AI models have gotten really good this year. I use Codex and Claude pretty much every day, and I wanted a harness and terminal I can customize however I want, so I made both.
r7-Harness is my agent harness, built on OMP. It has task titles with time estimates, separate profiles for each agent, and a reply style with colored markers for key points, finished work and questions.
r7-Shell is a lean custom terminal for Windows, made for running agents in WSL, as well as anything else you'd run in a terminal. Sessions keep running in the background, new text fades in and moved text glides into place, and a window flashes when its agent finishes a response.
Running r7-Harness in r7-Shell, you also get pictures inside replies, bigger titles and mouse editing in the input box.
At this point the models are so good that I have to stop myself from making more useful custom tools and actually work on my games. So, let me get to it.