What it changed
- A hosted registry and CLI for agent skills, live and used every day.
- Concepts the standard didn't have: dependencies between skills, a hash-checked lock file, three-way merge and kits.
- Works with Claude Code, Codex, Cursor and other agents.
The situation
Agent skills are how you teach an AI agent the way your company works. They're also just folders: copied between projects, edited in one place and not another, with no versions, no dependencies and no way for a team to share them properly. I hit that wall myself, with dozens of skills across every client project.
What I built
HappySkills, a hosted registry and CLI that does for agent skills what npm did for JavaScript, without changing the skill format. Skills declare dependencies with version ranges. A lock file pins every version and hash, so a setup can be rebuilt exactly. Local edits are merged, not overwritten. Teams get private catalogs with roles and per-skill access. And you drive all of it by talking to your agent, because the CLI was designed for agents first.
I built it end to end, as a company of one with agentic programming: product, design, infrastructure, security, payments and the CLI. The full verdict, with how it compares to the alternatives, is in The Stack.
Why it matters to you
HappySkills is the proof behind the claim: one senior engineer with agents can design, build and run a real product, not a prototype. It's also how I bring a company's know-how into its agents when I work with a team: as versioned skills, shared and kept in sync.