Harness Engineering 1: Install

​Most people keep their AI work in a pile. A prompt saved in a note. A good result nobody can reproduce. A pattern used four times and never named.

​A harness is the alternative. It is an environment that can create, test, and maintain reusable skills, so the work you do once becomes work the system can do again.

​In this first session, Israel Ayliffe installs one in front of you. Codex or Claude Code, running live. You watch the structure go in, see what each part is for, and leave with a running setup rather than notes about one.

​Running, not finished. Install is step one of a longer process: install, then audit and grill it, then update and improve. Session 2 covers the rest.

​One rule holds the whole thing together: the system may analyze on its own and recommend on its own, but it may only change the harness after a human says yes.

​Before you arrive:

​Have Codex or Claude Code installed and signed in.

​Windows users, install Git and Python first. Mac users can skip that step.

​Start on the strongest model you have access to. Starting lower saves credits in the moment and costs you cleanup passes later.

​Check your usage limits before the session. An install run can eat a full session window.

​This is Workshop 1 of a two-part series. For people already working with Codex or Claude Code. This is not an introduction to AI.