Give Claude a permanent memory of how you write, who you help, and what you sell.
Every new Claude Code session starts from zero. You re-explain your voice, re-paste your audience, re-describe your offer, and still get output that reads like everyone else's. This template fixes that once. You answer an interview, Claude writes your brain, and every session after that starts with your context already loaded.
Takes about 10 minutes. You don't need transcripts, recordings, or a content library to start.
your-workspace/
├── .claude/commands/ 3 slash commands: /build-brain, /process-brain, /session-wrap
├── ai-brain/ Your knowledge base
│ ├── My-Voice.md How you write and speak
│ ├── My-Audience.md Who you help and their exact words
│ ├── My-Business.md Offers, positioning, what's retired
│ ├── Inspiration-Vault.md Techniques swiped from other creators
│ └── [content folders] Drop new material here
├── projects/ Your actual work
└── workspace-CLAUDE.md Template for your global Claude context
The three brain docs are the point. Claude reads them before writing anything in your name, so the output sounds like you instead of like AI.
1. Drop the contents into your Claude Code workspace
Copy .claude/, ai-brain/, and projects/ into whatever folder you open in Claude Code.
They need to sit at the root, next to each other.
If you don't have a workspace yet, make one:
mkdir -p ~/my-workspace && cd ~/my-workspace
git clone https://github.com/justynberk/claude-code-brain-template.git temp
mv temp/.claude temp/ai-brain temp/projects temp/workspace-CLAUDE.md .
rm -rf temp
claude2. Run the interview
In Claude Code:
/build-brain
Claude asks one question at a time across four rounds: setup, voice, audience, business.
Answer specifically. Vague answers make a weak brain. At the end Claude writes
~/.claude/CLAUDE.md plus all three brain docs automatically. Nothing to fill in by hand.
Anywhere Claude doesn't have enough signal, it flags a [GAP: ...] instead of inventing
something. Those gaps close as you feed it more material.
3. Add working rules
Open ~/.claude/CLAUDE.md and fill in the Working Rules section. This is where you put the
non-negotiables Claude should follow in every session. Mine include "never use em dashes"
and "start with the answer, no preamble." Yours will be different.
new content → drop in the right folder → /process-brain → brain docs sharpen
The interview gets you a working brain. Feeding it real material is what makes it good. Anything you've written or said in your own voice is fuel:
| Folder | What goes in |
|---|---|
ai-brain/Voice-Recordings/ |
Voice memos, Wispr Flow exports, spoken content |
ai-brain/AI-Chats/ |
Exported Claude or ChatGPT conversations |
ai-brain/Transcripts/ |
YouTube, podcast, video, or coaching call transcripts |
ai-brain/Emails/ |
Emails you've written |
ai-brain/Inspiration/ |
Other creators' scripts, emails, landing pages |
ai-brain/Other/ |
Anything else worth feeding the brain |
Then run:
/process-brain
Claude reads every new file, pulls out voice, audience, and business signal, updates the
brain docs with verbatim quotes and citations, and moves what it processed into _processed/.
Run it again next month with new material and the docs get sharper.
Inspiration/ works differently. Those files are other people's work, so nothing from them
touches your voice. They feed Inspiration-Vault.md as a technique swipe file instead.
You can add your own folders alongside these. Drop a CLAUDE.md inside explaining what
belongs there and /process-brain picks it up automatically.
/build-brain - One-time interview. Writes your global Claude context and all three
brain docs. Voice is written in a separate pass from audience and business so the two
don't blend into generic marketing language.
/process-brain - Reads new files in your content folders and updates the brain.
Run it whenever material has piled up.
/session-wrap - End of a working session. Updates the active project's CLAUDE.md
with current state, next steps, and known issues, then asks whether anything from the
session belongs in the brain.
If you build voice and positioning in the same pass, the voice doc drifts toward how you
sell, not how you talk. So /build-brain writes audience and business first, then
deliberately re-reads your raw voice answers with that context dropped.
Same reason every claim in My-Voice.md carries a quote. It's evidence, not Claude's
opinion of how you sound. If a line in there feels wrong, you can trace it back to the
thing you actually said and cut it.
My-Voice.md, My-Audience.md, and My-Business.md are never finished. Edit them
directly whenever something reads wrong. Claude treats them as the source of truth, so a
stale line in there costs you on every piece of output.
My-Business.md has a Deprecated table. Put retired offers, old pricing, and dead
positioning in it. Claude checks that before referencing anything you sell, which stops it
from promoting an offer you killed six months ago.
Your brain is yours. It sits in plain markdown in your own workspace, and nothing in this
template sends it anywhere. If you version your workspace with git, add ai-brain/ to your
.gitignore before your first commit.
MIT. Use it, fork it, change it, ship it.
Built by Justyn Berkovits. If it works for you, tell me.