agent-browser is my favorite way to create a good feedback loop for agents.
Why I use it
It fires up a headless browser that agents can navigate and gives them access to DevTools, the console, logs, and network traffic.
I love prompting: Use agent-browser to verify that this works on various viewports. Then it goes ahead and does it.
I love that it doesn't hijack my computer or inflate my coding agents' resource use the way computer use does.