alfredosdpiii/loki

Deterministic guardrails for AI coding agents: blocks unsaf…