item-scraper/prd.md

PRD: AI-Powered Hardware Sourcing Assistant (Chrome Extension)

1. Role & Goal

You are an expert Chrome Extension Developer and AI Integration Specialist. The goal is to build a browser extension that allows a user to extract structured electronic product data (Laptops, Mouse, Keyboards, etc.) from any e-commerce website (especially non-standard local sites in Central America) into a Google Sheet using the Gemini API for semantic analysis.

2. Target Workflow

1. User browses a product page (e.g., a laptop on a Honduran retail site).
2. User opens the Chrome Side Panel.
3. User clicks a category button (e.g., "Extract Laptop") or types a custom command.
4. The extension captures the page text -> Sends it to Gemini API.
5. Gemini parses the unstructured Spanish/English text into a clean JSON.
6. The JSON is sent to a Google Apps Script Web App URL which appends it as a new row in a Google Sheet.

3. Technical Stack

• Frontend: Manifest V3, HTML5, Tailwind CSS (via CDN), JavaScript.
• APIs:
  • chrome.sidePanel (for the persistent UI).
  • chrome.scripting (to read page content).
  • Gemini API (Google AI Studio): For data extraction and translation.
• Backend: Google Apps Script (as a lightweight proxy to Google Sheets).

4. Functional Requirements

4.1 UI Components (Side Panel)

• API Key Input: A secure way to save the Gemini API Key to chrome.storage.local.
• Action Buttons: Quick-action buttons for different categories:
  • [Extract Laptop]: Target fields: Brand, CPU, RAM, Storage, Price, URL.
  • [Extract Peripheral]: Target fields: Brand, Model, Connectivity, Battery Life, Price, URL.
• Chat Interface: A simple chat box to give custom instructions to the AI about the current page.
• Status Indicator: Loading states, Success/Error messages.

4.2 Data Extraction Logic (The Content Script)

• The script must extract document.body.innerText but filtered to remove excessive whitespace or script tags to save token costs.
• It should also capture the current window.location.href.

4.3 AI Prompt Engineering (The "Brain")

The extension must send a system prompt to Gemini:

"Context: You are a professional shopping assistant. Input: Unstructured text from a retail website (might be in Spanish). Task: Extract [Fields] and translate technical terms to English/Chinese as requested. Constraint: Output ONLY valid JSON. Convert local currency (e.g., Lempira L.) to USD if possible, or keep original with symbol."

4.4 Data Storage (Google Sheets Integration)

• Instead of complex OAuth2, the extension will send a POST request to a Google Apps Script Web App URL.
• The Apps Script will handle appendRow() logic.

5. File Structure

• manifest.json: Configuration for permissions (sidePanel, scripting, storage, activeTab).
• sidepanel.html: The UI layout.
• sidepanel.js: Handling UI events, calling Gemini API, and communicating with Apps Script.
• content.js: Script to read the webpage text.
• google-script.js: The backend code to be pasted into Google Apps Script.

6. Security & Constraints

• Location Context: The user is in the US; use US-based Gemini API endpoints.
• Privacy: Do not send sensitive user data; only product-related text.
• Error Handling: Handle cases where Gemini cannot find product info (return "N/A").

7. Future Scalability

• Ability to capture product images (extracting <img> tags).
• Automatic currency conversion using a secondary FX API.