Why does this skill use a sub-agent?

Using the magi-system-orchestrator sub-agent prevents the main conversation from being flooded with raw CLI logs, keeping the main context clean and token-efficient.

What is the MAGI System in Claude Code?

It is a deliberation framework that uses three distinct AI perspectives (Claude, Codex, and Gemini) to reach a consensus on complex technical decisions or architectural designs.

Does it require specific CLI tools?

Yes, it requires locally configured Codex/GPT-5 and Gemini CLI tools accessible via your terminal with appropriate TTY settings.

How does it handle token limits?

It employs a specific pattern of redirecting verbose CLI output to temporary files and only returning the final agent summary to the main LLM context.

When should I use MAGI instead of standard Claude?

Use MAGI for large-scale architectural choices, investment decisions, or complex reviews where multiple viewpoints are required to identify trade-offs and risks.

MAGI System Orchestrator

Name: MAGI System Orchestrator
Author: kazuph

bykazuph

•

Productivity & Workflow

Orchestrates multi-AI deliberation between Claude, Codex, and Gemini to provide diverse perspectives on complex architectural and decision-making tasks.

The MAGI System Orchestrator is a sophisticated multi-agent skill designed for high-stakes decision-making, technical architectural reviews, and complex problem-solving. By coordinating parallel execution between Claude, Codex, and Gemini, it mitigates individual AI bias and provides a unified, structured action plan. The skill utilizes a dedicated sub-agent to manage heavy CLI outputs, ensuring token efficiency while maintaining a rigorous 'tri-calculation' process that includes availability checks, persona-driven insights, and synthesized synthesis of results.

Key Features

0115 GitHub stars

02Strict availability checks and TTY-aware execution for reliable cross-model coordination

03Automated sub-agent orchestration to prevent token explosion in the main context

04Token-efficient CLI patterns that capture only final agent messages

05Multi-perspective deliberation using Claude, Codex/GPT-5, and Gemini simultaneously

06Standardized output templates for situation snapshots and unified action plans

Use Cases

01Performing deep-dive implementation planning for fundamental system changes

02Evaluating complex software architectures where a single AI perspective might be biased

03Conducting high-level technical strategy comparisons involving multiple trade-offs

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add kazuph/dotfiles magi

For use in Claude.ai and ChatGPT

Download Skill