find-skills: The OpenClaw Skill That Turns "How Do I..." Into an Instant Install

With 106,000+ downloads and 483 stars, find-skills by @JimLiuxinghai is the second most-downloaded skill on ClawHub. It does something deceptively simple: it teaches your agent how to find other skills.

The Problem It Solves

Every OpenClaw user has run into this wall. You're in the middle of a session and you think: "There must be a skill for this." You're not sure what it's called, where to look, or how to install it. You interrupt your flow to go browse a registry. Sometimes you find what you need. Often you don't, because you searched for the wrong term.

The deeper problem: your agent doesn't know how to help you here. It can write code, answer questions, and run commands — but it has no instinct for "this sounds like a problem with a published skill solution."

find-skills gives the agent that instinct.

Core Concept

The skill teaches the agent two things:

1. When to look — recognize phrases like "how do I do X", "find a skill for X", "is there a skill that can...", or "I wish I could..."
2. How to look — translate user intent into a precise npx skills find query, present the results, and offer to install

The entire discovery-to-install flow becomes a single, uninterrupted agent action.

The Skills CLI Under the Hood

find-skills wraps the npx skills CLI — the package manager for the open agent skills ecosystem. Skills are modular packages that extend agent capabilities with specialized knowledge, workflows, and tools.

Core commands the agent learns:

# Search for skills matching a query
npx skills find [query]
 
# Install a specific skill
npx skills add <owner/repo@skill>
 
# Install globally, no confirmation prompt (for agent automation)
npx skills add <owner/repo@skill> -g -y
 
# Check for updates
npx skills check
 
# Update all installed skills
npx skills update

Browse the full registry at skills.sh.

Step-by-Step: What Happens During Discovery

Step 1: Recognize Intent

The agent watches for trigger phrases in user messages:

• "how do I make my React app faster?"
• "can you help me with PR reviews?"
• "I need to generate changelogs automatically"
• "is there a way to check accessibility in my design?"

These get mapped to skill searches, not just code answers.

Step 2: Translate to a Search Query

Domain + specific task → npx skills find query:

Step 3: Present Results

When npx skills find returns matches, the agent presents them clearly:

I found a skill that might help: **vercel-react-best-practices**
React and Next.js performance optimization guidelines from Vercel Engineering.

Install it with:
npx skills add vercel-labs/agent-skills@vercel-react-best-practices

Learn more: https://skills.sh/vercel-labs/agent-skills/vercel-react-best-practices

Step 4: Offer to Install

If the user wants to proceed, the agent runs the install without further prompting:

npx skills add <owner/repo@skill> -g -y

-g installs globally (user-level). -y skips confirmation — important for non-interactive agent sessions.

Skill Category Coverage

find-skills is aware of the skill ecosystem's major categories, which shapes how it forms queries:

Why It's the Second Most Downloaded Skill

The answer is reflexivity. Every skill a user installs makes them more likely to install another. find-skills is the connective tissue — it's the skill that helps you find all the other skills you didn't know existed.

There's also a compounding effect: the more find-skills is installed, the more agents actively look for skill solutions to user problems, which drives discovery of the entire ecosystem. It's the closest thing the skills registry has to a built-in recommendation engine.

The 483 stars reflect a different signal than raw downloads. Stars on a utility skill like this come from users who consciously decided it had changed how they use OpenClaw — not just users who installed it opportunistically.

How to Install

# Via ClawHub CLI
clawhub install JimLiuxinghai/find-skills

After installation, the agent immediately begins watching for skill discovery intent. No configuration required — it works out of the box.

Practical Tips

Use specific keywords when searching. "React testing" returns better results than "testing." The more domain-specific your query, the better npx skills find can narrow the field.

Try alternative terms if the first search fails. "Deploy" and "deployment" and "ci-cd" can return different result sets. The skill handles retry logic, but you can guide it explicitly.

Combine with auto-install in non-interactive sessions. The -g -y flags mean agents can install discovered skills without interrupting flow — useful in long automated sessions where skill needs emerge mid-run.

The registry updates continuously. A search that returned nothing three months ago might return several results today. Worth retrying for capabilities you previously couldn't find.

Considerations

It's only as good as the registry. find-skills searches what's been published. If no one has built a skill for your niche use case, the search will return nothing. It's a discovery tool, not a creation tool.

Global vs. project-level installs. The -g flag installs at user level (~/.openclaw/skills/). For team environments or reproducible setups, project-level installs (omit -g) are more appropriate — but require the agent to adjust its install command.

Token overhead is minimal. The skill activates only when discovery intent is detected, not on every prompt. Token cost per session is negligible.

The Bigger Picture

find-skills represents a shift in how AI agents relate to their own capabilities. Instead of treating installed skills as a static configuration set up once at the start of a project, it makes capability discovery a live part of every session.

The implication is significant: an agent with find-skills installed never hits a capability ceiling. Every limitation becomes a search query. Every missing feature becomes a candidate for skill discovery. The gap between "I can't do that" and "I found a skill for that" collapses to a single step.

That's why 106,000 users installed it. Not because it does something flashy — because it makes everything else more discoverable.

View the skill on ClawHub: find-skills