I’ve installed 47 OpenClaw skills over the past eight months. I actively use 12 of them. The other 35 range from “occasionally useful” to “what was I thinking?” Here’s what’s actually worth installing — and what sounds cool but wastes your time.
The 12 Worth Installing
1. GitHub integration. This is the first skill I install on every OpenClaw setup. It handles issues, PRs, CI status, and code review from within your AI workflow. When a team member mentions a bug in Slack, I can say “create an issue for that” and the skill handles it — filed, labeled, and assigned. Daily use.
2. Web search. Gives your agent access to web search. Essential for any task where the AI needs current information. Without it, your agent is limited to its training data. Install this immediately.
3. Summarization. Summarizes URLs, documents, videos, and podcasts. I use this 3-4 times per day. Paste a YouTube link, get a summary. Paste an article, get the key points. It’s the skill that non-technical people are most impressed by.
4. Cron/scheduling. Already covered this in my cron jobs post, but the scheduling skill is essential for any automation that needs to run on a timer. Morning briefings, weekly reports, daily cleanups — all depend on this.
5. Coding agent. Delegates complex coding tasks to a dedicated coding agent that can read files, write code, and run tests. This is for when you need more than a code snippet — you need a multi-file implementation or a complex refactor.
6. Browser automation. Controls a web browser for tasks that require visual interaction. Filling forms, scraping data from sites without APIs, testing web applications. I use it 2-3 times per week.
7. Database querying. Lets the agent query your databases with natural language. “How many users signed up last month?” without writing SQL. The skill translates your question into a query, runs it, and returns the results.
8. File system. Read and write files on your local or remote file system. Fundamental for any workflow that involves creating, editing, or organizing documents.
9. Discord/Slack integration. Connects your agent to team messaging. Send messages, read channels, respond to mentions. This is how most people interact with their OpenClaw agent daily.
10. Weather. Sounds trivial, but having weather in your morning briefing is surprisingly useful. “It’s going to rain this afternoon — bring an umbrella” is the kind of practical value that makes the whole system feel worth it.
11. Health check. Monitors the health of your servers and services. Pings endpoints, checks status codes, alerts when something’s down. Simple but essential for anyone running infrastructure.
12. PDF tools. Read, summarize, and extract data from PDFs. Business is still drowning in PDFs, and being able to say “extract the payment terms from this contract” saves genuine time.
The 5 to Skip
1. Auto-posting to social media. Sounds great in theory — AI generates posts and publishes them automatically. In practice, the generated content is generic, the timing is wrong, and the brand voice is off. Social media needs a human touch. Use AI to draft posts, but publish them yourself.
2. Email auto-responder. An AI that automatically responds to incoming emails without human review is a lawsuit waiting to happen. I installed this, turned it on for a day, and it sent a “sorry for your loss” style response to a client who mentioned a “deadline death.” Never again.
3. Voice transcription (local). The local transcription models are slow and inaccurate compared to cloud services. If you need transcription, use a cloud API. The local skill exists but isn’t worth the compute resources.
4. Crypto price tracker. Unless you’re actively trading, knowing that Bitcoin went up 2% doesn’t improve your life. It just adds noise to your morning briefing. I installed it, watched prices for a week, realized I was making no decisions based on the data, and removed it.
5. Joke/entertainment skills. “Tell me a joke” and “play a game” skills are fun for about five minutes. Then they sit unused, consuming space in your skill list and occasionally polluting agent responses when the AI decides to be “fun” at inappropriate moments.
How to Evaluate a New Skill
Before installing anything, ask three questions:
Will I use this more than once a week? If not, it’s probably not worth the setup. Running a command manually once a week is faster than configuring, testing, and maintaining an automated skill.
Does this replace something I currently do manually? The best skills automate existing workflows. The worst skills create new workflows that you didn’t need. If you weren’t doing the task before, you probably don’t need to automate it now.
What happens when it breaks? Every skill is a potential failure point. A broken GitHub skill means missed issues. A broken email auto-responder means embarrassing messages to clients. Evaluate the blast radius before installing.
The Meta-Lesson
Fewer, well-configured skills beat many poorly-configured ones. Every time. My OpenClaw instance with 12 skills is more useful than the version with 47 skills, because every installed skill is one I actually use, trust, and maintain.
Start with the top 5 on my list. Use them for a month. Then add more based on what you actually need, not what sounds cool in a README file.
🕒 Last updated: · Originally published: December 8, 2025