OpenClaw Skills Guide

OpenClaw ships with a powerful core - web search, file management, shell execution, and message handling. But the real flexibility comes from skills: modular npm packages that plug additional capabilities directly into your assistant.

Each skill exposes new tools to the underlying AI model. Install the Playwright skill and your assistant gains a full controllable web browser. Install the weather skill and it can fetch real-time forecasts for any city. Install the tmux skill and it can attach to running terminal sessions and send keystrokes.

Skills work by registering tool definitions that the AI model can call. When you ask your assistant to 'take a screenshot of this website,' it invokes the browser tool from the Playwright skill, which spins up a headless Chromium instance, navigates to the URL, captures the screenshot, and returns it to you - all automatically.

Installation is straightforward: run `clawhub install <skill-name>` or `npm install @openclaw/<skill-name>` in your OpenClaw directory. The assistant picks up new skills on the next session. No restart required in most cases, no config files to edit, no environment variables to set beyond what the skill's README specifies.

The OpenClaw skills ecosystem has grown quickly. Here are the skills that most power users install first.

Playwright is the most-requested skill. It gives your assistant a real Chromium browser it can navigate, click, fill forms, take screenshots, and extract content from. This is how you automate logins, scrape JavaScript-heavy sites, test web apps, or just have your assistant browse for you hands-free.

The weather skill connects to open weather APIs and gives your assistant current conditions plus multi-day forecasts for any location. It requires no API key and works globally. Useful for heartbeat automations like 'remind me if rain is forecast tomorrow morning.'

Video frames uses ffmpeg to extract still frames or short clips from video files. If you have video content on your server, your assistant can analyze specific moments, generate thumbnails, or extract clips based on timestamps - all via natural language.

The tmux skill lets your assistant attach to running terminal sessions. This is powerful for server management: your assistant can check what's running, tail logs, restart services, or run commands in existing shell sessions without spawning new ones.

Healthcheck is the system security skill. It audits your server configuration, checks for exposed ports, reviews SSH settings, scans for outdated packages with known vulnerabilities, and reports risk levels with remediation steps.

The clawhub skill itself is meta - it lets your assistant search and install other skills on your behalf. Ask 'find me a skill for monitoring disk usage' and your assistant will search clawhub, review the results, and install the best match.

Understanding skill structure helps you evaluate new skills and build confidence that they're safe to install.

Every skill is a standard npm package with a `SKILL.md` file at its root. This markdown file is critical - it's the instruction manual your AI reads to understand how to use the skill's tools. Well-written SKILL.md files include example prompts, parameter descriptions, and usage patterns. When OpenClaw loads a skill, it reads SKILL.md and incorporates those instructions into the assistant's context.

The actual tool implementations live in JavaScript or TypeScript source files. Tools are registered functions with typed parameters and return values. When the AI model decides to call a tool, OpenClaw validates the parameters, executes the function, and returns the result back to the model for interpretation.

Skills can declare dependencies on system tools. The Playwright skill lists Chromium as a requirement. The video-frames skill lists ffmpeg. OpenClaw's skill loader checks for these dependencies and will warn you if something is missing. On a barebones VPS, you might need to install these system packages manually - which is one reason people choose managed hosting.

Skill isolation is worth understanding. Skills run in the same Node.js process as OpenClaw, with the same filesystem access and network access. There is no sandbox. This is why reviewing skill code matters, especially for community-built skills. Stick to the official skill packages or skills with readable, well-reviewed source code.

Most skill installation failures come from missing system dependencies, not the skills themselves. Playwright needs Chromium plus a dozen shared libraries. Video-frames needs ffmpeg compiled with the right codecs. Healthcheck needs various system utilities. On a fresh VPS, setting these up takes time and expertise.

PlugAndClaw provisions each server with the full dependency stack pre-installed. Chromium, ffmpeg, system utilities, Node.js with npm - everything is ready from day one. When you install a skill, it just works. No 'missing shared library' errors, no codec compatibility issues, no permission problems.

Your PlugAndClaw server is also a dedicated VPS, not a shared container. Playwright can spin up full browser instances without hitting memory limits. Video processing doesn't get throttled. System-level skills like healthcheck can access real system information, not a virtualized view. The 2 vCPU and 4GB RAM allocation means skills that need compute actually have it.

Skill data persists, too. If a skill writes cache files or configuration, those files survive between sessions because they live on your persistent SSD. On serverless or shared setups, this data evaporates between runs.

For businesses running multiple skills simultaneously - browser automation running alongside email monitoring alongside competitor price tracking - the dedicated resource allocation means no contention. Each task gets what it needs without fighting other processes for CPU time.

Once your PlugAndClaw server is running, skill management is handled through your assistant via Telegram. Just ask: 'Install the Playwright skill from clawhub' and your assistant handles the npm install, reads the SKILL.md, and confirms when it's ready.

You can also SSH into your server directly to manage skills via CLI. Your server details are provided in your PlugAndClaw dashboard. From the terminal, `clawhub list` shows installed skills with version numbers, `clawhub search <query>` finds skills in the registry, and `clawhub update` upgrades all skills to latest versions.

Skill configuration, when needed, happens through environment variables or config files in your OpenClaw directory. Your assistant can guide you through this: 'How do I configure the GitHub skill with my token?' will walk you through the exact steps for your specific skill.

Removing skills is equally simple: `clawhub uninstall <skill-name>` removes the package and clears it from the assistant's available tools. Your assistant won't attempt to use tools it doesn't have, so removal is clean and immediate.

PlugAndClaw's $39.50/month plan includes $20 in AI credits, which means experimenting with skills - asking your assistant to try browser automation, test the weather skill, run a healthcheck - doesn't require separate API billing. The credits are included, the skills are pre-supported, and the server is already running.