luban-skill

Why we included this project

Most people who publish agent skills eventually hit the same wall: the skill runs fine on their own machine, but a stranger reading the repo cannot tell what it does or how it compares to similar skills. Luban treats that as a craft problem. It challenges whether the skill's premise is even worth defending, pulls up competing skills online, scores the structure and checks real output rather than trusting a green CI badge, and only keeps changes that pass a frozen baseline. Along the way it leaves you with a 13-section report, a screenshot-ready certificate showing before and after scores, and verification scripts committed to your repo, so your standards become documented project rules. It installs as a Claude Code plugin or a Hermes skill and has a real-world track record: its first job took ai-news-radar from v0.6 to v0.7.0 in one conversation.