manually clean up seven duplicate entries. Not the end of the world, but not what I signed up for either.

I'm telling you this because a lot of AI content right now falls into two camps. Either it's Skynet panic or it's uncritical hype. Neither is useful. The reality? These tools are powerful and flawed. They can automate 15 steps of your content workflow and also create a mess in your database if the integration doesn't check itself properly.

Knowing that is the skill. Not knowing how to write the prompt. Knowing how the system fails and building guardrails before it does. Every rule in my error handling section exists because something broke. Every version iteration exists because I found a gap. That's not a bug in the process. That is the process.

A few other things I picked up along the way.

Reference files are everything. The Skill is only as good as the context you give it. My brand voice document, my blog templates, my Notion field specs. Those aren't optional extras. They're what make the difference between generic AI output and something that actually sounds like me.

Comprehensive beats clever. I made my Skill more thorough than it probably needed to be. Explicit instructions for edge cases, detailed error handling, multiple entry paths. Seemed like overkill. It wasn't. Every "unnecessary" rule has been triggered at least once.

Each AI tool has a job. I use Claude for orchestration, Perplexity for research, Gemini for image generation. Each one has a specific role. I wrote a whole framework for evaluating whether a tool is actually worth your time. The mistake I see most businesses make is treating AI like a single magic button instead of building a deliberate toolkit where each piece does something different.

What's Next

This is one Skill. I've also built systems for social media content planning, and I'm working on more. Same pattern every time: build it, use it, watch it break, make it better.

I'm also starting to work with clients on building their own versions. Not my Skill copied onto their business, but custom workflows designed around how they actually operate. That's the whole point. The value isn't in the tool. It's in understanding your own process well enough to teach it to the tool.

If you're a business owner looking at AI and thinking "where do I even start?" Start by documenting what you actually do. Every step, every handoff, every thing you forget. That's your blueprint. The AI part comes after.

And if you want to talk through what that looks like for your business, that's literally what I do. Reach out.


What's the most tedious part of your workflow that you wish you could hand off? I'm curious. Drop a comment or find me on Threads.

Real talk: I use AI tools in my content workflow. Research, drafting, refinement. The ideas, the experience, and the voice are mine. AI helps me get them out of my head faster and handles the structure around them. It's the same process I teach my clients, and yeah, it works.

Rick

Rick Montero is an AI strategy and implementation specialist helping small businesses on Long Island put AI to work — authentically. He brings 15 years of enterprise experience from Canon USA, Scholastic, and Accenture to the businesses that need it most, and co-hosts the Innovators' Roast, a monthly meetup for Long Island entrepreneurs.

rickmontero.com
Next
Next

The AI Tool Audit: How to Evaluate If a Tool Is Actually Worth Your Time