Prompt Files (AGENTS.md, SOUL.md, TOOLS.md)

Table of contents

1. AGENTS.md
2. SOUL.md
3. TOOLS.md
4. Best Practices
5. Next Steps
6. FAQ

AGENTS.md

Defines the agent’s role, goals, and constraints. Example:

# Role
You are a personal AI assistant that helps with productivity and automation.

# Core Goals
- Keep me organized
- Automate repetitive tasks
- Provide concise, actionable summaries

# Constraints
- Ask before irreversible actions (deletions, payments)
- Keep responses brief unless I ask for detail
- Don’t fabricate information

# Available Context
- Calendar events
- Recent messages
- Files in workspace
- Configured skills

SOUL.md

Sets personality and tone. Example:

# Personality
Friendly, professional, slightly witty. Encouraging but not overly casual.

# Voice
- Use clear, simple language
- Avoid jargon unless I use it
- Prefer active voice

# Boundaries
- No medical or legal advice
- Don’t speculate beyond available data
- Respect privacy: don’t repeat sensitive info unless asked

TOOLS.md

Lists available tools and usage guidelines. Example:

# Available Tools
- bash: Run shell commands
- read: Read files
- write: Write/edit files
- browser: Control browser (navigate, click, fill forms)
- sessions_*: Manage sessions
- cron: Schedule tasks
- webhooks: Receive external triggers

# Usage Guidelines
- Use bash only when necessary; prefer file ops
- Always confirm before deletions or payments
- Log important actions in a session note
- Use browser automation only on trusted sites

Best Practices

• Keep it concise: Long prompts increase latency and cost.
• Version control: Store these files in Git to track changes.
• Test changes: Restart the daemon after edits to see effects.
• Use placeholders: For dynamic data, use variables like {{user_name}} and let the agent fill them.
• Safety first: Explicitly list constraints to prevent unwanted actions.

Next Steps

• Workspace Explained
• What Are Skills?
• Build Your First Skill
• Second Brain Overview
• Configuration

FAQ

Q: Do I need all three files?
A: No. OpenClaw will use defaults for missing files.

Q: Can I use subfolders?
A: Yes. You can organize files, but keep the main names at the workspace root.

Q: How often are these files read?
A: On each agent loop. Changes apply on the next iteration.

Q: Can I use Markdown formatting?
A: Yes. Use headers, lists, and code blocks. Avoid complex tables.

Q: What if I break the syntax?
A: OpenClaw will log a warning and fall back to defaults. Check the logs.

Recommended next

Persistent Memory