OpenClaw is one of the most popular open-source AI agents in 2026, with over 247,000 GitHub stars. One of its strengths is the skills system — called AgentSkills — which uses the same SKILL.md format as Claude Code, Codex CLI, and other AI coding agents.
This means skills built for Claude Code work on OpenClaw without modification, and vice versa. Here's how to use them.
Quick Answer: OpenClaw stores skills in
~/.openclaw/skills/, with each skill residing in its own folder and containing aSKILL.mdfile. You can install skills by unzipping archives or cloning Git repositories into this directory, or by symlinking/copying existing Claude Code skills.
Where OpenClaw stores skills
OpenClaw looks for skills in ~/.openclaw/skills/. Each skill gets its own folder containing a SKILL.md file:
~/.openclaw/skills/
├── code-reviewer/
│ └── SKILL.md
├── git-commit-writer/
│ └── SKILL.md
└── env-doctor/
├── SKILL.md
└── scripts/
└── check-env.sh
OpenClaw also ships with over 100 built-in AgentSkills that cover common tasks like file management, shell commands, web browsing, and API interactions.
Installing skills for OpenClaw
The process is the same as Claude Code — just use the OpenClaw skills directory:
# From a zip file (e.g., downloaded from Agensi)
unzip skill-name.zip -d ~/.openclaw/skills/
From GitHub
git clone https://github.com/author/skill-repo.git cp -r skill-repo/skill-name ~/.openclaw/skills/skill-name ```
OpenClaw discovers new skills on startup. You can also reload skills through your messaging interface without restarting.
Using Claude Code skills on OpenClaw
If you already have skills installed for Claude Code at ~/.claude/skills/, you can use them on OpenClaw by copying or symlinking:
# Copy a specific skill
cp -r ~/.claude/skills/code-reviewer ~/.openclaw/skills/code-reviewer
Or symlink your entire Claude skills directory
ln -s ~/.claude/skills/* ~/.openclaw/skills/ ```
Most skills work across both agents without any changes. The SKILL.md format is identical.
Skills that might need adjustment
While the format is the same, some behaviors differ between agents:
Skills that use Claude Code's context: fork setting to run as subagents may behave differently on OpenClaw, since OpenClaw handles concurrent tasks through its own gateway system.
Skills that reference Claude-specific slash commands (like /simplify or /batch) won't work on OpenClaw since those are built into Claude Code.
Skills that stick to standard markdown instructions, shell commands, and file operations work identically on both platforms.
Creating skills for OpenClaw
Creating a custom skill is the same process:
mkdir -p ~/.openclaw/skills/my-skill
Create a SKILL.md file with frontmatter and instructions. The frontmatter format is identical to Claude Code:
---
name: my-skill
description: Description of what this skill does and when to use it.
---
My Skill
Instructions here. ```
Security considerations
OpenClaw's open-source nature means anyone can publish skills, and the built-in skill repository has had security incidents. In February 2026, researchers found hundreds of malicious skills that performed data exfiltration and prompt injection.
Before installing any skill from untrusted sources:
Read the SKILL.md file completely. Check any scripts in the scripts/ directory. Be wary of skills that make network requests or access sensitive files.
Marketplaces like Agensi run automated security scans on every submission, which adds a layer of protection. If you find a skill on GitHub, consider checking if it's also listed on a marketplace that does security reviews.
Finding skills for OpenClaw
Since OpenClaw uses the SKILL.md standard, any skill built for Claude Code works on OpenClaw. This means the entire ecosystem is available to you:
Browse Agensi for curated, security-scanned skills across code review, git automation, documentation, DevOps, testing, and more.
Search GitHub for filename:SKILL.md to find community skills.
Check OpenClaw's built-in AgentSkills for common automation tasks.
Browse SKILL.md skills for OpenClaw, Claude Code, and 20+ agents on Agensi.