rollytucan/eve1 ? reverse-engineered prompt

Build me a framework called eve for creating durable AI agents in a really simple, inspectable way. I want the main idea to be that the filesystem is the authoring interface, so an agent lives in a normal folder with things like always on instructions, callable tools, reusable skills, message channels, and scheduled jobs in obvious places. It should feel easy to scaffold a new agent from the command line, work inside an existing project too, and start up in a local interactive terminal experience so someone can test the agent right away.

Please include a minimal working example, like a weather assistant with mocked data, plus a clear way to choose the model and runtime config. The framework should make it straightforward to add channels like HTTP, Slack, or Discord later, and support more advanced things like human approval steps, subagents, and recurring tasks. Keep the docs close to the code so coding agents can read them locally, and check current docs online if you need to fill in gaps.