quality

Critique loop

A critique loop has one model do the work and a second model review it before you accept it. A simple way to raise quality without more prompting.

Evals

An eval is a test for AI output: run a prompt or skill over real examples and score the results. Which checks to automate, and which need a judge.