Agent Safety

Harness engineering is the executable control layer around AI coding agents: repository knowledge, capability boundaries, context bundles, architectural invariants, runtime evidence, normalizing …