OpenClaw Skills CLI - Installing and Managing Agent Capabilities

One of the things that makes OpenClaw interesting as an AI agent platform is how it handles extensibility. Rather than baking every capability into the core agent, OpenClaw uses a skills system - discrete bundles of knowledge and behaviour that you can install, update, and manage independently. The skills CLI is the tool that makes this manageable.

If you've worked with package managers like npm or pip, the mental model transfers directly. You search for skills, install the ones you need, update them when new versions come out, and inspect what's currently available in your workspace. It's not complicated, but it's worth understanding how the pieces fit together.

What Skills Actually Are

Before getting into the CLI commands, it helps to understand what a skill is in OpenClaw terms.

A skill is a packaged set of instructions, tools, and configuration that gives an agent a specific capability. That might be something like understanding how to interact with a particular API, knowing how to follow a specific development workflow, or having domain knowledge about a subject area.

Skills sit in your workspace's skills/ directory. When an agent runs, it checks which skills are available and uses them to inform how it approaches tasks. The agent doesn't blindly follow every installed skill - it picks the relevant ones based on what you're asking it to do.

This is different from something like plugins in a traditional application. Skills aren't just code that runs. They're a mix of instructions, configuration, and sometimes tool definitions that shape the agent's behaviour. Think of them more like specialised training modules than software plugins.

Searching ClawHub

ClawHub is the central registry where skills are published. The search command is your starting point.

openclaw skills search "calendar"

This searches ClawHub for skills matching "calendar". You can also browse without a query:

openclaw skills search

That gives you the default feed - a curated set of popular and recently updated skills. If you want more results:

openclaw skills search --limit 20

And for scripting or piping results into other tools:

openclaw skills search --json

The JSON output is useful when you're automating workspace setup. We've built setup scripts for client environments where a new developer runs a single command that searches for and installs all the skills their team uses. Saves the "which skills do I need?" conversation that happens every time someone joins the project.

Installing Skills

Once you've found a skill you want, installation is straightforward:

openclaw skills install <slug>

The slug is the skill's identifier on ClawHub. It downloads the skill folder into your active workspace's skills/ directory and it's immediately available to agents running in that workspace.

If you need a specific version:

openclaw skills install <slug> --version <version>

And if you need to overwrite an existing installation (say, you've made local modifications you want to discard):

openclaw skills install <slug> --force

The --force flag overwrites the existing skill folder entirely. Any local customisations you've made to that skill are gone after this. I mention it because I've seen people use --force to "fix" a skill that wasn't working, only to lose modifications they'd forgotten about. If you've customised a skill, back up your changes before force-reinstalling.

Updating Skills

Skills evolve. The authors fix bugs, add capabilities, and update instructions. Keeping your installed skills current is good practice.

Update a specific skill:

openclaw skills update <slug>

Or update everything at once:

openclaw skills update --all

The --all flag only updates skills that were installed from ClawHub. If you've manually placed a skill folder in your workspace, it won't be affected.

How often should you update? It depends on how actively the skill is being developed. For production workspaces, I'd suggest pinning to specific versions and updating deliberately after testing. For development workspaces, running update --all periodically keeps you current without much ceremony.

Inspecting Your Workspace

The local inspection commands don't touch ClawHub. They look at what's in your workspace right now.

List all skills visible to the current workspace:

openclaw skills list

For more detail:

openclaw skills list --verbose

To see which skills are actually eligible for use (matching the current agent configuration):

openclaw skills list --eligible

The --eligible flag is the one I use most. Not every installed skill is necessarily active for every agent configuration. This command shows you what's actually available, which is what you care about when troubleshooting why an agent isn't using a capability you expected it to have.

Get details about a specific skill:

openclaw skills info <name>

And check the health of your skills installation:

openclaw skills check

The check command validates that your installed skills are properly structured and compatible with your OpenClaw version. Run this after major OpenClaw upgrades - skills that worked on the previous version might need updating.

How This Fits Into Agent Development

At Team 400, we use skills as part of our standard approach to building AI agent solutions. When we're setting up an OpenClaw workspace for a client, the skills configuration is one of the first things we work through.

The pattern looks like this: start with a minimal set of skills that match the agent's core job. Deploy. Watch how the agent handles real requests. Then add skills as gaps emerge. Trying to install every potentially useful skill upfront tends to create noise - the agent has too much context and starts applying skills in situations where they're not relevant.

We've also found that the skill system is a good way to standardise agent behaviour across teams. Rather than each developer configuring their agent differently, you define a set of skills that all agents in the project should have. New team members install the same skills and get consistent agent behaviour from day one.

The CLI's JSON output modes (--json on most commands) make it possible to automate skill management. We've integrated skill checks into CI pipelines - verifying that a workspace has the expected skills installed and that they're the right versions. It's a small thing, but it prevents the "it works on my machine" problem from extending to agent capabilities.

Tips from the Field

A few things worth knowing that aren't immediately obvious from the documentation.

The install command from the CLI downloads skill folders directly from ClawHub. There's a separate install path for skills triggered through the gateway (onboarding or settings), which uses a skills.install request. They achieve the same result but through different mechanisms. If you're scripting installations, the CLI is what you want.

When list is run without any subcommand, it's the default action. So openclaw skills and openclaw skills list do the same thing.

All the inspection commands (list, info, check) write their output to stdout. This is by design - it means you can pipe output to other tools or capture it in scripts. The --json flag on each of these gives you machine-readable output that's easier to parse programmatically than the formatted default.

If you're building agentic automation solutions and want agents that can grow their capabilities over time, the skills system is how you do it cleanly. It keeps agent capabilities modular, versioned, and manageable - which matters a lot more than you'd think once you're running agents in production across multiple teams and use cases.