CourseAgentic AI Engineer: The Complete Course · Module 2: Model Behaviour for Agents · part 14 of 97
Part 14 · Module 2: Model Behaviour for Agents

Lab: A versioned prompt, tested on two models

22 min read·13 Oct 2026
Module 2 · Lab · About 60 minutes · Needs: lessons 1 and 2 (rule checks, the pushback test, the six prompt sections, the OUTCOME: contract)

What you are building

system_v1.md reviewed file loader numbers, hash 7 cases suite course model second model swap report verdict rules.py AGENT_RULES, labels
The finished piece: a reviewed prompt file that BrightCart's code (blue) fills from rules.py and hashes, seven cases run on two models (amber), and a report with a verdict (green).

Lesson 2 ended with a prompt built inside a notebook. Dev's question still stands: where does it live, and who can change it? This lab gives the answer a team can review. The prompt becomes a file, returns_desk/prompts/system_v1.md, with placeholders where the numbers go, so every limit still comes from code. A loader fills it, names it by a hash, and refuses a file that changed without a new version. Seven fixed cases, including Tom pushing back and Tom miscounting his days, run against the course model and a second model, and a short report says whether the swap is safe. Everything at BrightCart is simulated.

The rest of this course is yours to keep

This course is bought on its own, once, and stays readable afterwards, including the parts added to it later.