Skillbox uses Jev to score agent skills for relevance
Skillbox, a self-hosted versioned skills library for AI agents, connects to the TypeSafe API so an agent can describe its task and Jev scores every skill for relevance, saving $500 a month in wrong tool calls.
AI agents are only as good as the skills you give them. right now those skills live in scattered markdown files, personal dotfiles, and tribal knowledge that disappears when someone leaves the team. Skillbox is a self-hosted versioned skills library for AI agents. built with React, Bun, Hono and PostgreSQL. what it does: every skill gets immutable revisions. markdown editor in the browser. conflict detection and restore. you can see exactly what changed and roll back. profiles with scoped permissions - who can read, write, propose, or archive each skill. revocable client keys with usage reporting. HTTP MCP endpoint and a Node/Bun stdio bridge. Claude Code, Codex, Cursor all connect to it. the Jev part that saves $500/month in wrong tool calls: optional task-aware recommendations powered by your own TypeSafe API key. an agent describes its task, Jev scores every skill in the library for relevance, and the agent gets the right one - not just the top-k by keyword match. your own key. your own data. never sent to a third party account.
This post quotes another post: x.com/p0lybender/status/2105657133702193458