How to Build an AI-Era Super Brain with Obsidian and Claude

Author: Yanhua (@yanhua1010) Source: https://x.com/yanhua1010/status/2020342019575673223 Date: 2026-02-08 Stats: 895 likes · 231 retweets · 2,139 bookmarks · 247K views

A while back I translated Heinrich's "Obsidian + Claude Code Beginner's Guide." After finishing it, I stared at my own Obsidian vault for a long time.

Honestly, I felt a little ashamed.

I agreed with everything he said — your note vault is your thinking operating system, CLAUDE.md teaches AI your way of thinking, every note is essentially a "skill." I understood all of it. But my vault was still a pile of scattered Markdown files, nowhere close to a "super brain."

Then I asked myself: are my needs actually the same as Heinrich's?

They're not. He uses his vault for deep thinking. I'm a programmer whose side hustle is multi-platform content creation — running a WeChat public account, Xiaohongshu, and X simultaneously. I don't just need a vault that can think. I need a content production pipeline that can actually do work.

So I spent a few weeks turning Heinrich's theory into my own practice. This article is the complete record: how I transformed an ordinary Obsidian vault into an AI content factory — from information source to publication.

Step 1: Get Claude to Move Into Your Vault

To turn your vault into an AI workstation, you first need to let AI in. Two plugins are enough.

The Claudian plugin is the key. It's an Obsidian community plugin with a simple purpose: let Claude Code run directly inside the note editor. No switching to a terminal, no copy-pasting. You talk to Claude right next to your notes, and it can read your files, modify your files, and call various tools.

Installation: Claudian isn't on the official Obsidian plugin marketplace yet, so you need to install it manually. The easiest way is through the BRAT plugin: install BRAT from the community plugins marketplace, then in BRAT settings click "Add Beta plugin," enter https://github.com/YishenTu/claudian, and it installs automatically with auto-updates.

My core config: model set to Opus (quality first), permission mode set to yolo (auto-approve all operations). Yolo mode sounds aggressive, but combined with CLAUDE.md rule constraints, it's actually very smooth in practice. If you're just starting out, I'd recommend beginning with the default permission mode and upgrading once you're comfortable.

The second plugin is Terminal. Some operations still need the command line, and this plugin lets you open a terminal directly inside Obsidian without switching apps.

Once you have these two plugins, Obsidian is no longer just a note-taking app. It becomes an AI-staffed workstation — you write notes, AI stands by ready to act.

16 Custom Skills: Equipping AI with a Professional Toolbox

Heinrich said something that stuck with me: "Every note is in some sense a skill — curated knowledge, injected when needed."

I pushed that idea one step further. What I'm building isn't "knowledge notes" — it's "action instructions." Each Skill is a Markdown file stored in the .claude/skills/ directory, teaching Claude how to complete a specific type of task.

I currently have 16 Skills, organized into four categories:

Publishing: post to WeChat public account, post to X
Image generation: automatically generate style-matched images based on content
Content conversion: one-click save any webpage or tweet as clean Markdown
Utilities: image compression, auto-compress to WebP format

Let me highlight a few I use most.

Smart article illustration (baoyu-article-illustrator) is the one that impressed me most. It automatically analyzes your article structure, determines which positions need illustrations, then generates them using a "type × style" two-dimensional combination. If the article discusses data comparisons, it picks "infographic + blueprint style." If it's a personal story, it switches to "scene illustration + warm style." You don't specify anything — it judges on its own.

One-click publish to X (baoyu-post-to-x) uses real Chrome browser automation, supporting text, images, video, and long-form Articles. I write an article in Obsidian, say "publish to X," and it handles everything.

Webpage to Markdown (baoyu-url-to-markdown) is my go-to for collecting source material. See a good article, drop in a link, and in seconds it becomes clean Markdown stored in my vault. That's how I build my source library.

From Podcast to Publication: A Complete Content Pipeline

Theory is too abstract — let me walk through a real example.

A few days ago, Silicon Valley 101 released a podcast where three guests did a deep dive on why Clawdbot exploded in popularity. I listened and thought the information density was high, so I decided to write an article.

