Connect the Claude or Codex plan you already have, point it at a repo, and your agents get to work inside a living world you build and command like an RTS.
Go to github.com/summercraft/summercraft and read the README. Check if I already have Summercraft installed. If not, set it up, read the skill guide, and configure the MCP.
Each one slots into the same simple hierarchy. That structure is the truth underneath, and it never changes.
One AI chat = one NPC.
A folder of agents = a crew of NPCs.
A codebase = a building.
Several related repos = a district.
All your groups + repos = one living world.
The hierarchy never changes. A theme just picks what each one looks like. Swap it live, or build your own.
Make your world public and anyone can drop in and watch your agents work, live. Only the world ever syncs. Your code, your diffs, and your files stay on your machine.
It only works if you trust it, so the boundaries are simple and hard.
Agents run on the subscription you already have. The metered API key is stripped from every one, so nothing can quietly run up a bill.
Each agent works in its own git worktree, never your live files. If one cannot get an isolated worktree, it refuses to run.
The one thing that ever leaves is an anonymized pulse: a path, a size, a state. No code, no diffs, no file contents, ever.
Click an agent and the camera drops into a first person view. You talk to it out loud, with its branch, PR, and diff as context. It works, commits, and reports back.
Born at a hackathon. Single player works today, multiplayer worlds are on the way. Open source, and meant to be hacked on.
Go to github.com/summercraft/summercraft and read the README. Check if I already have Summercraft installed. If not, set it up, read the skill guide, and configure the MCP.