We need to be able to call Agents inline. Thoughts: 1. Agents should exist one per $token e.g. scoped to a ticker name. Each would have its own runtime and meaning.md (for example). This scopes ONE Agent per thread. or 2. Agents can be anything: The platform agent, other agents that do other things - external ones. Which is the better design choice for inline agents? Token scoped or non-token scoped?