Aside from the fact that an agent can use this, how is this better than Claude for Chrome? With all the capabilities that Claude in Chrome has, it almost seems better to just have an agent come up with a prompt you can send CiC -- quickly send the prompt, get feedback and send it back to the agent. I know, it defeats the purpose of AI but seems like a bit of extra setup and I'm not finding a ton of value in it.