Senses · 01: Claude Code Setup

Senses is how the Cortex sees and acts. The first sense is the one you touch: Claude Code, the app you type into where Claude reads your Codex, runs your skills, and does the work. This chapter gets you from nothing to running a real skill on a real file and watching it work. Get this right and every later chapter has somewhere to run.

Why Senses comes after Memory

You built Memory first on purpose. The app is only useful once it has something to read. Claude Code with no Codex is a generic assistant. Reading your Codex, it drafts in your voice and names your real customers. So this chapter isn’t “install a tool.” It’s “point the tool at the Memory you built,” so the moment you open it, it knows who you are.

Three things turn a bare Claude Code into your main surface: the instruction file it reads each session, the auto-memory that remembers what it learns about you, and the folders where your skills live. Each feeds it context on its own, so you stop re-explaining yourself.

The install

1. Claude Code itself

Install it and open it in your business folder, the one holding your Codex. Ask it to read index.md and tell you what’s here. If it reads your Codex back, the app is connected to your Memory. That’s the whole test.

You can run it in a text window on your computer (the terminal), a code editor, or a desktop app. Pick the one you’ll live in; chapter 03 helps you choose.

2. The two-layer instruction file

Claude Code reads a plain text file of standing instructions called CLAUDE.md. It comes in two layers, and the layering is the trick:

~/.claude/CLAUDE.md       # GLOBAL: loaded everywhere, every session
<your project>/CLAUDE.md  # PROJECT: loaded only in that folder, adds on top

Global holds what’s true everywhere: who you are, how a session runs, your standing rules.
Project holds what’s true only here: this project’s quirks and conventions.

The global file is what makes every session start warm. Here’s the shape, not the contents:

## Identity
[Who you are and what your business is, in a few lines.]

## Session lifecycle
[How a session starts and ends. Match the folder to a project, lead
with what's queued, save progress at the end.]

## Preferences
[Defaults so Claude never re-asks: how you like things done.]

## House rules
[Standing conventions, e.g. "draft emails, never send them."]

3. Auto-memory

Claude Code keeps its own memory folder and reloads it each session. This differs from the Codex. The Codex is what you write down about your business; auto-memory is what Claude learns about working with you (your corrections, your preferences, where a project stands), saved on its own and recalled next time.

Together they’re why a session needs almost no warm-up: the Codex tells Claude about your business, auto-memory tells Claude about you.

4. The skills and commands folders

Skills and commands are shortcuts you trigger by typing a slash, like /kb-ingest. They live in two folders (~/.claude/skills/ and ~/.claude/commands/). Create them now; you’ll fill them in the Habits chapters.

Pitfalls (the “if this, then that” guide)

Situation	What to do
Your global `CLAUDE.md` is empty	Fix this first. Even three sections (identity, preferences, house rules) change every session. Brief it like a sharp new assistant.
You put project-only rules in the global file	Move them to that project’s `CLAUDE.md`. The global file is for what’s true everywhere; project details bloat it and leak into other work.
You mix up the Codex and auto-memory	Codex is what you write about your business. Auto-memory is what Claude learns about you. Keep them separate.
You keep re-explaining the same preference	That’s a memory that should be saved. Tell Claude to remember it, or add it to the global file if it’s a standing rule.
You stuff the global file with everything	Keep it to standing rules and identity. It loads every session, so length taxes every conversation.
Claude isn’t reading your Codex	Open it inside (or one level above) your business folder, and make sure `index.md` points to where things live.

Maya’s version

Maya is not a coder. She opened Claude Code as the desktop app in the folder holding her Codex and asked it to read her index.md. It came back knowing her online store, her service work, and her content channel. First test passed.

Her global CLAUDE.md is four short sections: who she is, how she likes a session to run, a couple of preferences, and one house rule (“draft emails, never send them”). Her store project has its own tiny CLAUDE.md with supplier names and shipping rules that only matter there.

She wrote almost nothing. A short global file, an even shorter project file, and Claude’s auto-memory filled in the rest as she worked. By the third session she’d stopped re-explaining herself.

Worth keeping: the global file encodes a working relationship, not just preferences. Describe how a session starts and ends, and Claude acts like an assistant who knows the routine, not a tool waiting to be told. Project files draw the boundaries, so each context gets the rules it needs and none it doesn’t.

You’ve got it when

Claude Code runs from wherever you’ll use it (terminal, editor, or desktop app)
A global CLAUDE.md loads on every session, and a fresh session can answer “what do you know about me?”
Auto-memory exists and has recalled something useful without you re-stating it
The skills/ and commands/ folders exist and hold your starter set
At least one project has its own CLAUDE.md that adds to the global one
You’ve run a real skill (/kb-ingest or /session-check) start to finish

If a fresh session opens already knowing your work, the sense is live. The last two prove it’s carrying context, not just installed.

The skills that help here

Two commands help with setup and upkeep: /update-config (adjusts Claude Code’s settings) and /session-check (an end-of-session check that nothing gets lost). Both are in the starter kit; the full list is in ../03-habits/01-skills-and-commands.md.

→ Next: 02-connectors.md. Give the Cortex eyes, ears, and hands in the tools you already use: the connectors that let it reach your email, notes, and files without copy-paste.