core-memory

name	core-memory
summary	Use Core Memory canonical write/read/flush surfaces from OpenClaw integrations without bypassing runtime authority.
description	Use when working with Core Memory from OpenClaw, including finalized-turn writes, memory retrieval, bridge behavior, semantic bead fields, promotion, claims, associations, or integration debugging.

Core Memory OpenClaw Integration

Before changing Core Memory/OpenClaw integration behavior, read the authoritative companion instructions:

../../../../docs/integrations/openclaw/core-memory-skill-instructions.md

Follow those instructions unless they conflict with higher-priority system/developer instructions or canonical Core Memory docs.

Key reminders:

• OpenClaw is not the memory authority; Core Memory is.
• Route finalized-turn writes through canonical ingestion (process_turn_finalized(...) / bridge event emission).
• Route retrieval through canonical read surfaces.
• Do not mutate .beads, .turns, archive, flush, or rolling-window files directly.
• Keep the OpenClaw bridge thin; do not move memory-engine semantics into plugin JS.
• Questions are retrieval/context turns, not declarative memories to promote.
• Treat because as grounded free-text support for applied semantic labels/state; short user-text quotes are valid support when grounded, but do not use guessed filler or long whole-turn dumps.
• Treat entity extraction/canonicalization as a Core Memory live-path judge responsibility, not bridge-side regex/NER; pass candidates only and let the runtime decide canonical labels/aliases.
• Use canonical association relationship types, not helper labels as durable relations.