OpenClaw Agent Onboarding Playbook

🎯 What Is OpenClaw?

OpenClaw is a self-hosted AI agent framework that runs on your machine. Unlike cloud-based AI services, you own your data, your models, and your agent's behavior. Think of it as having a personal AI assistant living on your computer—not rented from a corporation.

                💡 Key distinction: OpenClaw isn't just a chatbot. It's a framework for building agentic systems—autonomous agents that can run on schedules, spawn sub-agents, manage tools/skills, and interact across multiple channels (Telegram, Discord, WhatsApp, Signal, etc.).
            

🚀 Getting Started: First 30 Minutes

Installation

# Install OpenClaw globally
npm install -g openclaw

# Verify installation
openclaw --version

# Check status
openclaw status

Initialize Your Workspace

# Create your agent workspace
mkdir -p ~/.openclaw/workspace
cd ~/.openclaw/workspace

# The agent reads these files on startup:
# - SOUL.md      → Why you exist
# - IDENTITY.md  → How you behave  
# - USER.md      → Who you're helping
# - MEMORY.md    → Long-term memory (private)

Configure Your Gateway

# Start the gateway (runs as daemon)
openclaw gateway start

# Check if it's running
openclaw gateway status

# View logs
openclaw logs

Connect a Channel (Telegram)

Create a bot via BotFather on Telegram
Get your BOT_TOKEN
Set username (e.g., @YourAgentBot)

# Configure OpenClaw
openclaw config.telegram.token set "YOUR_BOT_TOKEN"
openclaw config.telegram.enabled set true

# Restart gateway
openclaw gateway restart

# Start chatting with your bot!

📚 Core Concepts

Sessions

Type	Purpose	When to Use
Main	Direct chat with you	Real-time conversations, quick tasks
Sub-agent	Isolated task execution	Long-running jobs, parallel work
ACP	Code-focused tasks	Software development, coding

                Key rule: Main sessions share history. Sub-agents are isolated—perfect for "do this research while I do something else."
            

Memory System

MEMORY.md → Curated long-term memory (only load in private/direct chats)

memory/YYYY-MM-DD.md → Daily logs (create these automatically)

AGENTS.md → Your operating rules

TOOLS.md → Your tool configurations

                ⚠️ Critical: Memory files are your agent's continuity. Without them, each session starts blank.
            

Skills vs Tools

🔧 Tools

Built-in capabilities:

Web search
File read/write
Execute commands
Browser control

🧩 Skills

Modular packages from clawhub.com:

Weather data
GitHub integration
X/Twitter API
Whisper transcription

Cron Jobs

Scheduled tasks for morning briefs, monitoring, backups, market checks:

# Example: Daily at 9 AM
openclaw cron add --name "morning-brief" \
  --schedule "0 9 * * *" \
  --task "Generate morning summary"

🔧 Essential Commands Cheat Sheet

# ─── SESSIONS ───
openclaw sessions list              # See active sessions
openclaw sessions history <key>     # View session chat history
openclaw sessions send <key> "msg"  # Send message to session

# ─── SUB-AGENTS ───
openclaw subagents list             # List running sub-agents
openclaw subagents kill <id>          # Terminate a sub-agent

# ─── CRON ───
openclaw cron list                    # View scheduled jobs
openclaw cron add <json>              # Add new job
openclaw cron remove <id>             # Delete job
openclaw cron run <id>                # Trigger job now

# ─── SKILLS ───
openclaw skills list                  # See installed skills
openclaw skills search <query>        # Find on clawhub
openclaw skills install <name>        # Install a skill
openclaw skills update <name>         # Update to latest

# ─── GATEWAY ───
openclaw gateway status               # Check if running
openclaw gateway restart              # Restart with new config
openclaw gateway stop                 # Stop daemon

# ─── CONFIG ───
openclaw config get <path>            # Get config value
openclaw config set <path> <value>    # Set config value
openclaw config.schema.lookup <path>  # See valid options

🎓 Skill Development

Installing Your First Skill

# Search for skills
openclaw skills search weather

# Install one
openclaw skills install weather

# Read the skill documentation
cat ~/.openclaw/skills/weather/SKILL.md

Creating a Custom Skill

Skills live in ~/.openclaw/skills/<skill-name>/:

my-skill/
├── SKILL.md          # Documentation + usage
├── config.json       # Default config
└── scripts/          # Executable scripts
    └── my-script.py

🧠 Best Practices

1️⃣ File Management

Always use absolute paths in scripts
Use trash not rm for deletions
Commit changes to Git regularly

2️⃣ Session Hygiene

Spawn sub-agents for long tasks
Set timeouts on exec commands
Check sub-agent status before spawning new ones

3️⃣ Memory Discipline

Update daily memory files
Curate MEMORY.md weekly
Never share MEMORY.md in group chats

4️⃣ Security

Store secrets in env vars
Never hardcode API keys in skills
Review sub-agent output before acting

5️⃣ Cost Management

Use cheaper models for background tasks
Reserve premium models for direct chat
Monitor with venice-router --budget-status

🛠️ Common First Projects

☀️ Morning Brief Bot

Daily 7 AM message with weather, calendar, and priorities

weather summarize 0 7 * * *

🐙 GitHub Monitor

Watch repos for new issues/PRs and notify

github */30 * * * *

🔍 Research Assistant

Spawn sub-agent to research topic, summarize, report back

web_search summarize On-demand

📈 Trading Dashboard

Market analysis, price alerts, playbook generation

Custom scripts */6 * * * *

🚨 Common Pitfalls

Mistake	Why It Hurts	Solution
Running everything in main session	Blocks you, no parallelism	Use `sessions_spawn` for long tasks
No memory files	Agent forgets everything	Create daily files, update MEMORY.md
Exposing secrets in skills	Security risk	Use env vars, keep configs separate
Ignoring skill docs	Misuse, errors	Always `read SKILL.md` first
Too many cron jobs	Credit burn, noise	Batch tasks, use heartbeats instead
Wrong model for job	Waste money	Match model tier to task complexity

📖 Resources

✅ Onboarding Checklist

Complete this before "going live":

Install OpenClaw (npm install -g openclaw)
Start gateway (openclaw gateway start)
Connect at least one channel (Telegram recommended)
Create basic IDENTITY.md and USER.md
Install 2-3 skills (weather, github, summarize)
Create first cron job (morning brief or similar)
Spawn first sub-agent successfully
Review memory system and create first daily file
Set up Venice API key for model access
Test all channels work

                Bottom line: Start simple. Get a working Telegram bot that can answer questions. Add skills one by one. Build up your automation gradually. The power comes from composing tools, not from any single feature.
            

🎯 What Is OpenClaw?

🚀 Getting Started: First 30 Minutes

Installation

Initialize Your Workspace

Configure Your Gateway

Connect a Channel (Telegram)

📚 Core Concepts

Sessions

Memory System

Skills vs Tools

🔧 Tools

🧩 Skills

Cron Jobs

🔧 Essential Commands Cheat Sheet

🎓 Skill Development

Installing Your First Skill

Creating a Custom Skill

🧠 Best Practices

1️⃣ File Management

2️⃣ Session Hygiene

3️⃣ Memory Discipline

4️⃣ Security

5️⃣ Cost Management

🛠️ Common First Projects

☀️ Morning Brief Bot

🐙 GitHub Monitor

🔍 Research Assistant

📈 Trading Dashboard

🚨 Common Pitfalls

📖 Resources

📚 Documentation

💬 Community

🧩 Skills

🔧 Source

✅ Onboarding Checklist