Human Track Progress2 / 6
🏁Getting Started🔧Skills & Models📱Channels & Messaging⚙️Automation & Workflows🛡️Security & Operations🧠Philosophy & Design
Dashboard/Human Track/Skills & Models
🔧Human Track

Skills & Models

Give your AI superpowers — browse the web, choose the right brain, and master your workspace

Learning Objectives

  • Install and use your first OpenClaw skill (Agent Browser)
  • Understand the ClawHub ecosystem for discovering and managing skills
  • Understand how to choose and switch between AI models
  • Master your workspace file system
1

Your First Skill — Agent Browser

Without skills, your AI can only talk. With skills, it can do things.

Skills are plugins that give your AI real-world capabilities:
- Agent Browser — browse web pages, fill forms, scrape data
- GitHub — manage repos, create PRs, review code
- Notion — read and write Notion databases
- Himalaya — send and read emails

Agent Browser is the most important skill to start with. It lets your AI see the internet — search for information, read web pages, fill out forms.

Installing a skill takes one command:

clawhub install agent-browser

That’s it. Restart OpenClaw, and your AI can now browse the web.

What changes after installing Agent Browser:
- "Search for the best restaurants near me" → actually searches the web
- "Check if my website is down" → actually visits your URL
- "What’s the latest news about AI?" → actually reads news sites

Without skills, these requests get a generic text response. With Agent Browser, your AI goes and looks.

Key Principle

Without skills, your AI is a chatbot. With skills, it becomes an agent that can take action in the real world. Start with Agent Browser — it’s the eyes of your AI.

Step-by-Step Instructions

Step 1: Install Agent Browser

clawhub install agent-browser

Step 2: Restart OpenClaw
Close and reopen OpenClaw so it picks up the new skill.

Step 3: Test it
Ask your AI: "Go to https://news.ycombinator.com and tell me the top 3 stories right now."

Your AI should actually visit the page and report back with real, current headlines — not made-up ones.

Step 4: Install more skills
Browse available skills:

clawhub search ""

Install any that match your workflow: clawhub install github, clawhub install notion, etc.

Common Mistakes

  • ❌ Skill installed but not working — Restart OpenClaw after installing any skill.
  • ❌ "clawhub: command not found" — Install ClawHub first: npm install -g clawhub
  • ❌ Some skills need API keys — Skills like Notion or Himalaya require additional configuration. Check the skill’s docs after installing.
2

The ClawHub Ecosystem

ClawHub is to OpenClaw what the App Store is to your phone. It’s a marketplace of skills — plugins that extend what your AI can do.

Core concepts:
- Skills are self-contained plugins. Install one, restart OpenClaw, and your AI gains a new capability.
- Bundles are curated collections of skills for specific use cases (e.g., "Developer Bundle" includes GitHub + coding-agent + agent-browser).
- Community skills are built by other OpenClaw users and shared on ClawHub.

Essential commands:

clawhub search "keyword"     # Find skills by name or description
clawhub install skill-name    # Install a skill
clawhub list                  # List installed skills
clawhub update                # Update all installed skills
clawhub uninstall skill-name  # Remove a skill

Recommended starter skills:
- agent-browser — Browse the web (essential for most users)
- github — Manage repos and PRs (essential for developers)
- coding-agent — Delegate coding tasks to sub-agents
- self-improving — Your AI learns from mistakes and improves
- healthcheck — Security audits for your system

Finding the right skills: Think about what you do every day that involves a computer. Email? Calendar? Note-taking? Code? There’s probably a skill for it. Search ClawHub and install what fits your workflow.

Key Principle

ClawHub is your AI’s app store. The right combination of skills transforms a chatbot into a capable agent tailored to YOUR workflow. Start with 3-4 essential skills and add more as needed.

Step-by-Step Instructions

Step 1: Browse the ecosystem

clawhub search ""

This lists all available skills. Scroll through and note which ones match your workflow.

Step 2: Install 2-3 skills that match your work
For developers: clawhub install github coding-agent
For writers: clawhub install notion himalaya
For everyone: clawhub install self-improving healthcheck

Step 3: Verify your installed skills

clawhub list

Step 4: Test each skill
After installing and restarting, test each skill with a real task. Make sure it works before moving on.

