Creator here. MikeBrain came from building AI agents that kept going off the rails. We needed:

1. *Trust Ladder* - agents start sandboxed, earn permissions through demonstrated behavior 2. *LEASH Protocol* - hard boundaries (rate limits, kill switches, no-go zones) 3. *Hall Monitor* - self-reporting discipline where agents flag their own violations 4. *Baton Protocol* - clean handoffs between sessions with no context loss

Been running this in production for 6 months managing construction documentation. Zero unintended external actions.

Happy to answer questions about the implementation or design decisions.