comparison
inference.sh skills vs manual management
copying SKILL.md files between repos works until it doesn't. the registry gives you versioning, security, and discovery.
| manual | inference.sh | |
|---|---|---|
| works in any runtime | ||
| version pinning | ||
| security scanning | ||
| one-command install | ||
| automatic updates | ||
| discovery and browsing | ||
| fork and evolution tracking | ||
| verified publishers | ||
| cross-team sharing |
the real cost of manual
you copy a skill file. the author improves it. you never know. someone shares a skill on twitter. you copy it without scanning. your team uses five different versions of the same skill. nobody knows which is current.
the registry solves all of this. belt skill use installs a versioned, scanned, verified skill in one command. belt skill upload shares it with everyone. updates flow automatically. forks are tracked.
when manual is fine
if you're one person with a few personal skills that you maintain yourself, manual management is simple and works. the registry becomes valuable when skills are shared across a team, when you want to discover new skills, or when trust matters (you're installing skills from people you don't know).
frequently asked questions
do I have to stop using my existing skills?
no. your existing SKILL.md and CLAUDE.md files keep working. publish them to inference.sh when you want versioning, scanning, and discoverability.
how do I publish an existing skill?
belt skill upload. it scans the skill, assigns a version, and makes it discoverable. one command.
what if I just have a few skills for myself?
manual management works fine for one person with a handful of skills. the registry shines when skills need updates, when teams share them, or when you want to discover what others have built.
ready to ship?
start with the hosted platform. deploy your own when you're ready.
we use cookies
we use cookies to ensure you get the best experience on our website. for more information on how we use cookies, please see our cookie policy.
by clicking "accept", you agree to our use of cookies.
learn more.