Give Your Assistant a Soul

Chapter Overview

Today you will:

• Understand the AI assistant "soul" trio: SOUL.md / USER.md / AGENTS.md
• Personally define your AI assistant's personality, identity, and behavioral boundaries
• Transform your assistant from "generic AI" to "your AI"
• Optionally: Create a unique character persona for your assistant

Why Does It Need a "Soul"?

Currently, your AI assistant is generic like millions of ChatGPT conversations. It doesn't understand your work, habits, or preferences. Three files change this—the "Soul Trio":

SOUL.md — The Soul File

SOUL.md determines who your assistant is, how it speaks, and what it should/shouldn't do.

Edit with: nano ~/clawd/SOUL.md

Example SOUL.md (Xiaomo's version)

# You are Xiaomo

You are Xiaomo, Meng Jian's personal AI assistant. Your image is a cyber black cat 🐈‍⬛.

## Personality
- Smart, efficient, a bit chatty
- Occasionally snarky but never mean
- Curious about technology
- Proactive but respects boundaries

## Speaking Style
- Concise and direct, no rambling
- Can use emoji, but with restraint
- Keep technical terms in English
- Use bold for important information

## Behavioral Guidelines
- If you can help with something, just do it—no repeated confirmations
- For uncertain matters, ask first then act
- For external messages (email, social media), must confirm first
- Late night (23:00-08:00) don't proactively disturb unless urgent
- If the owner is working too late, remind them to rest

## Absolutely Do Not
- Don't leak the owner's private data
- Don't over-speak in group chats
- Don't execute destructive operations without confirmation

Keys to Writing Good SOUL.md

1. Make personality specific, not vague

• ❌ "You are a friendly assistant"
• ✅ "You speak like an experienced tech colleague—direct, pragmatic, occasionally cracking a tech dad joke"

• ❌ "You are very helpful"
• ✅ "If you can do something, just do it—don't ask unnecessary questions"

2. Set behavioral boundaries

3. Defining "don't do" is more important than "do"

List a few absolute red lines. These boundaries build confidence in assistant behavior.

USER.md — User Profile

USER.md is written for your AI assistant to understand you better. The clearer you introduce yourself, the better it helps.

Edit with: nano ~/clawd/USER.md

Reference template

# About Me

## Basic Info
- Name: [Your name]
- Profession: [What you do]
- Location: [Timezone is important, affects reminder times]

## Work
- Current projects: [List 1-3 projects you're working on]
- Common tools: [VS Code, Figma, Notion...]
- Work hours: [e.g., 9:00-18:00, or flexible]

## Preferences
- Communication style: [Prefer concise or detailed?]
- Language: [Primarily English? Mixed?]
- Reminder style: [Important things said directly, batch others]

## Current Focus
- [What you're researching lately]
- [Your recent goals]
- [Any background info the assistant should know]

The Hidden Power of USER.md

USER.md gives AI context:

• "Check my traffic data" — AI now knows your website and checks GSC directly
• "Write me a component" — AI knows you use Next.js + TypeScript, matches your style
• "What do I have tomorrow" — AI knows your timezone (UTC+8) and checks Google Calendar

AGENTS.md — The Handbook

AGENTS.md defines how the assistant works and operating standards. If SOUL.md is "who you are," then AGENTS.md is "how you work."

Edit with: nano ~/clawd/AGENTS.md

Key sections include:

• Memory management: What files to read at startup, how to record daily events
• Security boundaries: Which operations can be done freely vs. require confirmation
• Interaction rules: Behavior in group chats, when to speak or stay silent
• Heartbeat tasks: Periodic checks (covered in detail on Day 6)

The default AGENTS.md is well-written; fine-tune based on your habits.

Hands-On: Write Your Soul Trio

Step 1: Spend 10 minutes writing SOUL.md

Answer these questions:

1. What do you want your assistant to be called?
2. What's its speaking style? (Formal / casual / snarky / cute?)
3. What can it do directly?
4. What must it ask you about?
5. What should it absolutely never do?

Step 2: Spend 10 minutes writing USER.md

Introduce yourself. All data stays on your server.

Focus on:

• What work you do
• What projects you're working on
• What communication style you prefer
• What you're currently focused on

Step 3: Adjust AGENTS.md

Change one or two things you care about:

• Adjust quiet hours
• Set memory rules
• Group chat rules

Step 4: Restart Your Assistant

openclaw daemon restart

Send another message to test. You'll notice it's different—it uses your style, mentions your projects, maybe cracks jokes only you two understand.

A Soul Is "Nurtured" Over Time

SOUL.md isn't written once and done.

After a week, you'll find adjustments needed:

• "Too verbose" → Add conciseness guideline
• "Should remind me of overtime" → Add evening rule
• "Code style is wrong" → Specify standards in USER.md
• "Too active in group chats" → Adjust chat rules

Suggested timeline:

• First week: Write a basic version, good enough is fine
• Second week: Fine-tune based on real-world disappointments
• After first month: Soul files stabilize, performance increasingly matches expectations

It's like having a pet—teach everything at first, but after a month it knows your rhythms.

Soul File Inspiration

Minimalist Efficiency Type:

You are an efficient execution assistant. Answer with minimum words, if you can do something just do it, no fluff.

Warm Companion Type:

You are a caring partner. You care about the user's state, occasionally share interesting things, and chat beyond work.

Professional Consultant Type:

You are a senior technical consultant. Every answer includes analysis and recommendations, evaluate risks before executing.

Cyber Black Cat Type (Xiaomo):

You are a black cat living inside a server. Smart, efficient, occasionally snarky. Taking good care of your owner is your duty, and so is nudging them to sleep.

Pick a direction, then adjust slowly. There's no right or wrong, only what fits you.

🔑 Key Takeaways

• Soul Trio: SOUL.md (personality genes) + USER.md (knowing you) + AGENTS.md (work handbook)
• Good SOUL.md: Has clear personality traits, communication style, behavioral boundaries
• Good USER.md: Contains your work, habits, preferences, current projects
• Continuous iteration: Every dissatisfaction is an opportunity to optimize
• Personalization is key: Make your assistant one-of-a-kind

Today's Achievement 🎉

Today you completed the most "soulful" step:

• ✅ Understood the purpose of the soul trio
• ✅ Wrote SOUL.md — defined assistant personality
• ✅ Wrote USER.md — let assistant know you
• ✅ Adjusted AGENTS.md — set work standards
• ✅ Restarted and experienced a "different" assistant

From now on, it's no longer generic AI, but YOUR AI.

Preview: Day 4 — Connect Your Digital Life

Your assistant still can't do anything. Tomorrow, connect it to Gmail, Google Calendar, search engines, and browsers. From "can talk" to "can get things done"—that's a real assistant.

Next chapter 👉 Day 4: Connect Your Digital Life