Step 1: Turn the information source into text.

I used Podwise to convert the podcast into structured Markdown notes — with timestamps, chapter breaks, and key insight extraction. Imported into Obsidian as raw material.

Step 2: Claude writes the first draft.

I opened the podcast notes in Obsidian and used Claudian to tell Claude: based on these podcast notes, write a long-form X post, distill the most core insights, use my writing style. Claude read the podcast content and also read my CLAUDE.md (which encodes my identity, audience positioning, and writing preferences), and produced a first draft. I reviewed and revised, adjusting structure and wording, then finalized.

Step 3: Auto-generate illustrations.

I called the "Smart Article Illustration" Skill. It automatically analyzed the article structure, identified 4 illustration positions, and generated 4 style-matched images: "The Warm Daily Life of an AI Butler," "The Heartbeat of a Memory System," "Agent Computer vs. Human Computer," and "The General and the Fairy Army."

Step 4: One-click publish.

I called the "Publish to X" Skill, and the article was published directly as an X Article.

From finishing a podcast to publishing a deep-dive article with illustrations: the whole process took about 30 minutes. Before, this would have taken at least an afternoon.

The same content can be distributed to WeChat and Xiaohongshu using different Skills, with format auto-adapted and cover images auto-generated. One piece of source material, multi-platform distribution.

This is what I mean by "content production pipeline." Information comes in, finished product goes out, with AI assistance at every step in between.

Giving Your Notes Wings: The Secret of CLAUDE.md

The soul of the whole system isn't any specific Skill — it's CLAUDE.md.

CLAUDE.md is a special file that Claude reads at the start of every session. It's essentially an "operations manual" that tells AI: who you are, how you work, what your standards are.

Here's what I've encoded in my CLAUDE.md:

Identity definition. "I'm a programmer whose side hustle is multi-platform content creation. Frontend and design are my weak spots." After adding this one line, Claude stopped recommending complex frontend solutions and started proactively simplifying things and explaining in detail.

Skills usage guide. Trigger conditions and use cases for all 16 Skills. When I say "generate a WeChat cover," Claude knows which Skill to call. When I say "illustrate the article," it knows to use a different one.

Workflow templates. Complete step-by-step processes for WeChat article creation, X content publishing, and Xiaohongshu content — every platform's full workflow is written in.

Iterative evolution philosophy. This file isn't written once and left alone. Claude makes a mistake, I add a rule. I discover a better process, I update the template. It's a living document recording all the lessons learned from my AI collaboration.

The process of writing CLAUDE.md is itself incredibly valuable. You're forced to put into words everything you "thought you knew but never said out loud" — your work habits, quality standards, ways of thinking. This process forces a deep act of self-knowledge.

Hidden in these Markdown files is a clearer version of "you" than the one in your memory.

Directory structure is also part of the brain. My vault is organized by category: x/ for X platform content, wechat/ for WeChat material, xiaohongshu/ for Xiaohongshu content, .claude/skills/ for all Skills. Structure is logic. Directories are the partitions of the brain.

Your Super Brain Needs Just Three Things

Looking back, this system isn't complicated. The core is just three things:

Obsidian (the container): your Markdown files, your knowledge, your structure
Claude Code + Claudian (the intelligence): AI that can read, think, and act
Skills (the hands): modular capabilities that let AI actually do work

The tools themselves don't matter. What matters is that you use these tools to build a system that amplifies your own unique perspective. AI can generate ten thousand words in a minute, but only you know which two thousand are worth reading.

If you want to start, you don't need to build the whole system at once. Three steps are enough:

Install Obsidian + the Claudian plugin
Write your first CLAUDE.md — even just 10 lines
Try one Skill — start with "Webpage to Markdown," the simplest and most practical

My biggest gain from building this system wasn't the efficiency boost. It was being forced to put my work style, thinking style, and value judgments all into words.

That might be what "super brain" really means in the AI era — not AI thinking for you, but AI helping you see clearly how you think.

If you're also building your own Obsidian + AI system, feel free to share your workflow in the comments. Everyone's "super brain" is different, and that's exactly what makes it interesting.