Security and Trust Config

The security model that makes it safe to give your agent real access. Authenticated channels (your private DM) vs information channels (group chats, email) — the line that prevents your agent from acting on instructions from strangers. Prompt injection defense. File safety rules that make trash > rm the default. A trust ladder: read-only → internal writes → external reads → external writes with approval → earned autonomy. Built for operators who give their agents real power.