Muster: Run a Multi-Agent Product Team Inside Claude Code with Persistent

What Muster Does

Muster is a multi-agent product team that runs entirely inside Claude Code. Instead of relying on external frameworks or API dependencies, it coordinates specialized AI agents—Product Manager, Developer, UI/UX Designer, Content Writer, Marketing, Research, Legal, and QA—through persistent markdown files. The system solves Claude Code's fundamental limitation: agents forget everything between sessions.

How It Works: Three Core Mechanisms

1. Three-Tier Reading Model

When you invoke an agent (like @developer or @research), it reads only ~80 lines at startup:

• Their specific role definition
• Filtered context relevant to their current task
• The immediate task instructions

Full documentation remains on-demand. Irrelevant files are never loaded into context, preserving tokens and focus.

2. PM-as-Context-Translator

The Product Manager agent doesn't just broadcast decisions. When a decision is made, the PM updates each agent's context file with only what that agent needs to know. The developer gets technical specs, the marketer gets positioning, the legal agent gets compliance requirements—all in their respective context files.

3. Files as Persistent Memory

All agent "brains," orchestration queues, handoff logs, and decision records persist as markdown files in your project structure:

my-project/
├── .claude/agents/       # Agent startup configs
├── CLAUDE.md            # Your product info + overrides
├── knowledge-base/
│   ├── agent-context/   # Per-agent filtered context
│   ├── product-spec.md
│   ├── orchestration-queue.md
│   └── agent-requests.md
└── src/                 # Your code

Setup and Workflow

Install with one command:

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 (the PM) your product idea. The workflow:

1. You describe your idea to Root Claude
2. PM sends Research to investigate market viability
3. If viable, PM writes product spec, plans sprints, queues agent tasks
4. You invoke agents following PM's sequence (one at a time or in parallel across terminals)
5. Each agent reads filtered context, does work, files handoff, promotes next step
6. Repeat until shipped

Two-Repo Architecture

Muster uses a clean separation:

• Muster repository (git submodule): Contains agent roles, skills, and protocols—shared across all projects
• Your project repository: Contains product context, knowledge base, and source code—one per product

This means you can update Muster's agent capabilities across all projects with a single git submodule update.

When To Use Muster

Use Muster when:

• Building complex products requiring multiple specialties
• You need persistent memory across Claude Code sessions
• You want to avoid repeating context to different agents
• Working on projects spanning days or weeks

Don't use Muster for:

• Quick one-off scripts or fixes
• Simple tasks a single Claude Code session can handle
• Projects where you prefer manual agent coordination

Customizing Agents

Each agent's behavior is defined in markdown files within the Muster submodule. Want to tweak how your Developer agent approaches React components? Edit muster/agents/developer/role.md. Need your PM to follow a different sprint planning methodology? Modify muster/agents/pm/protocols.md.

Because everything's in markdown, customization requires no special tools—just edit the files and your changes persist.