Build a ClawHub Skill from Scratch with Claude Code

The Problem

You've built workflows — file automation, hype tracking, reverse prompting. They work great on your machine. But what if someone else wanted the same hype tracker? They'd have to build it from scratch. What if you could package it so anyone with OpenClaw could install it with one command?

What You're Building

A ClawHub-ready skill. You'll take an existing workflow, package it as a portable SKILL.md, test it locally, and prepare it for the marketplace. From user to creator.

Milestone 1: Understand the Skill Anatomy

Before you build a skill, you need to know what one looks like.

Show me the structure of an OpenClaw skill. Explain what SKILL.md is, what the YAML frontmatter contains, and how the markdown body works. Find an example skill on ClawHub and break it down — show me every field and what it does. I want to understand the format before I write my own.

What Claude Code does: It researches the SKILL.md format and breaks it down. Every OpenClaw skill is a folder centered on a SKILL.md file — YAML frontmatter for metadata and markdown body for instructions. The frontmatter tells OpenClaw what the skill needs; the markdown tells the agent how to use it.

Try it: Read through the breakdown. You should understand: name, description, version, required env vars, required binaries, and the instruction body.

Milestone 2: Package Your Workflow

Turn your existing workflow into a portable skill.

Take my hype tracker workflow (from Lesson 7 — the tool tracker with JSON database, cron reminders, web research, and verdict reports) and package it as an OpenClaw skill. Create a SKILL.md file with proper YAML frontmatter and a clear markdown body that tells the agent how to use the hype tracker. Put it in a new folder called "hype-filter" under my OpenClaw skills directory.

What Claude Code does: It extracts the logic and writes it as a portable SKILL.md. The key shift is from "code on my machine" to "instructions any agent can follow." The markdown body describes the workflow; the YAML frontmatter declares requirements.

Try it: Read the SKILL.md — it should describe your hype tracker in a way any OpenClaw agent can follow.

Milestone 3: Write the Manifest

The SKILL.md is the brain. The manifest is the label on the box.

Complete the YAML frontmatter for my hype-filter skill. Set: name to "hype-filter", description to "Track AI tool launches and auto-check if they're still relevant after 8 weeks", version to "1.0.0". Add required binaries (curl for web research), primary emoji "magnifying_glass", homepage URL, and OS compatibility (macos, linux). Make the slug URL-safe and lowercase.

What Claude Code does: It fills in the metadata. The manifest is how your skill gets discovered and validated — name and description appear in search, version enables updates, requirements tell users what they need before installing.

Try it: Validate the frontmatter — all required fields present. Slug is lowercase with hyphens only.

Milestone 4: Test the Skill Locally

Before publishing, make sure it works from scratch.

Test my hype-filter skill by installing it locally in a clean OpenClaw workspace. Use the clawhub CLI to install it from the local folder (not from the marketplace). Verify: 1) The skill appears in my agent's skill list, 2) I can trigger it by messaging "Track: TestTool - AI - test.com", 3) The hype tracker JSON file gets created, 4) The cron job gets scheduled.

What Claude Code does: It installs your skill as if downloaded from ClawHub. Local installation simulates the real user experience — if instructions are unclear, the agent won't know what to do. This is your pre-flight check.

Try it: Message: "Track: TestTool - AI - test.com." Full workflow should fire — JSON entry created, cron scheduled.

Milestone 5: Prepare for Publishing

The skill works. Let's get it ready for the world.

Prepare my hype-filter skill for ClawHub publishing. 1) Write a clear marketplace description (2-3 sentences), 2) Add usage examples to the SKILL.md body, 3) Verify the bundle is under 50MB, 4) Run a format validation check, 5) Show me the exact publish command. Don't publish yet — just prepare.

What Claude Code does: It polishes for release. Publishing requires: valid SKILL.md, GitHub account at least one week old, bundle under 50MB. The publish command (clawhub publish) handles the upload when you're ready.

Try it: Validation passes. Description reads well. Command is ready.

What You Built

Remember when your workflows only worked on your machine? You just:

• Learned the SKILL.md format
• Packaged an existing workflow into a portable skill
• Wrote a manifest with requirements and metadata
• Tested locally to verify it works from scratch
• Prepared for ClawHub publishing

You went from user to creator. Every skill on ClawHub was made by someone who did what you just did.

Take It Further

• Publish it — run the command and share with the community
• Add configuration — let users customize the check interval
• Package more workflows — file automation, reverse prompting, security audit are all potential skills