Appearance
Day 4: Vibe Coding Agent Security and Evaluation
This is the day that decides whether what you've built for the last three days is a demo or something you'd actually trust. An agent that occasionally produces malformed data, or that can be talked into ignoring its instructions by something it read, isn't ready for a real user, no matter how good it looked in your last test run. Today you'll put real guardrails in place: schema validation the agent's output has to pass, a defense against the injection attacks that are unique to systems that read untrusted content, and an actual evaluation process instead of "it seemed to work when I tried it."
What you'll cover
- Getting Reliable Structured Data , schema validation with automatic self-correction, so bad output gets caught instead of passed downstream.
- Defending Against Prompt Injection , the attack vectors unique to agentic systems, and a real defense-in-depth stack against them.
- Testing & Evaluating Agents , building an eval suite that tells you whether a change made things better or worse, instead of guessing.