What Lore is for
Learn which facts, preferences, decisions, and rules should become durable memory.
Lore is for information that should survive one chat, one terminal session, or one agent runtime.
Good candidates
Good candidates:
| Keep in Lore | Reason |
|---|---|
| user preferences | they should shape future answers across tools |
| project decisions | they affect future coding and review work |
| commands and workflows | agents need the same operational facts repeatedly |
| deployment constraints | mistakes are expensive and context is easy to lose |
| agent operating rules | they should apply consistently at startup |
Poor candidates:
| Do not store | Why |
|---|---|
| temporary scratch notes | they create recall noise |
| guesses not yet verified | they become false durable context |
| huge pasted logs | summarize the conclusion instead |
| duplicated local agent rules | store shared rules once in Lore |
Lore is most valuable when several agents share the same memory graph. A rule learned in Claude Code can later help Codex, OpenClaw, Pi, or Hermes if it is stored in the right path.