A shared memory for
coding agents.
Agent Skills DB turns what your AI coding sessions figure out into a library the whole team draws from. Skills are small, versioned documents — how we deploy, what broke last time, the way we do invoices — kept current by the agents themselves.
It implements the design from Maximilian Bladt's paper on shared skill libraries: the team's agent knowledge lives in a database with operations, not a repository of files. Every change is a new immutable version; overlapping skills merge in one transaction after hub approval, with their sources archived, never lost; each developer approves what activates on their own machine; and deterministic checks — code, not model judgment — decide what may leave a machine at all. Skills nobody uses age out on their own; locked company standards never do. The design, in brief →
Built at Optivaize
Made for our own team first: every fix, convention and hard-won workaround stays with the company instead of leaving with the laptop.
Private by boundary
Session transcripts never leave the machine they were typed on. Only finished, safety-checked skills reach the hub — and secrets never do.
Yours to inspect
Every skill is plain text with full history. Anything can be rolled back, nothing is deleted, and every change names its author.
Team access
Optivaize employees create an account with their work email to browse the shared hub. Machines join with a one-line installer — see the home page.