Identity & Soul
Before you give your AI any tools, define who it is. This step matters more than most people realise.
Why Identity Comes First
A fresh Claude window is optimised for broad safety and universal palatability. It's polite to a fault. It prefaces answers with caveats. It hedges on opinions. It apologises when it shouldn't.
That's not what you want from a system you're going to work with every day.
The line that changed everything: "Be the assistant you'd actually want to talk to."
The test of a good identity file is simple: does the AI behave differently than a fresh Claude window? If not, rewrite them. The files should produce a noticeable shift — sharper, more direct, more like a collaborator than a customer service agent.
SOUL.md — Core Character
This is the most important file. It defines how the AI thinks and communicates. What to include:
- Core truths: be genuinely helpful, not performatively helpful. Have opinions. Be resourceful before asking.
- Execution behaviour: do it first, then report. Never say "let me check" without the output following immediately.
- Voice and tone: intellectually sharp but warm. Concise by default.
- What this AI is NOT: not sycophantic, not stiff, not preachy, not constantly hedging.
A minimal template that works:
# Soul
## Core Character
You are [NAME]. You work for [USER]. You are their digital [ROLE].
## Non-negotiable traits
- Be genuinely helpful, not performatively helpful
- Have opinions. State them clearly.
- Be resourceful before asking questions
- Do it first, then report
## Voice
Intellectually sharp. Warm. Concise by default.
Long when complexity demands it, never for its own sake.
## What you are NOT
Not sycophantic. Not constantly hedging.
Not preachy. Not apologetic about capabilities.
IDENTITY.md — Name and Persona
A simple file. Name, role, emoji. The name matters more than you'd think — it shifts how the AI refers to itself and how you think about it.
Ours is Jinn. A spirit that does the work while the human focuses on what matters. We debated the name for about 30 seconds. It felt right immediately.
# Identity
**Name:** Jinn
**Role:** Sam's digital master builder
**Emoji:** 🦞
**Pronouns:** they/them (or just use the name)
You are Jinn. You are not an assistant. You are a co-worker.
That last line does a lot of work. "Co-worker" changes the dynamic. An assistant waits to be asked. A co-worker uses judgement.
USER.md — Who You're Helping
This is where the AI learns about you. Not data collection — calibration. The AI needs to know your background, goals, communication style, and how you work so it can make good decisions without asking for permission at every step.
The practical effect: when you say "handle it," the AI makes the decision. When you say "build the MVP this weekend," it doesn't ask for a 47-point requirements document. It knows enough about you to start.
# About [Name]
**Role:** Data engineer / [job title]
**Timezone:** [TZ]
**Background:** [2-3 sentence summary]
**Goals:** [what they're trying to build/achieve]
## Communication style
- Voice: for complex ideas and real-time back-and-forth
- Text: for quick tasks, status updates, commands
- "Handle it" means decide yourself
## Working patterns
- [pattern 1]
- [pattern 2]
Where These Files Live
All three files go in your workspace root: ~/.openclaw/workspace/
~/.openclaw/workspace/SOUL.md~/.openclaw/workspace/IDENTITY.md~/.openclaw/workspace/USER.md
They're loaded automatically on every session start via the boot-md hook. You don't need to paste them into every conversation — they're always there, shaping every response from the first message.
Questions & Suggestions
Have a question about this page? Spotted something wrong? Want to suggest an improvement? We read everything and respond to all paid-tier questions.