One command. Full context for your AI agent. Any editor.
npx maestro-bundle ai-agents claudeAI agents (Claude Code, Cursor, Codex) are powerful, but without context they don't know:
- What stack the project uses
- Which coding standards to follow
- What the application does (requirements)
- What tools to use and when
- That they should plan before coding
The result is vibing code — the agent generates code with no direction, no standards, no planning. Fine for prototypes, unacceptable for real projects with teams.
maestro-bundle solves this by installing 4 layers of context that work together:
┌─────────────────────────────────────────────────────────┐
│ │
│ PRD.md WHAT to build │
│ (requirements) User stories, API spec, data │
│ model, acceptance criteria │
│ │
├─────────────────────────────────────────────────────────┤
│ │
│ AGENTS.md HOW to build │
│ (standards) Stack, architecture, conventions, │
│ project structure, git flow │
│ │
├─────────────────────────────────────────────────────────┤
│ │
│ Skills WITH WHAT to build │
│ (capabilities) RAG pipeline, clean architecture, │
│ deploy, testing, etc. │
│ │
├─────────────────────────────────────────────────────────┤
│ │
│ Spec Kit (SDD) IN WHAT ORDER to build │
│ (process) /speckit.specify → plan → tasks │
│ → implement │
│ │
└─────────────────────────────────────────────────────────┘
Each layer gives the agent a different type of context. Together, they make the agent work like a senior dev who knows the project, follows the standards, and plans before coding.
npx maestro-bundle ai-agents claude| What | Purpose | Who fills it |
|---|---|---|
| AGENTS.md | Standards, stack, conventions | Bundle (automatic) |
| Skills | Specific capabilities (RAG, clean arch, tests...) | Bundle (automatic) |
| PRD.md | Product requirements, user stories, API spec | Analyst / Dev |
| Spec Kit | SDD process — spec → plan → tasks → implement | Agent + Dev |
| Constitution | Non-negotiable project principles | Bundle (automatic) |
| LangChain Skills | 11 official LangChain skills (AI bundles) | Bundle (automatic) |
The dev only needs to fill in PRD.md. Everything else comes ready.
1. Dev installs the bundle
$ npx maestro-bundle ai-agents claude
2. Analyst/dev fills PRD.md with requirements
3. Dev opens the editor and requests a feature:
"Create the JWT authentication endpoint"
4. The agent already knows:
- The requirements (PRD.md)
- The stack and standards (AGENTS.md)
- How to do JWT (authentication skill)
- That it needs to create a spec before coding (Spec Kit)
5. Result: governed, consistent, planned code
| Bundle | Project type | Stack |
|---|---|---|
ai-agents |
Multi-agent AI systems | Python, LangChain, LangGraph, FastAPI, pgvector |
ai-agents-deep |
Deep Agent (Claude Code-like) | Python, Deep Agents SDK, LangGraph, Subagents |
jhipster-monorepo |
JHipster monolithic app | Java 21, Spring Boot, Angular, PostgreSQL |
jhipster-microservices |
JHipster microservices | Java 21, Spring Boot, Kafka, Consul, K8s |
data-pipeline |
Data & ML pipeline | Python, Pandas, Scikit-learn, MLflow |
frontend-spa |
Frontend SPA | React, TypeScript, Tailwind, Vite |
npx maestro-bundle ai-agents claude
npx maestro-bundle ai-agents-deep cursor
npx maestro-bundle jhipster-monorepo claude
npx maestro-bundle jhipster-microservices codex
npx maestro-bundle data-pipeline copilot
npx maestro-bundle frontend-spa windsurf| Editor | Command | Where it installs |
|---|---|---|
| Claude Code | npx maestro-bundle <bundle> claude |
CLAUDE.md + .claude/skills/ |
| Cursor | npx maestro-bundle <bundle> cursor |
AGENTS.md + .cursor/skills/ |
| OpenAI Codex | npx maestro-bundle <bundle> codex |
AGENTS.md + .agents/skills/ |
| GitHub Copilot | npx maestro-bundle <bundle> copilot |
.github/copilot-instructions.md |
| Windsurf | npx maestro-bundle <bundle> windsurf |
.windsurfrules |
| All | npx maestro-bundle <bundle> all |
All of the above in the same repo |
$ npx maestro-bundle ai-agents claude
Bundle: Multi-Agent AI System
Editor: Claude Code
✔ Claude Code: AGENTS.md, CLAUDE.md, 14 skills in .claude/skills/
✔ PRD.md template installed
✔ 11 LangChain Skills installed
✔ specify-cli v0.4.3 installed
✔ Spec Kit initialized (/speckit.* commands available)
✔ Bundle constitution integrated
Done!
Next steps:
1. Fill in PRD.md with your product requirements
2. Open the project in your AI editor
3. Use /speckit.specify to start your first feature
your-project/
├── CLAUDE.md # Points to AGENTS.md
├── AGENTS.md # Stack, standards, conventions (automatic)
├── PRD.md # Product requirements (fill this in!)
├── .claude/
│ ├── skills/ # Agent capabilities
│ │ ├── rag-pipeline/
│ │ ├── clean-architecture/
│ │ ├── testing-strategy/
│ │ └── ...
│ └── commands/ # Spec Kit commands
│ ├── speckit.specify.md
│ ├── speckit.plan.md
│ ├── speckit.tasks.md
│ └── speckit.implement.md
├── .specify/ # Feature specs, plans, tasks
│ ├── memory/constitution.md
│ ├── templates/
│ └── specs/
│ └── 001-feature-name/
│ ├── spec.md
│ ├── plan.md
│ └── tasks.md
├── skills/ # Canonical skills (Deep Agents)
└── references/ # Reference docs
The bundle installs GitHub Spec Kit which adds commands to your editor for Specification-Driven Development:
| Command | What it does |
|---|---|
/speckit.specify |
Specify WHAT to build and WHY |
/speckit.plan |
Plan architecture and technical decisions |
/speckit.tasks |
Break down into atomic tasks |
/speckit.implement |
Execute tasks following the plan |
The agent is instructed to follow this flow for new features. Simple bug fixes can go directly.
The ai-agents and ai-agents-deep bundles automatically install the 11 official LangChain skills covering LangChain, LangGraph, and Deep Agents.
- Node.js 18+
- Git
- uv or pip (for Spec Kit — installed automatically)
- AGENTS.md — Universal standard for AI agent instructions
- GitHub Spec Kit — Specification-Driven Development
- LangChain Skills — Official LangChain skills
- Agent Skills — Open standard for agent skills
