I Gave My AI Agent 60+ Superpowers — Here Are the Ones I Actually Use Daily

A few weeks ago I wrote about self-hosting OpenClaw as an AI agent on WhatsApp, Telegram, and Discord. The response was great — but the part that got the most questions was Skills.

So let me go deeper. Skills are what turn a basic chatbot into something genuinely useful. OpenClaw has 60+ of them, and I've tested most. Here are the ones that actually stuck in my daily workflow.

What Are Skills, Exactly?

Think of skills as plugins for your AI agent. Each one gives OpenClaw access to a specific tool or API — GitHub, Docker, your smart home, web search, and more.

The key difference from just asking ChatGPT to "help me with Docker": skills have real tool access. They don't just generate text about Docker — they actually run docker ps, pull images, and manage containers through your messaging app.

My Daily Driver Skills

1. GitHub CLI — The One I Can't Live Without

This replaced half my browser tabs. From Telegram, I can:

• List open PRs on any repo
• Review diffs and leave comments
• Create issues with labels
• Merge PRs after review

Me: "Show me open PRs on our main repo"
Agent: "Found 3 open PRs:
  #241 - Fix auth timeout (by @alex, 2 approvals)
  #239 - Add rate limiting (by @maria, needs review)
  #237 - Update deps (by @bot, CI passing)"

Me: "Review #239"
Agent: [provides diff summary and key changes]

The speed difference is real. Instead of: notification → open browser → navigate to PR → read → comment... it's just one message.

2. Tavily Web Search — Real-Time Answers

Unlike the AI's built-in knowledge (which has a cutoff date), Tavily gives your agent live web search:

• "What's the latest Node.js LTS version?"
• "Any outages on AWS us-east-1 right now?"
• "What did the React team announce this week?"

It searches, summarizes, and cites sources. I use this dozens of times a day.

3. Docker Essentials — DevOps From Your Phone

Managing containers from a messaging app sounds gimmicky until you're away from your desk and something goes down:

• "List running containers" → instant overview
• "Restart the postgres container" → done in seconds
• "Show logs for the api container, last 50 lines" → quick debugging

This has a Caution badge for good reason — it can stop/remove containers. But with allowed_users configured, only you can trigger it.

4. Linear — Tickets Without Context Switching

I used to keep Linear open in a tab all day. Now:

"Create a ticket: API returns 500 on /users endpoint when pagination exceeds 100. Priority high, assign to me."

Done. No tab switching, no clicking through dropdowns. My team thinks I've become incredibly responsive on tickets — I just reply from Telegram while making coffee.

5. Home Assistant — The Non-Dev Favorite

This one's for after work hours:

"Set office lights to warm white, 40%"
"What's the temperature in the living room?"
"Turn off all lights except bedroom"

My partner now uses the Telegram bot too. It's more natural than any smart home app.

The Underrated Ones

A few skills that don't get enough attention:

Remind Me — Dead simple but surprisingly useful. "Remind me to check the deployment in 2 hours" and it actually does. Through your messaging app, not some notification you'll ignore.

Cost Report — Shows your AI token usage and costs. I check this weekly. Keeps spending transparent — usually around $6-8/month for me.

Config Guardian — Validates your OpenClaw configuration and warns about potential issues. Saved me from a misconfiguration that would have exposed my instance.

Exa Web Search — Like Tavily but with a free tier. Good if you're just getting started and want to test web search without committing to a paid API key.

Safety Badges Matter

Every skill has a badge. Pay attention to them:

Badge	Meaning	Example
Recommended	Well-tested, low risk	GitHub CLI, Web Search
Caution	Can modify/delete things	Docker, Auto PR Merger
Experimental	Newer, community testing	Voice Reply, Proactive Agent

My rule: start with Recommended skills only. Add Caution skills once you're comfortable with how OpenClaw handles permissions. Avoid Experimental in production.

Installing Skills

It's straightforward:

# Browse available skills
openclaw skills list

# Install one
openclaw skills add github-cli

# Install multiple
openclaw skills add tavily-search docker-essentials linear

Each skill has its own setup requirements — usually just an API key. The CLI walks you through it.

The Skill I'm Most Excited About

Self-Improving Agent — This one lets OpenClaw learn from your corrections. If you tell it "no, when I say 'deploy' I mean push to staging, not production", it remembers. Over time, it gets noticeably better at understanding your intent.

It's still marked Recommended (not Experimental), which tells me the OpenClaw team has put real work into making it reliable.

Quick Setup if You're New

If you haven't set up OpenClaw yet:

npm install -g openclaw@latest
openclaw onboard

Then add skills:

openclaw skills add github-cli tavily-search

That's it. You'll have an AI agent with real GitHub and web search capabilities in your messaging app within 10 minutes.

Full Skills Directory

I've been using open-claw.me/skills as my reference — it has all 60+ skills organized by category (DevOps, Security, Productivity, AI & Agent, etc.) with setup guides and safety badges. Useful for browsing what's available before installing.