OpenClaw skills. Where they live and how to install them.

What skills are in OpenClaw

OpenClaw is an open-source AI assistant that runs on your own computer and talks to you through the chat apps you already use. Its documentation describes skills as markdown instruction files that teach the agent how and when to use tools. Each skill is a directory with a SKILL.md file: YAML frontmatter followed by a markdown body. OpenClaw follows the Agent Skills specification, so a skill written for another agent usually works here without changes.

OpenClaw loads its bundled skills plus any local ones, then filters them at load time based on environment, config and which binaries are installed. Required frontmatter is name and description. Optional keys include user-invocable, disable-model-invocation and command-dispatch, and a metadata.openclaw block can declare requirements such as binaries, environment variables, config paths or operating systems. A skill whose requirements are not met stays out of the agent's list.

Where OpenClaw looks for skills

When the same skill name appears in more than one place, the highest source in this table wins.

A skill in <workspace>/skills is visible only to that agent, while one in ~/.openclaw/skills is shared by every agent using that state directory. OpenClaw finds a SKILL.md anywhere under a skills root, up to six levels deep, so you can group skills in subfolders. The skill's name and slash command come from the name field, or from the folder name if name is missing.

Codex's own $CODEX_HOME/skills folder is not an OpenClaw skill root. To bring those skills over, run openclaw migrate plan codex to list them, then openclaw migrate codex to copy them into your workspace.

Install a skill

OpenClaw has a native installer. The examples below use frontend-design from anthropics/skills, located at skills/frontend-design.

Option 1: openclaw skills install

Git and local installs expect SKILL.md at the root of the source, so for a skill inside a larger repository, clone first and point the installer at the folder:

git clone --depth 1 https://github.com/anthropics/skills.git skills-src
openclaw skills install ./skills-src/skills/frontend-design --as frontend-design

This installs into the active workspace's skills/ directory. Add --global to install into ~/.openclaw/skills, where every local agent can see it unless an allowlist narrows that. If the repository root is itself the skill, you can skip the clone:

openclaw skills install git:owner/repo@ref

The slug comes from the name in SKILL.md when it is valid, and --as overrides it. Note that openclaw skills update tracks ClawHub installs only, so reinstall Git and local sources to refresh them.

Option 2: the skills CLI

npx skills add anthropics/skills --skill frontend-design -a openclaw

The skills CLI lists openclaw as a supported agent, with skills/ for project installs and ~/.openclaw/skills/ when you add -g. OpenClaw's own documentation does not mention this tool, so confirm the skill appears afterward.

Option 3: GitHub CLI

gh skill install anthropics/skills frontend-design --agent openclaw
gh skill install anthropics/skills frontend-design --agent openclaw --scope user

openclaw is one of the documented --agent values. GitHub's manual does not say which folder it writes to for OpenClaw, so check that the skill loaded. It needs GitHub CLI 2.90.0 or later, where gh skill is a public preview.

Option 4: copy the folder

git clone --depth 1 https://github.com/anthropics/skills.git skills-src
mkdir -p skills
cp -r skills-src/skills/frontend-design skills/frontend-design
rm -rf skills-src

Run this from the agent's workspace root. Use ~/.openclaw/skills/ instead of skills/ for a copy shared by all agents.

ClawHub

ClawHub, at clawhub.ai, is OpenClaw's public registry for skills and plugins. It stores versioned bundles with SKILL.md plus supporting files, and tracks versions, tags, changelogs and security scan summaries. You can use the native commands to search, install and update:

openclaw skills search "calendar"
openclaw skills install @owner/<slug>
openclaw skills update --all

Replace @owner/<slug> with the publisher and slug shown on the skill's ClawHub page. A separate clawhub CLI, installed with npm i -g clawhub, handles login and publishing. It also has its own clawhub install, which writes into ./skills under the current directory and records versions in .clawhub/lock.json.

Check that it loaded

Run openclaw skills check if a skill is installed but does not show up for an agent, or if a required binary or credential may be missing. A skill can be present on disk but not ready, because its requirements are unmet, or not visible, because an agent allowlist hides it.

OpenClaw takes a snapshot of eligible skills when a session starts and refreshes it when its watcher sees a SKILL.md change. In the Control UI, type $ in the composer to search the skills available to the current agent, or write a prompt such as "Use $frontend-design to review this page."

Update and remove

• Update ClawHub skills with openclaw skills update --all, or add --global for shared ones.
• Update Git or local installs by reinstalling them.
• Remove a skill by deleting its folder from the skills directory.
• Turn one off without deleting it by setting enabled: false for it under skills.entries in ~/.openclaw/openclaw.json.
• Limit what each agent sees with agents.defaults.skills or an agents.entries.<id>.skills list. A non-empty per-agent list replaces the defaults rather than adding to them.

Troubleshooting

• The skill does not appear: confirm the folder contains SKILL.md with name and description, then run openclaw skills check.
• It is installed but not ready: look at metadata.openclaw.requires. Binaries must be on PATH, and if the agent runs in a sandbox, they must exist inside the container too.
• An older skill ignores its requirements: the pre-July 2026 metadata.clawdbot block and .clawdhub metadata folder are no longer read. Rename them to metadata.openclaw and .clawhub.
• A different version loads than expected: a same-named skill higher in the table wins, so check the workspace skills/ folder first.
• Edits do not show up in a running chat: the list is refreshed on the next agent turn after the watcher notices a change, and an existing session keeps its snapshot until then.

Safety

OpenClaw's documentation says to treat third-party skills as untrusted code and to read them before enabling them. A skill can steer an agent that has access to your files, tools and chat accounts, so read SKILL.md and everything bundled with it before installing something you did not write. ClawHub shows scan results from VirusTotal and ClawScan on each skill page, and openclaw skills verify @owner/<slug> checks a skill's trust information from the command line. These are aids rather than guarantees. Prefer sandboxed runs for untrusted inputs and risky tools, and keep third-party skills in one workspace until you trust them rather than in the shared ~/.openclaw/skills folder.