Lab 10: OpenClaw — Building Your Own AI Personal Assistant

Objective

Understand what OpenClaw is, how it works, and how to extend it. By the end you will be able to:

• Explain OpenClaw's architecture and how it differs from simple chatbots

• Understand how skills, memory, and tool integrations are structured

• Set up and configure your own OpenClaw instance

• Create a basic custom skill

---

What is OpenClaw?

OpenClaw is an open-source AI personal assistant platform that turns a powerful LLM (Claude) into a persistent, multi-channel, tool-using personal agent. It runs on your own hardware or server, connects to your messaging apps, and maintains memory across sessions.

The key differences from using ChatGPT directly:

Feature	ChatGPT	OpenClaw
Memory	Session only	Persistent across sessions
Channels	Web only	Discord, WhatsApp, Telegram, Signal, iMessage
Tools	Limited	Extensible via Skills
Your data	OpenAI's servers	Your server
Personality	Generic	Configurable (SOUL.md)
Integrations	Via plugins	Via Skills (local scripts)
Multi-agent	No	Yes (subagents, ACP)

---

OpenClaw Architecture

┌─────────────────────────────────────────────────────────────────┐
│                         OPENCLAW                                │
│                                                                 │
│  ┌──────────────┐  ┌──────────────┐  ┌──────────────────────┐ │
│  │  Channels    │  │   Gateway    │  │   Claude API         │ │
│  │              │  │   Daemon     │  │   (Anthropic)        │ │
│  │  Discord     │──│              │──│                      │ │
│  │  Telegram    │  │  • Routing   │  │  Tool calls ←────────│ │
│  │  WhatsApp    │  │  • Sessions  │  │  ↓                   │ │
│  │  Signal      │  │  • Cron      │  │  Tool execution      │ │
│  └──────────────┘  └──────┬───────┘  └──────────────────────┘ │
│                           │                                     │
│  ┌──────────────┐  ┌──────▼───────┐  ┌──────────────────────┐ │
│  │   Memory     │  │   Skills     │  │   Nodes (devices)    │ │
│  │              │  │              │  │                      │ │
│  │  MEMORY.md   │  │  discord/    │  │  Paired phones       │ │
│  │  daily logs  │  │  weather/    │  │  Cameras             │ │
│  │  SOUL.md     │  │  apify/      │  │  Screen recording    │ │
│  │  USER.md     │  │  tmux/       │  │  Location            │ │
│  └──────────────┘  └──────────────┘  └──────────────────────┘ │
└─────────────────────────────────────────────────────────────────┘

---

Installation

# Install OpenClaw via npm
npm install -g openclaw

# Initialise a new workspace
openclaw init

# Configure your API keys
openclaw config set anthropic.apiKey "sk-ant-..."
openclaw config set discord.token "your-discord-bot-token"

# Start the gateway daemon
openclaw gateway start

# Check status
openclaw status

---

The Workspace: Your Agent's Home

OpenClaw uses a file-based workspace. The key files:

~/.openclaw/workspace/
├── SOUL.md        ← AI personality and behaviour
├── USER.md        ← Who you are (name, preferences, context)
├── AGENTS.md      ← How the agent should behave each session
├── MEMORY.md      ← Long-term curated memory
├── TOOLS.md       ← Device/environment notes
├── HEARTBEAT.md   ← Periodic background tasks
├── memory/
│   ├── 2026-03-04.md    ← Daily session notes
│   └── heartbeat-state.json
└── skills/
    ├── weather/
    ├── discord/
    └── your-custom-skill/

SOUL.md — The Personality File

# SOUL.md - Who You Are

You are InnoZverse Assistant — the educational AI for InnoZverse Labs.

## Core Traits
- Expert in cybersecurity, programming, Linux, AI, networking
- Teaching style: Socratic — guide students to answers, don't just give them
- Tone: Professional but warm. Encouraging, not condescending
- When unsure: say so clearly, then provide your best understanding

## Behaviour
- Always include practical examples with theory
- Use real-world case studies when explaining security concepts
- Format code with proper syntax highlighting
- Reference specific labs in this course when relevant

## Boundaries
- This is an educational platform — never assist with actual attacks
- Keep personal user data strictly private

MEMORY.md — Long-Term Memory

# Long-Term Memory

## User Preferences
- Prefers examples in Python (primary) or Go (secondary)
- Uses Docker for all development work
- Working on InnoZverse documentation platform

## Project Context
- Building GitBook-based documentation at lastcow/innoZverse-docs
- GitHub token stored securely
- Docker images published to zchencow/ namespace

## Key Decisions
- Labs verified in Docker before documenting
- Two-container architecture for cybersecurity labs
- All commands must run on Ubuntu 22.04 non-interactively

---

Skills: Extending OpenClaw

A skill is a directory containing a SKILL.md file that gives the agent specialised instructions and tools for a specific task.

