Module · Intermediate · 1h 15m
Module 2: Making it safe to run
The system prompt as a contract, an evaluation set with refuse cases, and approval gates for side effects.
- Claude
- ChatGPT
- Any assistant
- agents
- module
Inside this module
- Write a system prompt with a boundary and a refusal rule
- Build and run an evaluation set
- Gate a side-effect tool behind approval
The full module opens with a membership that covers AI Agent Development.
Sign in if you already subscribe.
Sign inLessons
- 01The system prompt as a contract
Lesson · 25 min · Writing the standing instructions that define what the agent may do, must hand over, and must never do.
- 02An evaluation set with critical failures
Lesson · 25 min · Building the fixed test set that decides whether an agent change ships.
- 03Approval gates for side effects
Lesson · 25 min · Adding a tool that changes the world, and making the agent stop and ask before using it.
