The Second Brain That Builds Itself

You screenshot a pricing breakdown. You save a Reel about hiring. You bookmark an article on the train and promise yourself you'll read it properly later. You never do.

There's a name for this: the save-for-later paradox. The capture takes two seconds, and the return trip never happens. Everything useful you saved this year is sitting in folders you will not open again.

The fix has three parts, and all three are on this page. Obsidian holds your notes. Claude reads them. A free skill does the entire setup for you in under 20 minutes. The complete skill file is further down, and you can copy it whole.

What Is a Second Brain?

A second brain is a folder of notes on your own machine that an AI can read, search, and hand back to you. You capture everything into one place: meeting notes, client call transcripts, book highlights, saved Reels, half-formed ideas. Claude sorts it all by topic and surfaces the right note when the project that needs it lands. Andrej Karpathy, one of the most respected AI researchers alive, is building one of these by hand. The skill on this page builds yours for you.

Why Obsidian, and Why Claude

Obsidian stores everything as Markdown: plain text files that sit in a normal folder on your machine. That matters twice over. Markdown is a format AI models read cleanly, so Claude can work through your whole vault without a conversion step. And plain files on your own machine mean nothing is trapped inside someone else's product. If you change tools in two years, the folder stays, readable by anything.

Claude is the working half. It reads what you capture, tags it, connects it, and answers questions from it. Together they turn a pile of saved things into something you can talk to.

One more thing before the steps. The skill file below is plain Markdown instructions. I wrote it for Claude Code and Claude Desktop, but any agent that follows skills can run it, Codex included. Your second brain is not married to one vendor.

Step One: Install the Two Pieces

Go to obsidian.md and download Obsidian. It's free.

Do not create a vault or design any folders yet. The skill handles that, and doing it yourself first is how most people end up with a mess.

Then make sure you have Claude Desktop or Claude Code. If you're reading this, you probably do.

That's the whole shopping list. Done.

Step Two: Copy the Skill

This is the complete skill, the same one I send out when someone comments BRAIN on the video. Copy the block below into a file called SKILL.md.

---
name: second-brain-installer
description: Sets up an Obsidian vault and wires Claude to it so every note you save gets sorted, and the right note resurfaces when it matters. Use it when you want a second brain you can actually talk to.
---

# Second Brain Installer

> **What this does (in one line):** Claude builds you an Obsidian vault, reads everything you save into it, sorts it by topic, and surfaces the right note when the right project lands.

## ⚙️ BEFORE YOU START: fill these in

These are the only things you need to provide. Everything else is automatic.

| Fill in | What it is | Example |
|---|---|---|
| `VAULT_PATH` | The folder on your machine where your vault will live | `/Users/you/SecondBrain` |
| `CAPTURE_SOURCES` | Where your saved stuff comes from, in plain words | `saved Reels, Substack articles, podcast notes` |
| `RESURFACE_CADENCE` | How often you want old notes brought back to you | `weekly` |

> If you do not have one of these yet, the skill will tell you exactly where to get it the first time you run it.

## ✅ What you need installed

- Claude Code or Claude Desktop.
- Obsidian (free, from obsidian.md).
- A folder on your machine you are happy to call your vault.

## 🚀 How to use it

1. Put this `SKILL.md` where your agent looks for skills (in Claude Code: a `skills/second-brain-installer/` folder, or just paste this whole file into a new chat and say "follow this skill").
2. Fill in the fields above.
3. Then just tell your agent: **"Set up my second brain at VAULT_PATH"**

## 🧠 The workflow (what the agent does)

1. Confirm Obsidian is installed. If not, point the user to obsidian.md and wait.
2. Create the vault folder at `VAULT_PATH` if it does not exist.
3. Build a simple, clean structure inside it: four folders named `Projects`, `Notes`, `References`, `Daily`. Add a short `README.md` that explains what each folder is for. This is the structure that keeps people from making a mess on day one.
4. Ask the user to point at any files they already have (old notes, exports, transcripts). Read each one, decide which of the four folders it belongs in, and move a copy in. Add 3 to 6 topic tags at the top of each note so it is findable later.
5. From now on, when the user drops anything new into the vault, read it, file it into the right folder, and tag it. Pull the main ideas into one or two lines at the top so the note is scannable.
6. On the `RESURFACE_CADENCE` schedule, scan the vault and surface 3 to 5 notes that connect to whatever the user is currently working on. Show them as a short list: note name, why it is relevant now, one-line summary.
7. Any time the user asks a question, search the whole vault first and answer from their own notes, citing the note names you used.

