Sync OS — Sync Live Claude Code Config to the Remote Repo

# Sync OS — Sync Live Claude Code Config to the Remote Repo

Sync the live `~/.claude/` configuration (skills, plugins, hooks, channels, settings) to the claude-code-os GitHub repo. Can be run from any project folder — it always targets the same repo.

## Arguments

`<optional commit message>` — If provided, used as the commit message. If not, auto-generates one from the changes detected.

Examples:
- `/sync-os` — auto-detects changes and generates commit message
- `/sync-os added weather and latex skills` — uses provided message
- `/sync-os` from any project folder — always syncs to the claude-code-os repo

## Constants

- **Repo path:** `/home/vivekkarmarkar/claude-code-os`
- **Remote:** `origin` (GitHub: VivekKarmarkar/claude-code-os)
- **Branch:** `main`

## Workflow

### Step 1: Diff Skills

```bash
diff <(ls ~/.claude/skills/ | sort) <(ls /home/vivekkarmarkar/claude-code-os/skills/ | sort)
```

For each skill in live but not in repo: copy the entire skill folder.
For each skill in repo but not in live: leave it (don't delete — it might be intentionally archived).

### Step 2: Diff Plugins

```bash
diff <(ls ~/.claude/plugins/cache/claude-plugins-official/ | sort) <(ls /home/vivekkarmarkar/claude-code-os/plugins/ | sort)
```

For new plugins: copy the entire plugin folder.
For existing plugins with version changes: remove old version folder, copy new one.
Check for nested `.git` directories and remove them before staging.

### Step 3: Diff Settings

```bash
diff ~/.claude/settings.json /home/vivekkarmarkar/claude-code-os/settings.json
```

If different: copy live settings.json to repo.

### Step 4: Diff Global CLAUDE.md

```bash
diff ~/.claude/CLAUDE.md /home/vivekkarmarkar/claude-code-os/CLAUDE.md
```

If different: copy live CLAUDE.md to repo. This is the global behavioral enforcement file — voice transcription, food image detection, and any future cross-project behavioral rules live here.

### Step 5: Diff Installed Plugins Manifest

```bash
diff ~/.claude/plugins/installed_plugins.json /home/vivekkarmarkar/claude-code-os/installed_plugins.json
```

If different: copy live manifest to repo.

### Step 6: Diff Hooks

```bash
diff <(ls ~/.claude/hooks/ | sort) <(ls /home/vivekkarmarkar/claude-code-os/hooks/ | sort)
```

Copy any new hook files.

### Step 7: Diff Channels

```bash
diff <(ls ~/.claude/channels/ 2>/dev/null | sort) <(ls /home/vivekkarmarkar/claude-code-os/channels/ 2>/dev/null | sort)
```

Copy any new channel folders.

### Step 7.5: Diff MCP Servers

```bash
diff <(ls ~/.claude/mcp_servers/ 2>/dev/null | sort) <(ls /home/vivekkarmarkar/claude-code-os/mcp-servers/ 2>/dev/null | sort)
```

For each MCP server in live (`~/.claude/mcp_servers/<name>/`) but not in repo (`~/claude-code-os/mcp-servers/<name>/`): copy ONLY the source files (`*.py`, `pyproject.toml`, `requirements.txt`, `README.md`). NEVER copy `.venv/`, `__pycache__/`, `*.pyc`, or any file with secrets. The repo's `.gitignore` already excludes `.venv/` and `__pycache__/` but be defensive.

Also re-run the MCP registration sync to refresh `mcp-servers.json` with the live registrations from `~/.claude.json` (with secrets redacted). The `~/.claude/hooks/sync-mcp.sh` hook does this; running it explicitly here ensures the manifest is current. It preserves the top-level structure of `mcp-servers.json` (`local`, `claude.ai`, `plugins` keys) and only refreshes the `local` block:

```bash
~/.claude/hooks/sync-mcp.sh
```

### Step 8: Update README Counts

Count the current totals:
```bash
ls /home/vivekkarmarkar/claude-code-os/skills/ | wc -l
ls /home/vivekkarmarkar/claude-code-os/plugins/ | wc -l
ls /home/vivekkarmarkar/claude-code-os/hooks/*.sh | wc -l
ls /home/vivekkarmarkar/claude-code-os/mcp-servers/ | wc -l
# MCP Servers count in README also includes claude.ai remotes + plugin servers
# from mcp-servers.json — count those separately from the directory list.
```

Update `## Skills (N)`, `## Plugins (N)`, `## Hooks (N)`, and `## MCP Servers (N)` in README.md with actual counts.

For each new skill, add a row to the skills table in alphabetical order:
```
| `skill-name` | One-line description from the first line of SKILL.md |
```

For each new plugin, add a row to the plugins table.
For each new hook, add a row to the hooks table.
For each new local MCP server, add a row to the MCP Servers table under the local-stdio section, alphabetical order.

### Step 9: Check for Large Files

```bash
find /home/vivekkarmarkar/claude-code-os -size +90M -not -path '*/.git/*'
```

If any files exceed 90MB (GitHub's limit is 100MB), add them to `.gitignore` and warn the user.

### Step 10: Stage, Commit, Push

```bash
cd /home/vivekkarmarkar/claude-code-os
git add -A
```

Before committing, check for `.env` files, credentials, tokens — do NOT stage those.

Generate or use the provided commit message. Format:

```
feat: sync — X new skills, Y plugin updates, Z config changes (TOTAL skills)

Details of what changed.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
```

Commit and push to origin main.

### Step 11: Report

Print a summary:
```
Sync complete.
  Skills: X → Y (+N new)
  Plugins: X → Y (+N new, M updated)
  Hooks: X → Y
  CLAUDE.md: [changed/unchanged]
  Settings: [changed/unchanged]
  Commit: <hash>
  Pushed to: origin/main
```

## Rules

1. **Always works from any directory.** The skill hardcodes the repo path. It does not depend on the current working directory.
2. **Never delete from the repo.** Only add or update. If a skill was removed from live, leave it in the repo — it's an archive.
3. **Never commit secrets.** Check for .env, API keys, tokens before staging.
4. **Handle large files.** GitHub rejects files over 100MB. Auto-gitignore anything over 90MB.
5. **Remove nested .git directories.** Plugins sometimes have these. Remove before staging or git will treat them as submodules.
6. **Don't touch the README structure.** Only update counts and add new rows. Never rearrange or remove existing rows.