Claude Skills

The Problem

I'd built up a folder of Claude Code skills while shipping real projects — but they were tangled into my monorepo: hardcoded paths, my own domain names, references to private projects. Useful to me, useless (or leaky) to anyone else.

The Collection

claude-skills is the cleaned-up, open-source version: self-contained skills you drop into any Claude Code setup. One skill is one folder — copy it, and the agent picks it up from the description in its SKILL.md. Seven so far, grouped by where they help across a product's life:

Build — SVG diagrams with collision-proof layout math, test-first enforcement, breaking out of repeated failed-fix loops
Ship — secret scanning, and platform-aware pre-deploy checks for Cloudflare Pages / Cloud Run / Vercel
Grow — a technical SEO audit and a post-launch growth playbook

The Publish Gate

The hard part of open-sourcing personal tooling is not leaking anything. So every skill has to pass an automated gate before it ships: a script scans for hardcoded secrets and for personalization — author names, private domains, machine paths — and refuses anything that points at files only on my laptop. The catalog is generated from the skills themselves, so the index can't drift, and an AGENTS.md plus llms.txt make the repo machine-readable for other AI tools.

Results

A clean, growing collection anyone can use, with a safety net that makes adding the next skill a three-command routine. Built on the same conventions as my Claude Code Workflow system.

View the repo on GitHub