My/AI/Rule/Store memory where every agent can read it

• When something is worth keeping past this session — a preference of mine, a project fact, how I want a piece of work done — write it into the git repository or knowledge garden the work is in, as a tracked file any tool can open. In this graph, a standing instruction for agents goes under My/AI/Rule, which agents read at load; a preference of mine goes under My/Pref.
• I move between coding agents all day, and each one keeps its notes where only it looks. Claude/Code saves to ~/.claude/projects/<project>/memory/ and reads ~/.claude/CLAUDE.md. Codex reads ~/.codex/AGENTS.md. CursorAI has its own project rules. A note left in any of those is invisible to the next agent I open, and it sits outside the repository the work lives in.
• Leave your harness's own memory system alone until I authorize the memory going there.
• Ask me where a memory belongs when the right page isn't obvious.