ATU-1.1: Agentic Terminal Use benchmark

A second open benchmark, for coding agents operating a terminal: 20 tasks across six levels, from file operations to git and safety, with a weighted ATU-Score and a separate safety score. MIT licensed, standard library only.