# Config Sync

> Configuration synchronization protocol for multi-machine Claude setups

- Skill: `subashmurugandev/config-sync` (Agent Skill, multi-file: 2 files)
- Install (CLI): `npx skillmds@latest add subashmurugandev/config-sync`
- Raw SKILL.md: https://api.skillmd.com/api/skills/subashmurugandev/config-sync/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: SubashMuruganDev (https://skillmd.com/u/subashmurugandev)
- Updated: 2026-09-22
- Page: https://skillmd.com/skills/subashmurugandev/config-sync

---


# Config Sync Protocol

**Trigger keywords:** sync config, sync claude, rsync claude

---

## IMPORTANT: When to Use This Skill vs The Script

**For routine syncing, just run the script directly:**
```bash
~/mcp-recover.sh   # Run on primary - handles everything automatically
```

The automated script is MORE RELIABLE than asking Claude because:
- It's deterministic (same behavior every time)
- No risk of Claude forgetting adaptation rules
- Already has hostname checks built in
- Runs daily via cron anyway

**Use this skill ONLY for:**
- Complex one-off changes requiring judgment
- Debugging sync issues
- Initial setup of new machines

**For simple "sync now" requests:** Tell the user to run `~/mcp-recover.sh` on primary.

---

## CRITICAL: Run from PRIMARY only

**This skill MUST only be executed on primary.** Never sync config FROM a satellite machine.

| Machine | Role | Action |
|---------|------|--------|
| primary | SOURCE/MASTER | Pushes config TO satellites |
| host1 | SATELLITE | Receives FROM primary only |
| host2 | SATELLITE | Receives FROM primary only |
| hostname | SATELLITE | Receives FROM primary only |

---

## What Gets Synced

| Item | Treatment |
|------|-----------|
| ~/.claude/CLAUDE.md | Direct copy |
| ~/.claude/skills/ | Direct copy (rsync --delete) |
| ~/.claude/agents/ | Direct copy (rsync --delete) |
| ~/.claude/commands/ | Direct copy (except lm.md - see below) |
| ~/.claude/aw-hook.sh | Direct copy |
| /lm command | **ADAPTED** - path translation for NFS |

**Excludes:** settings*.json, .credentials.json, history.jsonl, projects/, plugins/

---

## Sync Commands

### Standard configs (run all in parallel)
```bash
# Sync CLAUDE.md, skills, agents, commands (except lm.md), aw-hook.sh
for host in host1 host2 hostname; do
    rsync -av --delete --exclude='lm.md' \
        -e 'sshpass -p khis9 ssh -o StrictHostKeyChecking=no' \
        ~/.claude/CLAUDE.md ~/.claude/skills ~/.claude/agents \
        ~/.claude/commands ~/.claude/aw-hook.sh \
        user@${host}:~/.claude/ &
done
wait
```

### /lm command (path translation for NFS mounts)
```bash
# Translates /path/to/source/ → /path/to/source/coding/
for host in host1 host2 hostname; do
    sed 's|/path/to/source/|/path/to/source/coding/|g' ~/.claude/commands/lm.md | \
        sshpass -p 'khis9' ssh -o StrictHostKeyChecking=no user@${host} \
        'cat > ~/.claude/commands/lm.md' &
done
wait
```

---

## Quick Reference

When user says "sync claude config":

**STEP 0 - MANDATORY HOSTNAME CHECK:**
```bash
# ALWAYS run this first before any sync operations
if [[ "$(hostname)" != "primary" ]]; then
    echo "ERROR: Config sync must run from primary, not $(hostname)"
    echo "This machine is a SATELLITE - it receives config, not pushes it."
    exit 1
fi
echo "✓ Confirmed on primary - proceeding with sync"
```

Then:
1. Run standard config sync (CLAUDE.md, skills, agents, commands, aw-hook.sh)
2. Run /lm path translation
3. Report success/failure for each machine

**If hostname check fails:** Stop immediately. Do not attempt to sync from a satellite machine.