Skip to main content

OpenClaw Skills — How to Give Your AI Agent Superpowers

Here's something that blew my mind.

I told my OpenClaw agent: "Create a new skill called github-trending that shows me today's top GitHub projects."

Two minutes later, it had:

  • Written the skill code
  • Saved it to my workspace
  • Committed it to git
  • Run it immediately
  • Delivered a beautifully formatted report of today's trending repositories

The AI didn't just use a tool. It built a tool, installed it, and used it — all by itself.

This is what OpenClaw skills are about. And this guide will show you how to set them up, fix the inevitable dependency headaches, and unlock capabilities that will change how you think about AI assistants.


What Are Skills, and Why Should You Care?

Think of OpenClaw without skills as a brilliant mind trapped in an empty room. It can think, but it can't do anything.

Skills are what give it hands.

With the right skills installed, your agent can:

  • Read and analyze PDFs
  • Manage your GitHub repos — create issues, review PRs, check CI status
  • Process images and videos
  • Search the web and summarize findings
  • Access your Feishu/Lark docs, wikis, and drives
  • Monitor system health
  • And here's the wild part — create entirely new skills for itself

The Skills That Come Built-In

When you first set up OpenClaw, some capabilities are already there — no extra configuration needed:

CapabilityWhat It Does
Web browsingOpens pages, reads content, navigates sites
File systemReads, writes, and organizes files on your machine
Shell executionRuns terminal commands and scripts
MessagingSends messages, sets reminders, manages scheduled tasks
MemoryRemembers past conversations and context
Sub-agentsSpawns specialized agents for complex multi-step tasks

These aren't "skills" in the technical sense — they're core tools baked into OpenClaw. You don't need to install anything for these to work.


Choosing Your First Skills

During the onboard wizard (openclaw onboard), you'll see a screen listing 20+ optional skills. Here's the full list:

🔐 1password       📰 blogwatcher      🫐 blucli
📸 camsnap 🧩 clawhub 🎛️ eightctl
🧲 gifgrep 🐙 github 🎮 gog
📍 goplaces 📧 himalaya 📦 mcporter
🍌 nano-banana-pro 📄 nano-pdf 💎 obsidian
🎙️ openai-whisper 💡 openhue 🧿 oracle
🛵 ordercli 🗣️ sag 🌊 songsee
🔊 sonoscli 🧾 summarize 🎞️ video-frames
📱 wacli 𝕏 xurl

Don't install everything. Most of them require system dependencies (Homebrew packages, Python tools) and installing all of them will take forever and likely fail.

After trial and error, here's what I'd recommend for most users:

SkillWhat It DoesWhy You Need It
🐙 githubFull GitHub integration — issues, PRs, codeEssential if you write code
📄 nano-pdfRead and analyze PDF filesExtremely useful for document-heavy work
🧾 summarizeSummarize long texts and documentsPairs perfectly with nano-pdf
🧩 clawhubOpenClaw's official plugin hubMakes installing future skills easier
🍌 nano-banana-proImage generation and analysisGives your agent "eyes"

Skills You Can Skip (For Now)

SkillWhy Skip
💡 openhue / 🔊 sonoscliSmart home control — unless you want your agent turning lights on and off
🔐 1passwordOnly if you use this specific password manager
📍 goplacesLocation services — low priority for most setups
🎮 gogGaming platform integration — niche use case

How to Select Skills in the Wizard

This trips up a lot of people:

  • Arrow keys (↑/↓) to move between options
  • Space bar to select/deselect (the checkbox changes from ◻ to ◼)
  • Enter to confirm your selections

⚠️ Warning: Don't press Enter without selecting anything — the wizard interprets that as "Skip all" and moves on.


The Dependency Problem (And How to Fix It)

Here's the part nobody warns you about.

After selecting your skills and pressing Enter, you'll likely see a cascade of red error messages:

Install failed: github — brew not installed
Install failed: nano-pdf — uv not installed
Install failed: nano-banana-pro — brew not installed

Don't panic. This is completely normal. Most skills depend on system-level tools that aren't installed yet.