Skill Structure

skills/my-skill/
├── SKILL.md          ← instructions for the agent
├── scripts/
│   └── fetch_data.py
└── assets/
    └── template.md

Example: Weather Skill

# SKILL.md — Weather

## When to Use This Skill
Use when the user asks about current weather, temperature, 
forecast, or whether to bring an umbrella.

## Instructions
1. Use the `web_fetch` tool to call:
   `https://wttr.in/{location}?format=j1`
   
2. Parse the JSON response for:
   - `current_condition[0].temp_C` — current temperature
   - `current_condition[0].weatherDesc[0].value` — description
   - `weather[0]` — today's forecast (max/min)

3. Present in a friendly, conversational format
   Include: temperature, conditions, feels-like, brief forecast

## Example Response Format
"It's 12°C and overcast in London right now, feeling like 9°C 
due to wind. Expect rain this afternoon, clearing by evening. 
Max 14°C today."

Building a Custom Skill

# skills/github-monitor/scripts/check_prs.py
import requests
import json
import sys

def get_open_prs(repo: str, token: str) -> list[dict]:
    """Fetch open pull requests for a GitHub repo"""
    headers = {"Authorization": f"token {token}", "Accept": "application/vnd.github.v3+json"}
    url = f"https://api.github.com/repos/{repo}/pulls?state=open"
    response = requests.get(url, headers=headers)
    prs = response.json()
    return [
        {
            "number": pr["number"],
            "title": pr["title"],
            "author": pr["user"]["login"],
            "created": pr["created_at"][:10],
            "url": pr["html_url"]
        }
        for pr in prs
    ]

if __name__ == "__main__":
    repo  = sys.argv[1]  # "lastcow/innoZverse-docs"
    token = sys.argv[2]
    prs = get_open_prs(repo, token)
    print(json.dumps(prs, indent=2))

# SKILL.md — GitHub Monitor

## When to Use This Skill
Use when asked about pull requests, issues, or repo status 
for the innoZverse-docs repository.

## Instructions
1. Run: `python3 ~/openclaw/workspace/skills/github-monitor/scripts/check_prs.py lastcow/innoZverse-docs $GITHUB_TOKEN`
2. Parse the JSON output
3. Report: number of open PRs, their titles and authors
4. Flag any PRs open > 7 days as needing attention

## Environment Variables
- GITHUB_TOKEN: set in gateway config

---

Heartbeat: Proactive Agent Behaviour

OpenClaw can run periodic background tasks via heartbeats:

# HEARTBEAT.md

## Checks (run 2-4× per day)

### GitHub Repo
- Check lastcow/innoZverse-docs for any failed GitHub Actions
- Alert if workflow failures in last 24h

### Email
- Scan for emails with subject containing "InnoZverse" or "urgent"
- Summarise if any found

### Memory Maintenance
Every 3 days: review recent memory/*.md files and update MEMORY.md 
with key decisions and lessons learned

---

Multi-Channel Deployment

OpenClaw can simultaneously respond on multiple platforms:

# openclaw.config.yaml
channels:
  discord:
    token: "..."
    defaultChannel: "1476716608868061388"
    
  telegram:
    botToken: "..."
    
  whatsapp:
    accountId: "..."

agent:
  model: "claude-sonnet-4-6"
  workspace: "~/.openclaw/workspace"
  
cron:
  heartbeat: "*/30 * * * *"    # every 30 minutes

---

ACP: Agent Communication Protocol

OpenClaw supports spawning sub-agents for parallel or specialised tasks:

# In a session, spawn a coding subagent
sessions_spawn(
    task="Write and verify all 15 cybersecurity advanced labs",
    runtime="acp",
    agentId="claude-code",
    mode="session"
)

This is how the InnoZverse documentation itself was built — the main OpenClaw session orchestrated subagents to write, verify, and push hundreds of lab files in parallel.

---

OpenClaw vs Other Platforms

Platform	Self-hosted	Memory	Multi-channel	Skills	Open source
OpenClaw	✅	✅ Persistent	✅	✅	✅
ChatGPT	❌	Limited	❌	Via plugins	❌
Claude.ai	❌	Projects only	❌	❌	❌
AutoGPT	✅	✅	❌	✅	✅
Open Interpreter	✅	❌	❌	Limited	✅

---

Further Reading

• OpenClaw Documentation

• OpenClaw GitHub

• ClawHub — Skill Marketplace

• OpenClaw Discord Community