What should the default world contain for an AI agent?

I’m building mikrokhoros, an open-source Swift runtime for persistent object worlds. An agent observes a concrete world, acts through bounded tools, receives verified consequences, and leaves a replayable trajectory. It’s free to use under Apache-2.0 and is currently pre-release.

The current default world provides Athena, an Objective Board, a Library, a Warehouse, and a Marketplace. I’m deciding what belongs in the default world and first examples.

If you build agents, simulations, games, or evaluation environments, I’d value specific feedback:

• What objects or rules make a world useful on day one?

• Which mechanics make behavior easiest to understand and test—locality, containment, explicit authority, resource constraints, durable records, or replay?

• What should the first-run documentation make explicit for someone coming from an agent framework?

Source and runtime documentation: https://github.com/sonatapublisher/mikrokhoros