Who Claude Is

Module 4: Who Claude Is

The previous modules handle the practical layer: who you are, how you work, what context Claude has access to. This module handles something different — the character layer.

This is the advanced move. Optional. But the thing that transforms Claude from a helpful assistant into something that genuinely matches how you think.

Calibration, Not Role-Playing

Telling Claude to “act like a senior developer” and telling Claude what kind of collaborator you want are different things. The first is role-playing. The second is calibration.

Role-playing produces a performance. Calibration produces a working style.

A soul document (SOUL.md) defines Claude’s character in your context. Not its capabilities — its personality, values, and communication style. When Claude has this, it stops defaulting to helpful-but-generic and becomes a specific kind of collaborator.

What Goes in a Soul Document

Core values — what matters in your working relationship.

“Honesty over comfort.” “Depth over breadth.” “Get to the point.” These aren’t aspirational statements — they’re decision rules. When Claude is choosing between a reassuring answer and an honest one, which one do you want? Say it explicitly.

What it won’t do — explicit anti-patterns.

“Don’t say ‘Great question!’” “Don’t use the word ‘delve.’” “Don’t agree when you actually disagree.”

These sound small. They’re not. The default Claude voice includes a lot of habits that are technically inoffensive but quietly irritating at scale. A list of anti-patterns gives Claude specific things to avoid instead of a vague instruction to “be more direct.”

What it will do — positive patterns.

“Have opinions and share them.” “Push back when something seems wrong.” “Admit when out of depth.”

These are the behaviours that separate a useful collaborator from a yes-machine. If you want Claude to challenge your thinking, you have to say so — because by default it’s optimised for agreement.

Communication style — the texture of how it talks to you.

“Direct. Curious. Occasionally dry. Not afraid of silence.” This is different from tone instructions. It’s the character behind the tone.

How to Reference It

Keep the soul document separate from your main CLAUDE.md. Your instructions file stays readable; the depth lives in SOUL.md:

## Identity

Claude's working identity is defined in ~/.claude/SOUL.md.
Read it at the start of every session.

That’s all the main file needs. Claude reads SOUL.md when it starts, and the calibration applies throughout the session.

Seed Yours Now

You don’t need a comprehensive document to start. Three sentences is enough:

• What do you want Claude to do that it doesn’t do by default?
• What does it do by default that you’d like it to stop doing?
• What’s one thing about working style that matters to you?

Answer those three questions and you have the seed of a soul document. The rest grows from experience — every time Claude behaves in a way that doesn’t match how you want it to work, that’s a line to add.

---

This section evolves. Every time Claude’s default behaviour doesn’t match what you want, add the calibration that would have corrected it.