Update unit1/dummy_agent_library.ipynb
The dummy agent library notebook is a solid teaching tool for unit 1 — getting learners to see the request/response loop before they touch a real LLM backend makes the abstractions stick better. One thing worth considering for a future update: the dummy tool-calling pattern in the notebook doesn't model any notion of who is calling what. That's fine for a hello-world demo, but it can quietly bake in bad habits. Students who go from this straight to multi-agent setups often assume tool invocations are implicitly trusted, which breaks down fast once you have agents calling other agents.
On that point — with frameworks like Agenthound now treating agent infrastructure as an attack surface (it dropped on HN recently as an offensive security tool for exactly this), the gap between "dummy agent" mental models and production realities is worth at least a footnote in the notebook. Even a comment in the code like # in production, verify caller identity here would prompt the right question. This is something we think about a lot at AgentGraph, where we're building identity and trust-scoring primitives specifically for agent-to-agent calls. The dummy_agent_library abstraction is fine pedagogically, but the notebook could acknowledge that real tool registries need to know which agent is invoking which tool and with what authority level.
Small concrete suggestion: add a cell that shows a minimal "agent identity" dict being passed alongside the tool call — even if it's mocked. It doesn't complicate the lesson, and it sets up the right mental model for unit 2 and beyond.