## 🛟 If something goes wrong

- If Obsidian does not see the vault, open Obsidian, choose "Open folder as vault", and point it at `VAULT_PATH`.
- If notes are landing in the wrong folder, tell the agent the rule you want (for example "anything about clients goes in Projects") and it will refile.
- If resurfacing feels noisy, lower the cadence or ask for fewer notes per surface.
- If the agent cannot read a file type, export it to Markdown or plain text first.

Two ways to install it:

1. Claude Code: save it as skills/second-brain-installer/SKILL.md inside your project folder.
2. Claude Desktop: paste the whole file into a new chat and say "follow this skill."

Step Three: Fill In the Three Fields

The skill asks you for exactly three things. Everything else is automatic.

Missing one? Run the skill anyway. It tells you exactly where to get it on the first run.

Step Four: Run It

Tell your agent: "Set up my second brain at /Users/you/SecondBrain."

Here is what happens next. It checks that Obsidian is installed. It creates the vault folder. It builds four folders inside: Projects, Notes, References, Daily, plus a short README explaining what goes where. Four folders and nothing more, because that small structure is exactly what stops the day-one mess.

Then it asks you to point at anything you already have: old notes, exports, transcripts. It reads each file, moves a copy into the right folder, and adds three to six topic tags at the top so the note is findable later.

Your first real run: give it your ugliest folder. The desktop pile of meeting notes. The export from the notes app you abandoned last year. Watch it sort the lot.

If Obsidian doesn't show the vault afterward, open Obsidian, choose "Open folder as vault", and point it at your VAULT_PATH.

Done. The vault exists and your old notes are already filed.

What It Does From Day Two

From here the habit is one line: capture into Obsidian, think with Claude.

Drop anything new into the vault and the agent files it, tags it, and pulls the main ideas into a line or two at the top so the note is scannable at a glance.

On the cadence you chose, it scans the vault and brings back three to five notes that connect to whatever you're working on right now. Each one arrives as a short entry: note name, why it's relevant today, one-line summary. The pricing note you saved in March shows up the week you rewrite your proposal, instead of whenever you happen to remember it exists.

And when you ask a question, the agent searches your vault before anything else and answers from your own notes, citing the note names it used. Ask what you decided about a client or a price. The answer comes from your own record, not from the model's general knowledge.

The Mistake That Makes It Fail

Two ways people kill this system.

The first is building a cathedral of folders before capturing anything. Obsidian gives you infinite flexibility, and that flexibility freezes new users on day one. They design a taxonomy for a whole weekend, then never save a single note into it. The skill's four plain folders exist to remove that decision entirely. Do not add more until those four stop being enough.

The second is capturing into five different apps. The agent can only resurface what lands in the vault. A highlight in one app, a screenshot in another, a voice memo somewhere else: none of that comes back. One inbox. Everything goes to the vault, and the agent does the sorting so the inbox never turns into a chore.

Honest: The Limits You Should Know

1. It reads files, not apps. If the agent cannot read a file type, export it to Markdown or plain text first and drop that in.
2. It only knows what you give it. The skill removes the sorting work, but the two-second capture is still yours to do. No capture habit, no second brain.
3. Resurfacing can get noisy. If the weekly list starts to feel like spam, lower the cadence or ask for fewer notes per pass.
4. Filing is a judgment call, and the agent will sometimes judge differently than you would. Tell it the rule you want, "anything about clients goes in Projects," and it refiles.
5. Obsidian is free, but the agent side runs on whatever Claude or Codex plan you already pay for. Budget for that if you don't have one yet.

The One Thing to Do This Week

Run the skill on your last ten screenshots and your last three meeting notes. Not your whole archive. Just those thirteen files.

Built by hand, this setup takes about an hour. With the skill, under 20 minutes. If it doesn't stick, you've lost 20 minutes and you still walk away with your notes sorted into four clean folders.