What Your Skills Actually Need

Behind the scenes, skills depend on a chain of system tools:

Skills (github, nano-pdf, summarize...)
↓ depends on
System tools (gh, poppler, ffmpeg, imagemagick...)
↓ installed via
Package managers (Homebrew, uv)
↓ which need
Your operating system (WSL/Linux/macOS)

If any link in this chain is missing, the skill fails to install.

Step 1: Install Homebrew

This is the foundation. Most skills need it.

# Standard installation
/bin/bash -c "$(curl -fsSL https://raw.githubusercontent.com/Homebrew/install/HEAD/install.sh)"

# If you're in China and the above is slow/fails:
/bin/bash -c "$(curl -fsSL https://gitee.com/cunkai/HomebrewCN/raw/master/Homebrew.sh)"

Critical: After installation, run the "Next steps" commands that appear on screen:

(echo; echo 'eval "$(/home/linuxbrew/.linuxbrew/bin/brew shellenv)"') >> ~/.bashrc
eval "$(/home/linuxbrew/.linuxbrew/bin/brew shellenv)"

Verify: brew --version should show a version number.

Step 2: Install uv (Python Package Manager)

Many skills use Python internally. uv is the modern, blazing-fast Python environment manager:

curl -LsSf https://astral.sh/uv/install.sh | sh
source ~/.bashrc

Step 3: Install Core System Packages

One command to install everything most skills need:

brew install gh uv poppler ffmpeg imagemagick coreutils

What each package unlocks:

  • gh → GitHub skill (issues, PRs, code operations)
  • poppler → nano-pdf skill (PDF reading and analysis)
  • ffmpeg → video-frames skill (video processing)
  • imagemagick → Image manipulation capabilities
  • uv → All Python-based skills (summarize, etc.)

Step 4: Let OpenClaw Fix the Rest

openclaw doctor --fix

This command scans your system, finds what's missing, and automatically installs what it can. Run it after installing the packages above.

Tracking Your Progress

Here's what my skill journey looked like:

StageEligible SkillsWhat Changed
First install (no Homebrew)10 / 55Only built-in tools worked
After Homebrew + basic packages14 / 55GitHub, PDF, summarize came online
After full package install + doctor fix16 / 55Additional tools activated

The remaining 39 "missing" skills need more specialized dependencies. You don't need all 55 — 16 eligible skills is already plenty for daily use.


The Wow Moment: Your Agent Can Build Its Own Skills

This is where it gets wild.

OpenClaw comes with a meta-skill called skill-creator — a skill whose entire purpose is to create other skills.

I tested this by saying:

"Create a new skill called github-trending that searches today's top GitHub Trending projects and formats them with Star count, language, description, and why they're worth watching."

What happened next:

  1. The agent analyzed what I wanted
  2. Wrote a complete SKILL.md file with the skill definition
  3. Saved it to skills/github-trending/
  4. Committed it to git (commit: 5dc1368)
  5. Immediately ran the new skill
  6. Delivered this result:

Today's GitHub Trending (Top 6 visible)

  1. obra/superpowers — ⭐ 90,831 — Shell An agentic skills framework & software development methodology that works. → Agent/skill framework, directly relevant to the AI agent ecosystem.

  2. codecrafters-io/build-your-own-x — ⭐ 479,073 — Markdown Master programming by recreating your favorite technologies from scratch. → Massive developer learning resource, consistently trending.

  3. langchain-ai/deepagents — ⭐ 13,613 — Python Agent harness with planning, filesystem, and subagent support. → Multi-agent orchestration becoming standard infrastructure.

(and 3 more...)

From "I want a tool that doesn't exist" to "here are today's results" — in under 2 minutes.

The agent didn't just use a pre-built plugin. It invented a new capability for itself on the spot. This is the fundamental difference between a chatbot and an AI agent.


