PatrickStarVibe/VibeCodingAssistant-Connect-Your-Codex-And-Claude-Code

What it solves

It addresses the risk and lack of traceability when using a single AI to handle complex web coding tasks. Instead of relying on one model to plan and execute, it creates a collaborative AI team where different roles review and verify each other's work, ensuring that the user's original prompt remains the primary source of truth.

How it works

The system orchestrates a multi-agent workflow consisting of specialized roles:

• Architect: Breaks down requirements into detailed execution units.
• Plan Reviewer: Identifies risks and gaps in the plan.
• Developer: Implements the code only after the plan is approved.
• Final Reviewer: Independently verifies code changes and test results.
• VibeCodingAssistant: Manages user communication, task routing, and state tracking.

It utilizes a tiered model strategy (e.g., DeepSeek for orchestration, Claude for planning/review, and Codex for implementation) and integrates with Lark as a remote interface, allowing users to manage local coding tasks from a mobile device.

Who it’s for

Web developers who want a more rigorous, reviewable, and remote-manageable AI coding workflow for tasks ranging from simple style tweaks to core module refactoring.

Highlights

• Multi-Agent Collaboration: Prevents "self-approval" by using different models for planning, reviewing, and executing.
• Tiered Workflow Modes: Offers four levels of rigor (Low, Medium, High, Extra High) based on task complexity and risk.
• Remote Management: Integrated with Lark for mobile approvals, progress tracking, and task control.
• Full Traceability: Generates queryable artifacts for every stage, including implementation logs and final reports.
• Flexible Model Routing: Optimizes cost and quality by assigning specific models (like DeepSeek v4 Flash) to high-frequency orchestration tasks.