The Memory Design Prompt

Design persistent memory for this agent.

Agent purpose: [DESCRIBE]
What should be remembered: [DESCRIBE]
Who the memory belongs to: [USER / ORGANISATION / AGENT / SHARED]
Privacy constraints: [DESCRIBE]

Produce:
1. Memory schema: what a memory record contains
2. Write policy: what triggers a write, and what the agent should
   NOT write
3. Read policy: when memory is consulted, and how relevant records
   are selected
4. Organisation: flat, hierarchical, tagged, or indexed
5. Update vs append: when a memory is revised rather than added to
6. Expiry: what goes stale and how it is retired
7. Conflict: what happens when a new memory contradicts an old one
8. Scoping: what prevents one user's memories reaching another
9. Auditability: how a human inspects and corrects what was
   remembered
10. Deletion: how a user removes their memories, and what that
    guarantees

CRITICAL constraints to build in:
- Never store credentials, tokens, or secrets in memory
- Memory is replayed into future contexts — anything written once
  is repeated indefinitely
- Personal data in memory is subject to the same regulation as
  personal data anywhere else

Also tell me: what am I proposing to remember that a system of
record already stores? Those should be queries, not memories.