How does the MAGI voting system work?

Each of the three engines provides a score, a list of pros/cons, and a vote (APPROVE, REJECT, or ABSTAIN). The final report summarizes these to help developers reach an informed consensus.

Which AI models are used in the review process?

The system utilizes MELCHIOR-1 (Claude Opus) for technical depth, BALTHASAR-2 (Gemini) for modularity, and CASPER-3 (Codex/GPT) for security analysis.

Do I need extra plugins for this to work?

Yes, this skill requires the Gemini and Codex-CLI MCP tools to be searchable and available within your Claude Code environment.

Can this skill help with security audits?

Absolutely. The CASPER-3 engine specifically focuses on credential leaks, trust boundaries, and permission management to identify critical security vulnerabilities.

What is the MAGI Architecture skill?

It is a Claude Code skill that uses three different AI models simultaneously to review code and architectural designs from technical, modular, and security perspectives.

MAGI Architecture Review

Name: MAGI Architecture Review
Author: weiting-tw

byweiting-tw

•

Security & Testing

Conducts parallel multi-engine architectural reviews using Claude, Gemini, and Codex to reach a consensus on system design and security.

The MAGI Architecture skill implements a triple-redundancy review system inspired by the MAGI supercomputer from Evangelion. It leverages three distinct AI engines—Claude Opus (MELCHIOR-1) for technical depth, Gemini (BALTHASAR-2) for modularity, and Codex (CASPER-3) for security—to analyze architectural proposals in parallel. By aggregating scores and votes from these diverse perspectives, it provides a robust, low-bias evaluation of complex designs, ensuring that scalability, maintainability, and security risks are thoroughly identified through a structured consensus mechanism.

Key Features

01Automated multi-tool execution for rapid design auditing

02Dedicated security and trust boundary identification

03Triple-engine parallel analysis (Claude, Gemini, and Codex)

04Consensus-based voting system (Approve/Reject/Abstain)

052 GitHub stars

06Detailed scoring (0-10) for architecture, modularity, and safety

Use Cases

01Evaluating new system architectures for scalability and technical depth

02Identifying security risks and permission boundaries in new feature designs

03Reviewing modularity and interface clarity in complex refactors

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add weiting-tw/magi-system magi-arch

For use in Claude.ai and ChatGPT