The problem

Skills accumulate in ~/.claude/skills and ~/.config/opencode/skills by copy-paste. Two machines diverge, nothing is versioned, and there is no way to tell what a given machine is actually running.

Approach

A registry plus a CLI. Skills are declared, versioned, and cached locally, so rsk can answer “what does this machine have, and what should it have” instead of leaving it to memory.

What I wanted to learn

Building the whole framework around the skills, not only the skills: the CI/CD pipeline, the versioning, the rsk CLI, and now its MCP, which helps discover skills, suggests them for each project, and installs them.