Overview
By the end of this act you can: make parts of the agent’s behaviour non-negotiable, and scope what it’s allowed to touch. So a powerful, probabilistic tool stays one you can trust.
Act 1 gave the agent the right knowledge. This act answers the other half. What is the agent allowed to do? You wrap deterministic guardrails around its probabilistic output.
- Hooks fire automatically at points in the agent’s lifecycle. They’re the one place the agent doesn’t get an opinion. (hands-on)
- Custom agents and MCP scope which tools and files an agent can reach (least privilege), and wire in fresh external knowledge. (demo, via the “adding CDK the right way” vignette)