Extension Architecture

The project uses a no-build architecture so it can be loaded directly into Chrome as an unpacked extension.

Manifest Strategy

• Manifest Version: MV3
• Background worker initializes defaults
• Content scripts are loaded in order on supported AI domains
• Popup and options pages manage user configuration

Script Order

1. content/aiOverlay.js
2. content/liveCoachBubble.js
3. content/quickBuilder.js
4. content/monitor.js
5. content/interactionTracker.js

This order ensures toast runtime exists first, UI surfaces are injected, then prompt and behavior monitors run.

Main Runtime Components

• content/quickBuilder.js: in-page Prompt Builder launcher + panel, role selector, Build + Insert, Build + Send, keyboard shortcuts (Ctrl/Cmd + O and command fallback Ctrl/Cmd + Shift + O) when focused in chat composer
• content/monitor.js: prompt capture, send detection, realtime scoring, bad-prompt detection, dependency metrics
• content/liveCoachBubble.js: floating realtime HUD for prompt score + habit snapshot, draggable, persisted, and auto-remounted on dynamic UI rerender
• content/interactionTracker.js: copy/paste monitoring, fast-copy detection, large-paste warnings
• content/aiOverlay.js: toast rendering layer (top-right notifications)

Reliability Considerations

• MutationObserver re-attaches listeners when AI UI nodes rerender
• prompt deduping prevents duplicate warnings on fast UI updates
• prompt snapshot is captured before send clears the composer
• live draft scoring runs on debounced input updates without inflating send counters
• all settings are merged with defaults to avoid corrupt local state