Muster – <.md> 파일을 기반으로 구축된 Claude Code용 다중 에이전트 제품 팀

A multi-agent product team that runs inside Claude Code. Muster coordinates specialized AI agents — PM, Developer, UI/UX, Content, Marketing, Research, Legal, QA, and growing — through persistent markdown files. No external frameworks, no API dependencies. Just Claude Code and the filesystem. You (Founder) --> Root Claude (PM) --> Dev | UI/UX | Content | Marketing | Legal | QA | Research The PM reads everything. Each specialist reads only what's relevant to their current task. That's the core idea. Most multi-agent frameworks optimize for agent communication — how agents pass messages to each other. But the real bottleneck is context. Claude Code agents forget everything between sessions. If you're building a product with design, dev, legal, marketing, and QA, you need persistent memory and a way to keep each agent focused on what matters. Muster solves this with three mechanisms: - Three-Tier Reading Model — Agents read ~80 lines at startup (their role + filtered context + current task). Full docs are on-demand only. Irrelevant files are never loaded. - PM-as-Context-Translator — When a decision is made, the PM doesn't tell every agent "go read the decision log." The PM updates each agent's context file with only what that agent needs. - Files as Persistent Memory — Agent brains, orchestration queues, handoff logs, and decision records persist as markdown files. No session starts from zero. curl -fsSL https://raw.githubusercontent.com/sandhuka/muster-ai/main/scripts/setup-project.sh | bash -s my-project cd ~/Desktop/my-project claude Then tell Root Claude your product idea. It acts as your PM — plans sprints, coordinates agents, and tells you who to invoke next. See getting-started.md for the full step-by-step walkthrough. - You describe your idea to Root Claude (the PM) - PM sends Research to investigate the market - If viable, PM writes the product spec, plans a sprint, and queues up agent tasks - You invoke agents following the PM's sequence — one at a time or in parallel across separate terminals when tasks are independent - Each agent reads its filtered context, does the work, files a handoff, and promotes the next step - Repeat until shipped | Agent | Role | |---|---| | PM (Root Claude) | Plans sprints, cascades context, reviews deliverables, makes decisions | | Research | Market analysis, competitive teardowns, user insights, product validation | | Developer | Production code, architecture, testing — iOS, backend, and generic skills | | UI/UX | Wireframes, user flows, component specs, design tokens | | Content | In-app copy, blog, email, store listings, help docs | | Marketing | Growth strategy, campaigns, user acquisition, analytics | | Legal | Compliance, privacy, terms of service, IP protection (guidance, not legal advice) | | QA | Test strategy, bug tracking, release validation | | CrewAI / AutoGen / LangGraph | Muster | | |---|---|---| | Solves | Agent communication | Agent memory and context efficiency | | Built on | Python libraries, API layers | Markdown files and Claude Code | | Requires | Code to define agents | No code — file-based configuration | | Built from | Framework design theory | Real product development (an iOS app) | | Includes | Agent orchestration primitives | Full operational patterns — growth caps, cascade lag prevention, decision autonomy, pre-handoff self-review | Muster uses a two-repo model: my-project/ ├── .claude/agents/ # Agent startup configs (invoke with @developer, @research, etc.) ├── CLAUDE.md # Your product info + overrides ├── muster/ # <-- Git submodule (this repo) ├── knowledge-base/ │ ├── agent-context/ # Per-agent filtered product context (PM writes, agents read) │ ├── product-spec.md # Product specification │ ├── orchestration-queue.md # Who to invoke next │ ├── agent-requests.md # Inter-agent communication (handoffs, reviews) │ └── ... └── src/ # Your code Muster (this repo) contains agent roles, skills, and protocols — shared across all projects via git submodule. Your project contains product context, knowledge base, and source code — one per product. | Doc | What It Covers | Read When | |---|---|---| | getting-started.md | Step-by-step setup and first sprint | Setting up a new project | | architecture-and-design.md | Architecture deep dive — data flow, context management, how agents communicate | Evaluating whether to adopt Muster | | system-guide.md | Templates, extensibility, verification checklist | Adding agents, skills, or modifying the framework | MIT License. See LICENSE.