Don't store trust, derive it: a trust model for agent-written docs

Your coding agent writes persistent knowledge now. CLAUDE.md files, memory directories, runbooks,...