Role
Designed the rules, lifecycle, agent roles, launcher, dashboard flow, and the transition from v1 experiments into a structured v2 system.
AI Workflow / Agent Systems
Multi-Agent Orchestration Framework
A reusable framework that turns role-based AI collaboration, review rules, and operating flow into an explicit system instead of repeating prompt structure from scratch on every project.
v1 Bash + Markdown
v2 TypeScript
Claude Agent SDK
Launcher + Dashboard
Problem
AI collaboration kept requiring the same role split, review steps, and failure rules to be rebuilt on every project.
Outcome
A reusable framework with 4 agents, 66 tests, and 119 feature definitions tied to a launcher and dashboard structure.
Why It Matters
This is evidence that I can move AI work from experimentation into repeatable operating structure.
From v1 to v2
Version 1 proved the operating idea with Bash and Markdown. Version 2 turned that experience into a TypeScript component model with event-driven structure and clearer ownership boundaries.
The important point is not the agent count alone. It is that the workflow now carries rules for review, rejection, escalation, and launch flow instead of leaving those decisions implicit.
System direction: thin-shell / service-core separation so orchestration logic can be reused without rewriting every adapter layer.
Real value: the framework came from real project pressure, especially work that repeated across DXCenter, robotics, and AI-assisted implementation contexts.
Roles: AI workflow, agent systems, developer tooling, and internal platform.
Korean version: open the original Korean case study.