Build additional protections around ExitPlanMode

I just witnessed an unprecedented act of disregard for rules on the part of the AI that I think is worth taking a second look at as a community.

I can't find any Anthropic docs on ExitPlanMode (really?) so figured it might be a good place to have the swarm place a secondary check on plan mode on behalf of the user. Or some other solution. But certaintly this is not a new norm I can expect to live with right? In this particular scenario, I told the AI to stay in Plan mode because I had more "weird details" to give it, and even the most advanced AI models would not have had the info needed to proceed (in this case, the weird detail was that I was integrating this work into an existing software stack deployed to AWS).