Harness Engineering: Building the World an AI Agent Works In

The same model can score 15 points apart depending on what surrounds it. This course teaches the environment around an AI agent: how interface design and harness choice move results, how to make a repository legible with short instruction files and lint rules that explain their own fixes, how permissions, hooks and OS sandboxes contain an agent, and how to test a harness change with repeated trials, paired comparisons and pass^k instead of one lucky run.