CODEX WORKED 59 HRS ON THIS MAC APP

I got Codex to work for 59 hours, 59 hours, two days and 11 hours to build me out a full Mac OS app that has a complex Superbase backend, Railway API, services without any intervention. This would not have worked without engineering the loop that allowed Codex to work for so long, specifying exit condition goals, actually configuring the agent's environment to help it use those external services, and sharding a large plan into micro tasks. To do this, I use my orchestration skill M2C1, the unique part about this skill walks you through configuring your agent's access to external services before you start and how it specifies the testing and verification conditions before you specify that exit condition. In this example, I wanted to use the database and off services of Superbase, so I had to give my agent access to that Superbase project, as well as custom skills built to teach that agent how to use the database for this project correctly. I also wanted to use Railway to host worker services, so I had to give it access to the Railway project and similar custom skills. Similarly, for all the other software packages that you will use in your project, you create custom skills to make it easy for your agent to progressively disclose that information, it's working. Secondly, and most importantly, is specifying the verification and test conditions to actually pass that exit goal. First, M2C1 guides you through all the potential logging and use surfaces of that final app. So in this case, it's going to be a Mac OS app, the actual desktop which will be navigated by an agent through computer use. Through all of the user flows of that app. There are many logging services, including the local runner logs from that Mac app, the Superbase logs, and the Railway logs. From all these different verification and testing services, you make a custom verification and testing skill that must be run after each iterative PR in your project before proceeds to the next. And after all code changes have been made, one comprehensive end-to-end test observing all of those log and usability services before the agent completes its loop. If you want to use this technique to get your agents to do long running autonomous looping coding, this whole system is available in my school community. The link is in my bio and I'd love to get working.