Re-embed OpenClaw memory into a fresh LanceDB store
Plan and verify an OpenClaw memory embedding cutover using a complete store backup, a dry run and matching row counts.
npx skills add agentskillexchange/skills --skill re-embed-openclaw-memory-into-a-fresh-lancedb-store
Invoke when an OpenClaw operator changes embedding model, dimensions or Jina task hints and needs a controlled memory cutover. The agent records the existing database path and configuration, prepares a fresh target and compares migration counts before restarting normal retrieval.
What this skill actually does
Use the documented reembed workflow on a plugin version that exposes it. Stop the gateway to prevent concurrent writes and make a complete filesystem copy of the old LanceDB directory. A default memory-pro export is insufficient as the sole backup because it is limited by row count and scope.
Configure the new embedding settings and a new empty dbPath. Run openclaw memory-pro reembed –source-db –dry-run, record the source count, then run the same command without –dry-run. Keep source and target distinct. Compare the imported count and openclaw memory-pro stats against the dry-run count; retain the old store for rollback before restarting and checking representative recall.
Deliver source/target paths, configuration differences, backup location, counts and recall evidence. The boundary is an OpenClaw plugin embedding migration, rather than generic persistent-memory setup. Do not force same-path re-embedding: upstream documents append-by-id duplication. Source: https://github.com/CortexReach/memory-lancedb-pro/blob/master/README.md .