Welcome

uze is a compatibility and distribution layer for agent tooling. You install a plugin once; uze stores its bytes centrally and delivers it through each harness's most native mechanism — a real plugin where one exists, native capabilities where it doesn't, a safe adapter only as a last resort.

The problem

Coding agents — Claude Code, Codex, OpenCode, Antigravity — each have their own plugin format, their own skills directory, their own way of reading project instructions. Maintaining the same skill or the same project context four times over is how agent setups rot.

The approach

1. One plugin bytes, one place. The Store owns the canonical package; nothing a harness reads is ever written from anywhere else.
2. One project context. AGENTS.md is the portable baseline, projected into each harness's own bridge/config format instead of maintaining four files.
3. Per-harness native delivery. Every capability is delivered through the most native mechanism that preserves its semantics — and when a harness genuinely cannot preserve them, uze says so honestly instead of degrading silently.

The loop

$ uze setup            # detect + provision the four harnesses through official routes
$ uze flow@ai          # add a plugin from a marketplace to this project
$ uze doctor           # diagnostics: store, harnesses, attachment health
$ uze context reconcile   # project context across every detected harness

Or just run uze with no arguments for the terminal UI.

Next steps