# OpenClaw Prompting Guide: How to Talk to Your AI Agent

OpenClaw is not ChatGPT. You don't write prompts - you shape an agent. This guide covers SOUL.md, AGENTS.md, USER.md, memory, and the prompting techniques that actually matter.

## Why OpenClaw Prompting Is Different

In ChatGPT, you write a prompt, get a response, and start over. OpenClaw uses persistent identity files that load every session. Instructions compound over time instead of resetting.

| Aspect | ChatGPT / Claude | OpenClaw |
|--------|-----------------|----------|
| Instructions | System prompt (resets) | Persistent files (SOUL.md, AGENTS.md) |
| Memory | Limited or none | MEMORY.md + daily files |
| Your context | Repeat every time | USER.md (reads automatically) |
| Personality | Generic | SOUL.md (fully customizable) |
| Tools | Fixed plugins | Skills (SKILL.md files) |
| Improvement | Starts fresh | Compounds across sessions |

## The Four Files That Shape Your Agent

### SOUL.md - Who your agent is
Your agent's identity. Personality, tone, values, boundaries. Loaded every session.

### AGENTS.md - How your agent operates
Operating manual. Rules for memory, safety, file handling, group chats, external actions.

### USER.md - Who you are
Your name, timezone, preferences, context. Agent reads this to personalize responses.

### MEMORY.md - What your agent remembers
Long-term memory that persists across sessions. Agent reads and updates it automatically.

## Prompting Techniques That Actually Work

1. **Be specific about output format** - "Summarize in 3 bullet points with one key stat"
2. **Give context, not just commands** - include audience, tone, length constraints
3. **Use "remember this"** - agent writes it to memory for next session
4. **Chain tasks** - "Research competitors, summarize pricing, save to file"
5. **Tell it what NOT to do** - "No marketing language. No filler."

## SOUL.md Examples by Use Case

| Use Case | Personality | Key Instructions |
|----------|------------|-----------------|
| Developer | Technical, concise | "Prefer TypeScript. No verbose explanations." |
| Business | Professional, structured | "Use bullet points. Include action items." |
| Creative | Expressive, experimental | "Take risks. No cliches." |
| Crypto | Data-driven, cautious | "Include risk factors. Cite sources." |
| Content | Engaging, audience-aware | "Write for humans, not algorithms." |

## Skills: Task-Specific Prompts

For recurring tasks, create custom skills. A skill is a SKILL.md file with task-specific instructions that loads when relevant.

## Common Prompting Mistakes

- Being too vague
- Overloading SOUL.md (start with 10-15 lines)
- Not using memory
- Treating it like a search engine
- Ignoring USER.md
- Not iterating

## Advanced: Cron Prompts and Automation

Set up cron jobs with prompts that run on schedule:
- Morning brief: emails, calendar, alerts
- End of day: memory file with open tasks
- Weekly review: update MEMORY.md

## Quick Start

1. Edit SOUL.md - 3-5 lines about personality
2. Edit USER.md - name and timezone
3. Start chatting
4. Say "remember this" for important context
5. Iterate - update SOUL.md when needed

## FAQ

**What is SOUL.md?**
Your agent's identity - personality, tone, behavior. Loaded every session.

**How is it different from ChatGPT?**
Persistent files that compound over time vs system prompts that reset.

**Can I change the personality?**
Yes. Edit SOUL.md. Changes take effect next session.

---

Try OpenClaw: https://www.ampere.sh/setup