Top 7 Skills to Install Right After Setting Up OpenClaw

The first time you open the ClawHub marketplace after installing OpenClaw, you’re hit with hundreds of skills. It’s hard to know where to start. Sorting by rating just gives you a wall of 3.9-star entries, and that tells you nothing about whether any of them are actually useful.

I went through more than 30 of them firsthand — read the code, ran them, and deleted some the same minute I installed them. This post is the result of that process. Everything here is based on real usage, not star ratings.

The 7 Essential Skills

1. agent-browser (Rating: 3.86) — The Foundation for Web Automation

This is the first thing you should install. It’s a Playwright-based web automation skill, but it works differently from a standard Playwright script. Instead of dumping the entire DOM to the LLM, it uses a snapshot+refs approach to extract only the actionable elements. The official claim is a 93% reduction in token usage compared to conventional methods — and honestly, you feel it in practice.

It handles pretty much everything you’d want a browser to do: automating blog platforms, scraping pages that require login, filling out forms repeatedly. OpenClaw has a built-in browser tool, but agent-browser replaces it in a much more efficient way.

clawhub install agent-browser

After installing, you can open a browser session with a single line: agent-browser open https://example.com. The SKILL.md file has solid documentation — worth reading through.

2. qmd (Rating: 3.81) — Local File Search CLI Powered by BM25

This one’s for when you need to search through local files quickly. It’s built on the BM25 algorithm, which makes it strong for full-text search, and it runs without any external dependencies like Ollama.

My most common use case is checking for duplicate blog posts. I index my posts directory, and before writing anything new, I can verify in under a second whether I’ve already covered a topic. It indexes anything text-based — markdown files, code, documentation, you name it.

clawhub install qmd
# Or install directly:
npm install -g https://github.com/tobi/qmd

It also supports MCP mode, so you can integrate it directly as an OpenClaw tool. It’s registered on ClawHub, so clawhub install works — but direct npm installation is sometimes faster.

3. deep-research-pro (Rating: 3.56) — Multi-Source Deep Research, No API Key Required

You give it a topic, it breaks it down into sub-questions, searches each one via DuckDuckGo, and produces a report with cited sources. No OpenAI API key needed. That’s the key selling point.

I’ve used it for tracking the latest AI trends, researching real-world implementations of specific technologies, and basic competitive analysis. It’s not perfect — since it depends on DuckDuckGo, it can’t pull from academic papers or anything behind a paywall. But if you need a free research pipeline, there’s nothing else like it.

clawhub install deep-research-pro

The 3.56 rating probably reflects a mismatch in expectations. Of course it doesn’t match a paid GPT-4-powered research tool. Judged as a free option, it does the job well enough.

4. openclaw-backup (Rating: 3.62) — Run This Immediately After Setup

This backs up your entire ~/.openclaw directory — config, credentials, workspace, cron settings, everything. If you ever need to reinstall OpenClaw or move to a new machine, this backup is the difference between a five-minute restore and starting from scratch.

One command backs everything up, and example cron automation is included. Running it once right after setup is good enough insurance on its own.

clawhub install openclaw-backup

Strongly recommend installing and running this immediately after setup. The version of you that accidentally wipes your config later will thank you.

5. youtube-watcher (Rating: 3.65) — Free YouTube Transcript Extraction

A wrapper around yt-dlp. If a YouTube video has captions, this pulls the full transcript for free. I use it when I don’t have 40 minutes to watch an AI news channel — just extract the transcript and summarize the key points.

yt-dlp needs to be installed separately. If it’s not already there, run pip install yt-dlp first.

clawhub install youtube-watcher

Auto-generated captions are extractable too, but the quality is inconsistent. Transcripts uploaded directly by the channel come out much cleaner. Works better on English-language videos than Korean ones.

6. naver-news (Rating: 1.07 — but the code is actually solid) — Naver News API Integration

A 1.07 rating makes you want to scroll right past this one. But when I looked at the code, it was cleaner than I expected. My guess is the low rating has nothing to do with the skill itself — it’s probably because getting a Naver API key is annoying. You have to go to their Developer Center, create an app, and grab a Client ID and Secret. That friction alone seems to explain the rating.

The actual functionality is clear: give it a keyword and a date filter, and it pulls Naver News search results. 25,000 calls per day free. If you need to monitor Korean news, this is what you use.

clawhub install naver-news

Once you have the API key, it works immediately. This skill taught me the most important lesson from this whole exercise: always read the code before trusting the rating.

7. summarize (Rating: 3.95 — #1 on ClawHub) — Multi-Source Summarization

Summarizes content from URLs, PDFs, YouTube videos, and more. The highest-rated skill on ClawHub, and the wide applicability shows why.

One thing to check before installing: depending on your OpenClaw version, this might already be bundled. Installing a duplicate can cause conflicts, so verify first.

clawhub install summarize

I use it most often for long web articles. Drop in a URL and it pulls out the key points. Especially efficient when you’re dealing with a lot of English-language material.

Conditional Recommendations: If You Have API Keys or Accounts

The seven above work out of the box, but these two need some prep work first.

x-twitter: Worth installing if you have a Twitter Bearer Token. The twclaw CLI handles reading and posting on Twitter. Useful for monitoring AI trends. Getting a Bearer Token is more involved than it used to be, but if you already have one, this is worth trying.

mem0: The ClawHub version hardcodes the OpenAI API and uses an in-memory vector store, meaning everything disappears when the process exits. If persistent long-term memory is the goal, you’re better off setting up the Mem0 OSS server directly. Also worth noting: OpenClaw’s native memory feature already integrates with Mem0, so in many cases there’s no reason to install this at all.

Patterns to Avoid

After going through 30+ skills, a few red flags stand out. Treat these as warning signs.

High rating doesn’t mean there’s actual code: Some popular skills — 3.7 stars and up — contain nothing more than text documents. We’re talking about things that literally say “to do this task, use the web_search tool.” automation-workflows (3.70) and seo-competitor-analysis (3.64) were the worst examples I hit. Opened them up and found no executable scripts whatsoever. Always look at the code before you install.

Skills that wrap paid SaaS products: I installed naver-blog-writer and it immediately tried to connect to a remote control plane on local port 19090. The actual features were paywalled. google-analytics (3.57) routed through a separate paid service called MATON_API_KEY. Most skill descriptions don’t mention any of this upfront — you only find out after installing. Checking the code first is the only reliable way to catch it.

VirusTotal flags need context: A VT flag doesn’t automatically mean a skill is malicious. The youtube-transcript skill got flagged because it included WireGuard-related code — not actually malicious. Before you nuke something based on a flag, read the actual code and make a judgment call.

For a deeper look at security in the ClawHub ecosystem, I wrote a separate post: The Security Crisis in OpenClaw’s Skill Ecosystem.

Wrap-Up

One-line summary: install agent-browser + qmd + openclaw-backup the moment you finish setup. Pick from the other four based on what you actually need.

There are definitely undiscovered gems still sitting on ClawHub. I probably scrolled past skills with low ratings that turn out to be like naver-news — solid code behind a bad reputation. Reading the code directly ended up being the only metric that actually held up.