Stop waiting on one agent at a time.

Three tasks, one terminal

One terminal makes agent work serial, yet the inventory bug, the enemy AI and the save system are separate jobs. Here each agent gets a real shell in panes, tabs or windows, with the game streaming beside them, so checking a result takes a glance.

Pre-wired to Unity

Claude Code, Codex and Cursor get the Unity MCP server registered automatically, so they open scenes, press Play, read the console and take screenshots. The other three run as plain shell agents.

A checkout each

Two agents in one checkout step on each other, so a session can start in its own git worktree with your uncommitted changes and the Library copied in.

Or an agent makes a scratchpad with unity_scratchpad: an empty project at your Unity version, streamed beside your game.

Agents that start agents

Through satura_exec an agent opens helper agents in new tabs with a task, messages them and gets their reports back in its own terminal. It can run at most four and close only its own.

It can also click through the app while your cursor stays put. Ask it to show you, and Drift, the Satura Space mascot, does each step on screen.

Runs that survive the app

Terminals live in a detached daemon that owns the pty, so quitting or updating the app doesn't kill an agent. Reattaching replays the scrollback, and Claude Code tabs resume their sessions after a relaunch.

Knowing when to look

A finished turn flashes its tab green with a sound; a permission stop flashes amber, even in background tabs. Nine sounds ship, including the Half-Life HEV beeps.

The feedback overlay sends an agent your current view with a note, so you can point at a problem instead of describing it.