cursor-sync

name	cursor-sync
description	N-repo bidirectional sync of agent assets (commands, skills, rules, agents) under BOTH .cursor/ and .claude/ across all 7 managed repositories via research hub. (Name is historical — it syncs .claude/ too.)
disable-model-invocation	true

Sync agent assets under .cursor/ AND .claude/ across all managed repositories.

Naming note: this skill was originally named cursor-sync when it only handled .cursor/. It now syncs .claude/ assets too. The name is kept for stability.

Repository Registry

Hub: research (merge canonical)

Targets (under ~/thaki/ or ~/work/thakicloud/):

1. ai-platform-strategy (sylvanus4/ai-strategy)
2. macro-factor-dashboards (sylvanus4/macro-factor-dashboards)
3. github-to-notion-sync (sylvanus4/github-to-notion-sync)
4. ai-template → origin: sylvanus4/cursor-template (renamed remote, local dir kept)
5. ai-model-event-stock-analytics (hyojunguy/ai-model-event-stock-analytics — different account)

realtime-translator is intentionally NOT a sync target (user directive 2026-07-05: "싱크 하지마"). Do not re-add.

Last updated: 2026-05-22.

Sync Process

Run for BOTH trees: .cursor/ and .claude/. Two-phase hub model.

Standard mode: -avuc (archive + newest-wins + checksum).

• -u newest-wins: protects a receiver's genuinely newer file.
• -c checksum: transfers only files whose CONTENT differs, so identical files with differing mtime do NOT churn (avoids 600+ phantom git diffs).
• New files are added; updated files (e.g. a revised SKILL.md) DO propagate — unlike --ignore-existing, which skips every existing file regardless of content.

Always exclude runtime/per-repo data (see Excludes below).

Phase 1: Pull (absorb changes from all targets into hub)

EXC="--exclude=state/ --exclude=outputs/ --exclude=*.log \
     --exclude=__pycache__/ --exclude=*.pyc --exclude=.DS_Store"
for t in <targets>; do
  for tree in .cursor .claude; do
    for sub in commands skills rules agents; do
      [ -d "$t/$tree/$sub" ] && rsync -avuc $EXC "$t/$tree/$sub/" "research/$tree/$sub/"
    done
  done
done

Phase 2: Push (distribute merged hub to all targets)

for t in <targets>; do
  for tree in .cursor .claude; do
    for sub in commands skills rules agents; do
      [ -d "research/$tree/$sub" ] && rsync -avuc $EXC "research/$tree/$sub/" "$t/$tree/$sub/"
    done
  done
done

Scope

Syncs (both .cursor/ and .claude/):

• commands/
• skills/
• rules/
• agents/ (.claude/ only in practice)

Does NOT sync:

• hooks/ — coupled to per-repo settings.json wiring (absolute paths). Propagate manually.
• settings.json / settings.local.json — project-specific config.
• Runtime/state data: state/, outputs/, *.log, caches.
• Project-specific code, configs, .env, credentials.

Rules

• Research repo acts as the canonical merge hub
• -avuc: newest-wins for real conflicts, checksum to skip identical content
• Always apply the Excludes (runtime state must not cross repos)
• Confirm direction before bulk sync operations
• Verify convergence (md5 / recursive file counts) after sync
• Never sync .env, credentials, project-specific configs, or hooks wiring