Summary

gbrain sync --source <id> updates source bookkeeping for the requested source, but the import/write path can still write pages/chunks/tags under the default source.

Repro / observed behavior

From an OpenClaw Support KB install:

gbrain sync --repo ~/.gbrain/sources/openclaw-support-kb --source openclaw-support-kb

gbrain sources list --json

Observed before patch:

• default had ~636 pages
• openclaw-support-kb existed but had 0 pages
• search worked only because data had accidentally landed in default

Root cause

sync --source carried sourceId into sync state/bookkeeping, but performFullSync() called runImport() without that source id. Then runImport() / importFile() / engine CRUD defaulted page writes and slug lookups to the default source.

The same source-blind slug path affected page lookup, tag reconciliation, chunk upserts/deletes, and version creation during import.

Expected behavior

Named-source sync/import should write and reconcile content under the requested sources.id, while calls that omit sourceId preserve the current default-source behavior.

Patch direction

Thread optional sourceId through:

• sync.ts → runImport() / importFile()
• import.ts → importFile()
• import-file.ts → getPage, putPage, tags, chunks, versions
• PGLite/Postgres engine page/chunk/tag/version CRUD

Default behavior should remain source_id='default' when no source id is provided.

Validation target

• bun run typecheck
• source-scoped sync of a multi-file repo lands pages under the named source
• gbrain search ... --source <id> returns expected hits