gbrain-sync

name: gbrain-sync description: >- N-repo selective sync of gbrain structure artifacts (agency org chart, client pod skeletons) across ThakiCloud repos via the research hub, while gbrain DB content (entity pages) stays accessible repo-independently through the gbrain CLI. Use when the user runs /gbrain-sync, asks to "sync gbrain artifacts", "share agency org chart across repos", "copy client pod structure", "gbrain 구조 동기화", "에이전시 조직도 다른 리포에 공유". Do NOT use for .claude/ assets (use claude-sync), .cursor/ assets (use cursor-sync), or gbrain DB↔KB entity sync (use gbrain-bridge).

gbrain-sync

Propagates lightweight gbrain structure artifacts — the agency org chart and client pod skeletons — across all managed ThakiCloud repos, using research as the merge hub (same pattern as claude-sync).

Why this skill exists (the data model)

gbrain data lives in two places. They sync differently:

The design rule: copy structure, reference bulk, query DB. Never rsync the 126M knowledge base into 5 repos (= 630M duplication + git bloat). Query it via the CLI.

Sync Scope (whitelist — structure only)

Target Repositories

Hub: research. Source of truth for these artifacts is ai-platform-strategy (where the agency-org chart and client pods are authored). Sync is bidirectional through the hub but in practice flows ai-platform-strategy → research → others.

macro-factor-dashboards is listed in eod-ship managed-repos but is NOT present locally — skip any target whose dir is absent (do not fail).

{BASE} resolves at runtime: $HOME/work/thakicloud if it exists, else $HOME/thaki.

Workflow

Step 1: Environment Detection

if [ -d "$HOME/work/thakicloud" ]; then BASE="$HOME/work/thakicloud"; else BASE="$HOME/thaki"; fi
EXCLUDES="--exclude brain/ --exclude memory/ --exclude node_modules/ --exclude __pycache__/ --exclude .DS_Store"
TARGETS="ai-platform-strategy realtime-translator github-to-notion-sync ai-template ai-model-event-stock-analytics"

Step 2: Pull Phase (Targets → research, newest-wins)

for T in $TARGETS; do
  [ -d "$BASE/$T" ] || continue
  # agency-org KB
  if [ -d "$BASE/$T/knowledge-bases/agency-org" ]; then
    mkdir -p "$BASE/research/knowledge-bases/agency-org"
    rsync -au "$BASE/$T/knowledge-bases/agency-org/" "$BASE/research/knowledge-bases/agency-org/"
  fi
  # client pod structure (brain/ + memory/ excluded)
  if [ -d "$BASE/$T/clients" ]; then
    mkdir -p "$BASE/research/clients"
    rsync -au $EXCLUDES "$BASE/$T/clients/" "$BASE/research/clients/"
  fi
done

Step 3: Push Phase (research → Targets, checksum)

for T in $TARGETS; do
  [ -d "$BASE/$T" ] || continue
  if [ -d "$BASE/research/knowledge-bases/agency-org" ]; then
    mkdir -p "$BASE/$T/knowledge-bases/agency-org"
    rsync -ac "$BASE/research/knowledge-bases/agency-org/" "$BASE/$T/knowledge-bases/agency-org/"
  fi
  if [ -d "$BASE/research/clients" ]; then
    mkdir -p "$BASE/$T/clients"
    rsync -ac $EXCLUDES "$BASE/research/clients/" "$BASE/$T/clients/"
  fi
done

Step 4: Verification & Report

echo "=== agency-org ==="
for R in research $TARGETS; do
  [ -d "$BASE/$R" ] || continue
  C=$(find "$BASE/$R/knowledge-bases/agency-org" -type f 2>/dev/null | wc -l | tr -d ' ')
  echo "  $R: $C files"
done
echo "=== clients (structure, no brain/memory) ==="
for R in research $TARGETS; do
  [ -d "$BASE/$R" ] || continue
  C=$(find "$BASE/$R/clients" -type f ! -path '*/brain/*' ! -path '*/memory/*' 2>/dev/null | wc -l | tr -d ' ')
  echo "  $R: $C files"
done

Report format:

gbrain-sync Complete
- agency-org/: 4 files (N repos aligned)
- clients/ (structure): M files (N repos aligned)
- brain/ + memory/ excluded: ✓ (no live data propagated)
- bulk KB (126M): not copied — access via `gbrain query` ✓

Implementation Rules

• Whitelist only — never rsync the whole knowledge-bases/ tree
• brain/ + memory/ ALWAYS excluded (live per-client data — keep repo-local)
• Pull uses -au (newest-wins by mtime); Push uses -ac (checksum, idempotent)
• Forbidden flags: --delete (never remove target-only files), -i/-v (no output on openrsync)
• Skip any target dir that does not exist (do not fail the run)
• When invoked by eod-ship/sod-ship, run without confirmation

Relationship with other sync skills

No overlap: claude-sync owns .claude/, gbrain-sync owns repo content artifacts. Running both in eod-ship is safe and idempotent.

DB access from other repos (no sync required)

The 127 gbrain entity pages need no copying. From any repo/dir:

gbrain query "what do we know about <entity>"   # hybrid search
gbrain search "<keyword>"                          # keyword
gbrain get <slug>                                  # full page
gbrain doctor --json                               # connectivity check

This is the canonical way to use gbrain knowledge in another repo. gbrain-sync only handles the file structure that the CLI cannot serve.