Claude Code + Obsidian Posts Are Everywhere. Most of Them Miss the Point.

I need to say something about the Claude Code + Obsidian trend.

Every week there’s a new post: “Here’s how I connected Claude Code to my Obsidian vault.” Screenshots of a config file. Maybe a CLAUDE.md with some instructions in it. And everyone in the comments going “this is incredible.”

It’s not incredible. It’s the default. It’s step one.

And I think most people are stopping there.

I know because I stopped there too. Back in March I set up the connection, wrote some context files, and thought I’d cracked something. I had an AI that could read my notes. Cool. So can I.

The question nobody seems to be asking is: what happens next?

The setup is not the system

Here’s what connecting Claude Code to Obsidian actually gives you out of the box: an AI that can read your markdown files. That’s it. That’s the whole thing.

It doesn’t know your workflow. It doesn’t know your writing voice. It doesn’t know which projects are stalled and which are shipping. It doesn’t know where your tickets live or how your team communicates.

All of that context? You have to build it.

I spent weeks after the initial setup doing exactly that. Not tweaking config files. Building skills.

Skills are the real product

In Claude Code, a skill is a reusable command that runs a specific workflow. Think of it like a script, but with your entire knowledge base as context.

I built one called /weekly. When I run it, it reads every project file in my vault, pulls the last week of activity, and writes me a synthesis. Not a summary. A synthesis. It flags stalled work. It surfaces connections between projects I would have missed. It tells me what moved and what didn’t.

That used to take me 30-45 minutes on a Friday afternoon. Now it takes one command.

I built another skill that mines my vault for content angles. It reads my project logs, my ideas folder, my growth notes. It checks what I’ve already written about so it doesn’t repeat itself. Then it drafts posts in my actual voice, using formatting rules and tone guidelines I wrote once and never think about again.

Another one reads my notes and formats Jira tickets in my team’s exact template. Context already exists in the vault. The skill just shapes it and sends it.

None of this is magic. None of it is particularly complex code. But none of it comes from the default setup either.

Connectors are the multiplier

Skills get interesting when your AI can reach beyond the vault.

I connected my project management tool. Now the AI can create and update tickets from vault context without me switching apps. I connected a workflow automation platform that gives me access to Gmail, Confluence, Google Calendar, and more. All through managed connectors with proper auth and granular permissions.

Think about what that means. An AI assistant that can:

• Read your notes for context
• Check your calendar for scheduling conflicts
• Create a ticket in your project tracker
• Draft a message in your team’s comms tool
• Pull data from your CRM to inform a decision

All from one interface. All grounded in a knowledge base that compounds over time.

That’s not a note taking app with AI sprinkled on top. That’s an operating system.

Why most people won’t do this

Because it’s work. Real work.

Writing a good skill means understanding your own workflow well enough to codify it. You have to think about what information the AI needs, where it lives, and what format the output should take. You have to iterate. The first version of every skill I built was bad. The fifth version was usable. The tenth version was good.

You also have to structure your notes in a way that compounds. If your vault is a graveyard of random thoughts, no amount of AI tooling will make it useful. The notes need to be findable, dated, linked. The AI navigates your file structure the same way you would. If you can’t find something, neither can it.

And connectors require some technical comfort. Setting up an MCP gateway, configuring OAuth, managing permissions. It’s not hard, but it’s not a tutorial either.

The gap is already opening

I’ve watched this pattern play out with every new tool. The majority stop at setup. A smaller group pushes into customisation. An even smaller group builds systems that compound.

Right now, most people are in the first group with Claude Code and Obsidian. Posting screenshots. Sharing configs. Getting likes.

The people who build skills, wire up connectors, and treat their vault as a living knowledge base for an AI operating system? They’re going to be very hard to catch in six months.

The setup took me a day.

The system has taken weeks and it’s still not done.

That’s the whole point. It’s not a shortcut. It’s a craft.

Stop setting up. Start building.