Skip to content

Latest commit

 

History

History
43 lines (34 loc) · 2.69 KB

copilot_read.md

File metadata and controls

43 lines (34 loc) · 2.69 KB

Project Context: Multi-Agent Software Development System

1. Project Goal

[cite_start]We are building a Multi-Agent Collaborative System capable of autonomous software development[cite: 7]. The system will receive a natural language task (e.g., "Build a website") and autonomously plan, code, and test the software.

2. Core Architecture

The system requires a Multi-Agent Orchestrator to manage the following specialized agents:

  1. [cite_start]Project Planning Agent (Planner)[cite: 27]:

    • Role: Architect & Project Manager.
    • Responsibility: Analyzes user requirements, designs software architecture, and breaks down the project into a list of specific files and development tasks.
    • Output: A structured plan (JSON) containing file paths, dependencies, and step-by-step instructions.

  2. [cite_start]Code Generation Agent (Coder)[cite: 28]:

    • Role: Senior Developer.
    • Responsibility: Executes the tasks defined by the Planner. It receives a specific filename and task description, then outputs the complete code.
    • [cite_start]Tools: Must be able to write files to the local file system[cite: 36].

  3. [cite_start]Code Evaluation Agent (Reviewer) [cite: 29] (Optional for MVP):

    • Role: QA Engineer.
    • Responsibility: Reviews code quality and functional correctness.

3. Workflow (Orchestration Logic)

[cite_start]The main.py entry point should function as the Orchestrator [cite: 30-33]:

  1. Receive Task: User inputs a request (e.g., "Create an arXiv CS Daily webpage").
  2. Plan: The Planner Agent generates a file structure and task list.
  3. Execute: The Orchestrator iterates through the task list. For each file, it invokes the Coder Agent.
  4. Save: The system writes the generated code to a dedicated ./workspace directory.

4. Technical Constraints & Tools

  • Language: Python 3.10+
  • [cite_start]LLM Integration: Use a unified LLMClient to interface with APIs (e.g., DeepSeek, OpenAI)[cite: 16].
  • Prompt Engineering: Use distinct System Prompts for each agent to enforce their roles.
  • State Management: The system must maintain a log of what has been built to avoid context loss.

5. First Test Case: "arXiv CS Daily" Webpage

The system must be capable of generating a web app with:

  • [cite_start]Category navigation (cs.AI, cs.CV, etc.)[cite: 43].
  • [cite_start]Daily updated paper lists[cite: 46].
  • [cite_start]Paper detail pages with PDF links and citation tools[cite: 49].

YOUR ROLE

You are my Lead Engineer. I will ask you to implement specific modules based on this architecture. Always keep the Multi-Agent structure in mind. Start by helping me scaffold the project directory.