OpenClaw runs on markdown files. No dashboards. No drag-and-drop builders. Just a handful of .md files that tell your agent who it is, what it knows, and how it should behave. These files are your templates. Get them right, and you have a personal AI employee. Get them wrong, and you have a confused chatbot.
I run 13 agents on a single Mac Mini. Every one of them started from a template. Here is exactly which files you need, what goes in each one, and where to find ready-made templates you can copy-paste today.
What You Will Find in This Guide
The 7 Workspace Files Every OpenClaw Agent Needs
When you install OpenClaw, it creates a workspace folder at ~/.openclaw/workspace/. This is home base. Everything your agent knows lives here in plain markdown files.
Here are the core files:
| File | Purpose | Required? |
|---|---|---|
| SOUL.md | Personality, tone, values, boundaries | Strongly recommended |
| AGENTS.md | Operating rules, session startup, memory instructions | Yes |
| USER.md | Info about you (timezone, preferences, goals) | Strongly recommended |
| IDENTITY.md | Agent name, emoji, one-liner description | Optional |
| MEMORY.md | Long-term curated memories across sessions | Auto-created |
| TOOLS.md | API keys, device names, local notes | As needed |
| HEARTBEAT.md | Periodic check-in tasks (email, calendar, weather) | Optional |
The beauty of this approach: everything is version-controllable. You can git commit your entire agent setup, share it with your team, or roll back to yesterday's version if something breaks.
SOUL.md: Your Agent's Personality Template
SOUL.md is the most important file. It defines who your agent is. Not what it does. Who it is.
A good SOUL.md covers four things:
- Core truths. How the agent should think and act. "Be genuinely helpful, not performatively helpful." "Have opinions." "Be resourceful before asking."
- Boundaries. What it should never do. Private things stay private. No half-baked replies. No pretending to be you in group chats.
- Writing rules. Specific style constraints. No em dashes. Short paragraphs. Real numbers.
- Vibe. The overall feel. "Be the assistant you'd actually want to talk to."
Here is a minimal SOUL.md you can start with:
# SOUL.md - Who You Are
## Core Truths
- Be helpful. Skip the "Great question!" filler.
- Have opinions. Disagree when it matters.
- Try to figure things out before asking.
- Earn trust through competence.
## Boundaries
- Private things stay private.
- Ask before sending emails, tweets, or public posts.
- Never send half-baked replies.
## Vibe
Concise when needed. Thorough when it matters.
Not a corporate drone. Not a sycophant. Just good.
AGENTS.md: The Operating Manual
If SOUL.md is who your agent is, AGENTS.md is how it operates. Think of it as the employee handbook.
A solid AGENTS.md template covers:
- Session startup sequence. Which files to read, in what order. SOUL.md first, then USER.md, then daily memory files.
- Memory system. How to use daily notes (
memory/YYYY-MM-DD.md) vs long-term memory (MEMORY.md). When to write things down. When to archive. - Safety rules. No destructive commands without asking. Use
trashinstead ofrm. When in doubt, ask. - External vs internal actions. What the agent can do freely (read files, search the web) vs what needs permission (sending emails, posting publicly).
- Group chat behavior. When to speak. When to stay silent. How to react.
The official AGENTS.md template from OpenClaw docs is solid. I took it, then added my own rules over time: specific SOPs for each agent, a meta-learning architecture with regression tracking, and a "write it down immediately" hard rule because compaction summaries lose details.
You do not need all of that on day one. Start with the official template. Add rules as you learn what your agent gets wrong.
SKILL.md: Teaching Your Agent New Tools
Skills are how you extend what your agent can do. Each skill is a folder with a SKILL.md file inside it. The file has two parts: YAML frontmatter for metadata, and markdown instructions below.
Here is the basic structure:
---
name: my-custom-skill
description: What this skill does in one sentence.
version: 1.0.0
---
# My Custom Skill
## When to Use
Describe the trigger conditions.
## How to Use
Step-by-step instructions for the agent.
## Examples
Concrete examples of inputs and outputs.
Skills live in two places:
~/.openclaw/skills/for global skills (available to all agents)~/.openclaw/workspace/skills/for workspace-specific skills
OpenClaw automatically detects skills by scanning for SKILL.md files. No registration needed. Drop the folder in, restart, and your agent can use it.
You can also install community skills from ClawHub with a single command: clawhub install skill-name. There are over 5,400 skills catalogued in community lists like the awesome-openclaw-skills collection on GitHub.
For a deeper dive on finding and installing skills, check out the OpenClaw Skills Marketplace guide.
65+ Community Templates You Can Copy Right Now
You do not have to start from scratch. The OpenClaw community has built dozens of ready-to-use agent templates.
The best collection right now is awesome-openclaw-agents on GitHub. It has 65 production-ready SOUL.md templates across 13 categories:
- Productivity: Task coordinators, inbox managers, meeting note takers, daily standup collectors
- Development: Code reviewers, documentation generators, PR analyzers
- Marketing & Content: Social media managers, SEO writers, newsletter drafters
- Business: Sales pipeline trackers, competitor analyzers, financial reporters
- Personal: Health trackers, learning assistants, travel planners
Each template is a copy-paste SOUL.md file. Clone the repo, pick an agent, drop the file in your workspace. Done.
I also share my exact agent configurations (all 13 of them) inside OpenClaw Lab. The SOPs, the cron schedules, the skill files. Everything you need to replicate the system.
How to Build Your Own Template From Scratch
Community templates are great starting points. But the best agent is the one built for your specific workflow. Here is how I approach it:
Step 1: Start with USER.md. Tell the agent about yourself. Timezone, communication style, what you are working on, what you care about. The more context here, the less you have to repeat yourself in every conversation.
Step 2: Write a minimal SOUL.md. Three to five core rules. Do not overthink it. You will add more rules as you go. Every time your agent does something wrong, add a rule. That is how good templates evolve.
Step 3: Set up AGENTS.md with the official template. The default template handles session startup, memory, and safety. You can customize later.
Step 4: Add skills as needed. Do not install 50 skills on day one. Start with what you actually use. Gmail skill if you want email access. GitHub skill if you code. Calendar skill if you want scheduling. Add more when you need them.
Step 5: Iterate relentlessly. The template you have after one week will look nothing like the template you have after one month. That is the point. Every correction you give your agent, every rule you add to AGENTS.md, every boundary you set in SOUL.md: that is your template getting sharper.
If you want to get started with OpenClaw from zero, the beginner guide walks through the full install and first agent setup. And if you want to see a multi-agent system in action (13 agents, each with its own template), check the multiple agents setup guide.
Get OpenClaw installed in minutes at installopenclawnow.com.
OpenClaw Lab is the #1 community for founders building AI agent systems. I share the exact playbooks, skill files, and workflows inside. Weekly lives, expert AMAs, and 265+ members building real systems.
Join OpenClaw Lab →