Skip to content

CLAUDE.md

Latest commit

 

History

History
100 lines (70 loc) · 3.59 KB

CLAUDE.md

File metadata and controls

100 lines (70 loc) · 3.59 KB

CLAUDE.md

This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.

Project Overview

CookingAgent is a TypeScript-based intelligent chatbot that serves as a personal cooking assistant. It combines LLM capabilities with the HowToCook open-source recipe repository to provide:

  • Smart menu recommendations based on user preferences
  • Ingredient and tool shopping lists
  • Optimized cooking workflow planning
  • Interactive conversation support

Key Documents

  • Requirements: docs/prd.md - Product Requirements Document v1.1
  • Architecture: docs/sad.md - System Architecture Document v1.3
  • Task List: docs/todo.md - Detailed development tasks breakdown
  • Data Source: HowToCook/ - Recipe repository (Markdown files)

Architecture Overview

The system follows a 3-layer architecture:

  1. Data Layer: Offline preprocessing pipeline that converts HowToCook MD files to structured JSON
  2. Application Core: Conversation engine, knowledge base, and AI agent modules
  3. Presentation Layer: CLI interface built with Ink + React

Key Components

  • Offline Data Processing:

    • RecipeParser: MD → native JSON conversion
    • LLM-basedTagger: Incremental smart tagging using LLM
    • DatabaseBuilder: Generates recipes_index.json and recipes_data.json

  • Online Agent Modules:

    • extractIntent: Parse user preferences from natural language
    • recommendMenu: Generate personalized menu recommendations
    • planWorkflow: Create optimized cooking workflows
    • KnowledgeBase: In-memory recipe data management

  • CLI Interface: React-based Ink components for rich terminal UI

Tech Stack

  • Language: TypeScript
  • AI Framework: Vercel AI SDK
  • UI Framework: Ink (React for CLI)
  • Data Processing: Native Node.js with crypto for hashing
  • Schema Validation: Zod (for LLM output validation)

Development Commands

Since the project is not yet implemented, these commands will be added during development:

# Data processing
npm run process-data        # Process HowToCook recipes into JSON

# Development
npm run dev                 # Start development server
npm run build               # Build for production
npm start                   # Run the CLI application

# Code quality
npm run lint                # Run ESLint
npm run type-check          # Run TypeScript compiler

Data Flow

  1. Offline: HowToCook MD files → RecipeParser → LLM-basedTagger → JSON database
  2. Online: User input → extractIntent → recommendMenu → planWorkflow → CLI display
  3. Incremental: Only new/changed recipes are processed by LLM to optimize API costs

LLM Integration Strategy

The system uses structured prompts with specific personas:

  • LLM-basedTagger: "美食数据分析师" for recipe tagging
  • extractIntent: "智能点餐员" for preference extraction
  • recommendMenu: "高级营养师和米其林餐厅行政总厨" for menu creation
  • planWorkflow: "效率大师和厨房总指挥" for workflow optimization

All LLM interactions use JSON schemas with Zod validation to ensure output consistency.

CLI UI Requirements

The terminal interface should display:

  • Welcome message at the top
  • Message history in the middle
  • Input box at the bottom that clears after submission
  • Status/error line below the input box

Development Notes

  • Use incremental processing to minimize LLM API calls
  • Preserve original markdown text for ingredient lists and cooking steps
  • Implement proper error handling for LLM API failures
  • Follow the detailed task breakdown in docs/todo.md