Agents often fail at boundaries: when context moves from one agent to another, from a prompt to a tool, or from local code to an MCP-backed capability. This session introduces the idea of an “agent passport”: a typed, traceable handoff envelope that carries intent, permissions, memory scope, guardrail status, and trace identity across agent workflows.

Using LLM4S, we will walk through a compact maintainer-assistant workflow: an open source issue is classified, routed to a documentation, code, or safety specialist, checked through guardrails, connected to an MCP capability, and recorded as an inspectable trace. The focus is a production pattern for making agent delegation auditable, portable, and safer to operate.

Attendees will learn how Scala’s type system, functional state, typed tools, provider abstraction, MCP integration, and observability support enterprise-grade agent systems where humans can understand why a handoff happened and what evidence shaped the final output.