If you've just installed OpenClaw and stared at the terminal wondering "okay, now what?" — this guide is for you. Out of the box, OpenClaw is already powerful: it can chat, run shell commands, browse the web, and read and write files. But the moment you want your agent to actually do something specific — manage your GitHub PRs, draft emails from your Gmail, query a database — you need skills.
That's where ClawHub comes in.
What Even Is a Skill?
Before we touch the CLI, a quick mental model that will save you a lot of confusion later.
OpenClaw has two distinct concepts that beginners routinely mix up: Tools and Skills.
Tools are the agent's organs — the raw capabilities it physically has. The read and write tools give it file access. exec lets it run shell commands. browser lets it click buttons and fill forms. Without a tool enabled, the capability simply doesn't exist.
Skills, on the other hand, are instruction manuals. A skill teaches OpenClaw how to combine its tools to accomplish a task. The obsidian skill teaches it how to organize notes. The github skill teaches it how to interact with repos. Installing a skill does not grant new permissions — if the write tool is disabled, even the most detailed file-management skill will do nothing.
Three conditions must be met for a skill to actually work:
-
Configuration — the relevant tool is enabled (e.g.,
execfor running commands) - Installation — any bridge software the skill depends on is present
- Authorization — the skill has credentials for the external service it integrates with
Keep that mental model handy. It'll explain a lot of "why isn't this working?" moments later.
What Is ClawHub?
ClawHub (clawhub.ai) is the official public registry for OpenClaw skills and plugins — think of it as npm, but for AI agent capabilities. It hosts over 13,000 community-built skills, uses vector search (so natural language queries like "email automation" work better than exact package names), and tracks usage signals like stars and download counts to help surface quality work.
Every skill on ClawHub is a versioned bundle: primarily a SKILL.md file with instructions and metadata, plus any supporting configs or scripts. Because it's open by default, anyone with a GitHub account older than one week can publish — which means the quality varies, and you should treat skill installation the same way you'd treat installing an npm package from an unknown author.
Step 1: Install OpenClaw
If you haven't set up OpenClaw yet, the fastest path is:
npm install -g @openclaw/openclaw
Requirement: Node.js ≥ 22.16.0
Then run the guided setup wizard:
openclaw onboard --install-daemon
The wizard walks you through configuring your LLM provider (Anthropic, OpenAI, local Ollama, etc.), setting up the Gateway as a background service, and connecting your first channel. The recent 2026.4 releases cleaned up the onboarding flow significantly — there's a loading spinner during the model catalog fetch now, so the wizard no longer goes blank mid-setup.
Security note before you proceed: By default, the Gateway binds to
0.0.0.0and listens on port18789. If you're deploying on a VPS, lock this down with a firewall rule immediately — past security incidents have involved publicly exposed instances. Don't skip this step.
Step 2: Install the ClawHub CLI
The native openclaw CLI handles skill installs for most use cases. But if you want to publish skills, manage registry authentication, or sync your workspace, you'll want the separate ClawHub CLI as well:
npm install -g clawhub
clawhub login # authenticates via GitHub OAuth
For day-to-day skill management, you can use either interface. The native OpenClaw commands install into your active workspace and record source metadata, so future updates know to check ClawHub automatically.
Step 3: Find Skills
From the terminal:
openclaw skills search "email automation"
# or
clawhub search "file storage persistent"
ClawHub's vector search means you're querying by meaning, not exact keywords. Broad terms work well.
From the browser:
Head to clawhub.ai and browse by category or search naturally. Each listing shows the SKILL.md contents, version history, install count, and community feedback. Read the SKILL.md before installing — it tells you exactly what the skill does, what tools it requires, and what credentials it needs.
Before installing a third-party skill, quickly check:
- Does the GitHub repo exist and look maintained?
- Does the
SKILL.mddeclare its dependencies clearly? - Does it have a reasonable install count and community feedback?
Since February 2026, ClawHub runs automated VirusTotal scanning on every published skill, but that's a floor, not a ceiling. Reviewing before installing is still recommended.
Step 4: Install a Skill
# Using native OpenClaw CLI (recommended — records source metadata for updates)
openclaw skills install github
# Using ClawHub CLI
clawhub install steipete/github
Skills install to ~/.openclaw/skills/ by default, with each skill in its own subdirectory. You can inspect the files at any time to see exactly what instructions are being loaded.
