This is a real world Ralph loop, shipping a new feature. Check out how it’s using the agent-browser skill to do acceptance tests in the browser. I’m using /handoff in Amp instead of a bash loop with amp -x. This gives me observability into the actual execution of each loop, while still getting a clean context slate each time. I can visualize all the tasks in the browser as they execute. I feel like I’m living in the future. The Ralph and agent-browser skills are here: https://t.co/9Rsg3NbSfw