How Vibe Guard fits together.

Counts tokens and catches loops, prompt-injection, and overspend on the web AI tools — ChatGPT, Claude, Gemini, bolt, v0, Replit. Tags everything source=browser.

1. Install from the Chrome Web Store when approved — or load the unpacked vibeguard-extension/ folder via chrome://extensions → Developer mode → Load unpacked.
2. Open the popup and sign in with your account.
   The popup shows your live count and which site it's watching.
3. Keep coding — it pushes to your row automatically.

A menu-bar app — no terminal needed. Reads your local coding-agent logs (Claude Code, Codex, Claude-in-Xcode, Codex-in-Xcode, Gemini CLI, Cline, Aider), plus your plan limits and live context window. It floats the Context HUD over Xcode, and the Qwen Coach runs on-device via Apple Core AI (with Private Cloud Compute for heavy lifts).

1. Download vibeguard-desktop-meter.zip from the downloads page and unzip.
2. Double-click to run. A bolt appears in your menu bar with your token count.
3. Pair it: the menu shows your code — type it into the app.
4. Float the Context HUD over Xcode — see the ~200K wall coming:

   python3 vibeguard-context-hud.py

   The Qwen Coach float runs on-device through Apple Core AI (escalating to PCC for heavy lifts), so it ships inside the native companion app rather than a script.

A statusline that shows tokens, cost, your Claude rate-limit %, and your context fill right in the terminal — and a Stop hook that mirrors your daily total to your phone.

1. Install the plugin:

   /plugin marketplace add BTSInnovationsLLC/vibe-guard-plugin
   /plugin install vibe-guard

2. Turn on the statusline — add this once to ~/.claude/settings.json:

   {
     "statusLine": {
       "type": "command",
       "command": "python3 ~/.claude/plugins/marketplaces/vibe-guard/statusline.py"
     }
   }

3. Sign in once with your Vibe Guard account (open the app or run the meter) — the plugin then reports to your account automatically.

A Stop hook that sums today's Codex usage from ~/.codex/sessions and pushes your running total and Plus/Pro rate-limits to your phone.

1. Copy the hook:

   mkdir -p ~/.codex/vibeguard
   cp hooks/push.py ~/.codex/vibeguard/push.py

2. Add the Stop entry — copy hooks/hooks.json to ~/.codex/hooks.json (or merge it).
3. Enable hooks in ~/.codex/config.toml:

   [features]
   codex_hooks = true

4. Sign in with your Vibe Guard account on this computer — it links automatically.

No desktop needed. The app queries each provider's usage API directly — Anthropic, OpenAI, and xAI — for org/API usage, with the account email for Max-style plans.

1. In the app, open Token Count and pick a platform's page.
2. Tap Sign in and paste that provider's API key (and account email where the plan needs it).
3. That platform gets its own ring and stats page. Keys are stored in the device keychain.

A Model Context Protocol server so any MCP client — Claude, Cursor, ChatGPT — can ask Vibe Guard "how much context and limit do I have left?" as a tool, mid-task, and back off before it hits the wall. Not shipped yet — this is on the roadmap, and we're building it honestly rather than faking it.