After installing, restart OpenClaw — skills only load at session start:
openclaw restart # or restart the daemon via your system service manager
Step 5: A Practical Starter Stack
Rather than installing skills at random, a focused set of 3–5 skills covering your core workflows is more effective than a sprawling collection. Here's a good starting point for developers:
Persistent Storage — clawhub install fastio
Gives your agent cloud-based file storage with search. Without something like this, everything the agent creates vanishes when the session ends. After installing, start a chat and say "Set up Fastio storage for me" — it opens a browser window for auth and handles the rest.
GitHub — openclaw skills install github (or clawhub install steipete/github)
PR reviews, CI status checks, issue management via the gh CLI. If your work involves code, this one earns its keep fast.
Agent Browser — clawhub install TheSethRose/agent-browser
Web navigation, form filling, and screenshots using headless Chromium. Good for any workflow that involves scraping or automating web interfaces.
Google Workspace (Gog) — openclaw skills install gog
Integrates Gmail, Calendar, Tasks, Drive, Docs, and Sheets under one skill. Once authorized, the agent has access to your entire Google account through those services — so only enable this if you're comfortable with that scope.
For non-developer workflows, swap GitHub and Agent Browser for agentmail (email) and notion.
Step 6: Keep Skills Updated
Skills are actively maintained by their authors — bug fixes, new features, and compatibility patches ship regularly. A one-liner updates everything:
clawhub update --all
To see what's installed and whether updates are available:
clawhub list
If an update breaks something, roll back:
clawhub install <slug>@<previous-version>
This overwrites the current version with the one you specify. The registry keeps full version history, so you're never stuck.
Step 7: Build and Publish Your Own Skill
This is where things get genuinely interesting. A skill is just two required pieces:
1. A SKILL.md file — the instruction set your agent reads. Written in Markdown with a YAML frontmatter block declaring metadata:
---
name: my-jira-skill
description: Helps OpenClaw query and update Jira issues
tags: [jira, project-management, issues]
---
# Jira Skill
Use this skill to search for Jira issues, update their status, and add comments.
## Tools Required
- exec (to run the Jira CLI)
## Setup
1. Install the Jira CLI: `npm install -g jira-cli`
2. Run `jira login` to authenticate
## Usage
- "Find all open P1 issues assigned to me"
- "Move issue ENG-123 to In Progress"
- "Add a comment to ENG-456"
2. Optionally, supporting files — configs, helper scripts, or anything the SKILL.md references.
To publish:
clawhub skill publish ./my-jira-skill/
That's it. ClawHub assigns a version, indexes the skill for vector search, and makes it publicly discoverable. If you later want to iterate:
# Bump version and republish
clawhub skill publish ./my-jira-skill/ --version 1.1.0
A few things to be aware of when writing a SKILL.md:
- Be explicit about which tools must be enabled — users shouldn't have to guess
- Declare environment variables and external binaries in the frontmatter
- Include example prompts; they dramatically lower the barrier for new users
- Avoid overlap with other popular skills (e.g., don't define a "screenshot" tool if Agent Browser is commonly used — conflicts confuse the agent)
The /tasks Panel: A Quick Bonus
One feature worth knowing before you start running complex skills: the /tasks in-chat panel, introduced in the 2026.4.1 release. Type /tasks in any OpenClaw chat and you get a unified view of all background tasks — sub-agents, cron jobs, long-running execs — with live status. If you're running a skill that kicks off a multi-step background workflow, this is the quickest way to see what's actually happening without digging through logs.
Wrapping Up
The skill system is what takes OpenClaw from "interesting demo" to "thing I actually use daily." The install path is deliberately simple — one command, one restart — because the real work is in choosing the right skills for your workflows and understanding the Tool vs. Skill boundary that governs what your agent can actually do.
Start with a small, focused stack. Read the SKILL.md before you install. Keep things updated. And when you've built something that solves a problem worth sharing, publish it. ClawHub is only as good as what the community puts into it.
Have questions or found a skill worth highlighting? Drop a comment below.
ClawCon Michigan
I wasn't able to attend ClawCon Michigan in person, but followed the livestreams and community recaps closely — the sessions on agent security and the skill ecosystem were particularly worth watching if you can find recordings.
Top comments (0)