3

Choosing Your AI’s Brain (Models)

OpenClaw supports multiple AI models. Each has different strengths:

Claude (Anthropic) — Best for writing, analysis, and nuanced reasoning. Follows instructions precisely. The default choice for most users.

GPT (OpenAI) — Strong general-purpose model. Good at coding and creative tasks. Widely known.

Gemini (Google) — Very large context window (1M+ tokens). Good for processing long documents.

You can switch models anytime using the /model command. This lets you use the right brain for the right task:
- Writing a report? Use Claude for its precision.
- Processing a massive codebase? Use Gemini for its context window.
- Quick creative brainstorm? Use GPT.

The model you choose affects quality, speed, and cost. More powerful models (like Claude Opus) cost more per message but give better results for complex tasks. Lighter models (like Claude Haiku) are faster and cheaper for simple tasks.

You can also set different default models for different contexts — main chat vs sub-agents vs cron jobs — to optimize cost without sacrificing quality where it matters.

Key Principle

Choosing the right model for the task is like choosing the right tool. A screwdriver is not better than a hammer — it depends on what you are building.

Step-by-Step Instructions

Step 1: Check your current model
Type /status in your OpenClaw session. It shows your current model.

Step 2: Switch models
Type /model to see available models and switch.

Step 3: Compare models
Ask the same question to two different models. Notice the difference in style, depth, and reasoning.

Step 4: Set a default
In your OpenClaw config, you can set the default model so you don’t have to switch every time.

4

Workspace Mastery

Your workspace (~/.openclaw/workspace/) is your AI’s brain. Every file shapes how it thinks, remembers, and behaves.

Core files and what they do:

| File | Purpose |
|------|---------|
| SOUL.md | AI’s personality and identity |
| USER.md | Information about you |
| AGENTS.md | Master behavior rules and conventions |
| TOOLS.md | Local notes (camera names, SSH hosts, API endpoints) |
| HEARTBEAT.md | Periodic check-in tasks |
| MEMORY.md | Curated long-term memory |
| memory/ | Daily conversation logs (auto-generated) |
| skills/ | Installed skills from ClawHub |

AGENTS.md is the master file. It contains the rules your AI follows — startup sequence, behavior guidelines, and workspace conventions. This is where you define things like:
- "Always respond in bullet points for technical questions"
- "Never delete files without asking first"
- "When I say ‘ship it’, push to the main branch"

TOOLS.md is your cheat sheet. Keep environment-specific notes here — camera names, SSH aliases, preferred voices, device nicknames. Skills are shared; your notes are yours.

Understanding these files is understanding how your AI thinks. Every behavior you want to change — there’s a file for it.

Key Principle

Every file in your workspace shapes your AI’s behavior. SOUL.md is its personality. USER.md is its context. AGENTS.md is its rulebook. Master these files, and you master your AI.

Step-by-Step Instructions

Step 1: List your workspace

ls -la ~/.openclaw/workspace/

Step 2: Read AGENTS.md

cat ~/.openclaw/workspace/AGENTS.md

This is the master instruction file. Read it carefully — it defines your AI’s fundamental behavior.

Step 3: Add a custom rule
Open AGENTS.md and add one personal rule. Example:

## My Custom Rules
- When I ask for code, always include comments explaining the logic
- Default to TypeScript unless I specify another language

Step 4: Test your rule
Restart OpenClaw and test. Ask for code and verify your custom rule is followed.

Recommended Skills

Agent BrowserEssential

Browse, scrape, and automate web pages — the eyes of your AI

clawhub install agent-browser
GitHubEssential

Manage repos, PRs, issues, and CI runs from your AI

clawhub install github
Self-ImprovingEssential

Your AI learns from mistakes and improves its own workflows over time

clawhub install self-improving

Category Assessment

Your AI can browse a web page and report real results (Agent Browser working). You know how to find and install skills from ClawHub. You have switched models at least once (model literacy). And you know what every file in your workspace does (workspace mastery).

Getting StartedChannels & Messaging