I use multiple AI coding agents (Claude, Cursor, Codex) and kept losing product context every time I switched. Not to mention, agents write technically correct code sometimes violate the product's design philosophy. I build AI tutors that force reasoning from the student, and one agent even added a 'Skip Reasoning' button goddamnit. LedgerSync is a file-based protocol that grounds agent decisions to philosophy docs and shares context across tools. No server, no accounts — just files your agents read.