Real-World Skill Demo: GitHub Inbox Triage

Here's another example that shows what configured skills can actually do in practice.

I asked my agent:

"Search today's OpenClaw GitHub activity — commits, issues, discussions — and triage them into: needs attention, good to know, and just FYI."

In about 30 seconds, it delivered a structured analysis of 20+ items:

Needs Attention (4 critical issues):

  • Gateway restart fails to recover from stale lock files
  • Local mode exec completely broken on macOS companion
  • Telegram duplicate session race condition
  • Terminal left in broken state after exit

Good to Know (8 items):

  • Multiple test infrastructure improvements (Feishu, Google OAuth)
  • SSH sandbox and macOS service fixes
  • Feature request: Feishu cards and streaming output

Just FYI (6 items):

  • New community plugins (social publishing, LanceDB memory)
  • Edge case bugs (sticker cache, markedjs freeze)

For each item, it explained why it categorized it that way. Then it offered to further prioritize into "verify now / track this week / just watch."

A human doing this manually would need 30-60 minutes of GitHub browsing. The agent did it in 30 seconds with better structure than most people would produce.


Let Your Agent Install Skills for You

Here's a shortcut most tutorials don't mention.

Instead of manually fighting with dependencies and configuration, you can simply ask your agent:

"Check what skills you have, what's missing, and install the three you think would be most useful for me."

When I did this, my agent:

  1. Audited its own skill inventory
  2. Identified gaps in its capabilities
  3. Recommended three new workflow skills based on my usage patterns:
    • calendar-planner — daily/weekly planning, focus blocks, meeting prep
    • inbox-triage — message classification, blocker identification
    • cross-team-comm — alignment messages, risk syncs, status updates
  4. Created all three skills with complete SKILL.md definitions
  5. Added reference templates for each (planning templates, triage boards, communication patterns)
  6. Committed everything to git (two commits: skill creation + template enhancement)

The agent essentially said: "Based on how you use me, here's what I'm missing and here's how I'm fixing it."


Feishu/Lark Skills: A Special Category

If you use Feishu (Lark), OpenClaw has deep integration through dedicated skills:

SkillWhat It Does
feishu-docRead and write Feishu documents
feishu-driveManage cloud files and folders
feishu-permHandle permissions and sharing
feishu-wikiNavigate and update knowledge bases
feishu-chatSend group messages, @mentions, formatted cards

These are configured automatically during the Feishu channel setup in the onboard wizard. No additional installation needed.


Troubleshooting Skills

"Missing requirements: 43"

This is normal after a fresh install. Follow the dependency installation steps above, then run openclaw doctor --fix.

Skills installed but not working

Check if the underlying tool is actually accessible:

# For GitHub skill
gh --version

# For PDF skill
pdftotext -v

# For Python-based skills
uv --version

If any of these fail, reinstall via brew install <package>.

"openclaw doctor" shows Eligible count not increasing

Make sure your shell has loaded the latest paths:

source ~/.bashrc
openclaw doctor --fix

Want to see exactly what's missing?

openclaw doctor

Look at the "Skills status" section. It shows Eligible (working), Missing requirements (need dependencies), and Blocked by allowlist (disabled by policy).


The Big Picture: Skills Are Just the Beginning

After configuring skills, your OpenClaw agent transforms from a chatbot into something much more capable:

Without skills: Can chat, answer questions, and think deeply — but can't do anything on your behalf.

With basic skills: Can read your PDFs, manage your GitHub repos, process images, and summarize documents.

With workflow skills: Can plan your day, triage your inbox, draft cross-team communications, and manage your schedule.

With skill-creator: Can identify its own capability gaps and build new tools to fill them — without you writing a single line of code.

The progression feels less like "configuring software" and more like "training a new team member who gets better every day."


What's Next?


Based on real configuration experience with OpenClaw 2026.3.8 on Windows 11 WSL2. Skill status progression (10 → 14 → 16) and all demos (inbox-triage, github-trending, self-installed workflow skills) are from actual usage.