Maestro

Full-Stack Autonomous Development Orchestrator for Claude Code

Build features with /maestro. Build entire products with /maestro opus. Track progress on your kanban board. Persist knowledge in your second brain.

---

What is Maestro?

Maestro is a Claude Code plugin that orchestrates autonomous software development across three layers: Vision and Strategy (product research, competitive analysis, market positioning), Tactics and Architecture (decomposition, dependency graphs, tech stack decisions), and Execution (implement, self-heal, QA review, git craft). Each layer delegates to purpose-built agents that receive exactly the context they need.

Maestro tracks token costs, learns from every session, and progressively earns your trust through demonstrated reliability. It integrates with your existing tools — Asana, Jira, Linear, GitHub Issues for project management, and Obsidian or Notion as a persistent knowledge base.

---

Quick Start

# Install the plugin
claude plugin install maestro

# Initialize for your project
/maestro init

# Build a feature
/maestro "Add user authentication with email/password"

# Build an entire product
/maestro opus "Personal finance dashboard with budget tracking"

---

Commands

Command	Description
/maestro "task"	Build a feature autonomously
/maestro opus "vision"	Build an entire product (Magnum Opus)
/maestro init	Initialize for this project
/maestro status	View progress, resume, pause, abort
/maestro model	View/change model assignments
/maestro help [topic]	Contextual help and FAQ
/maestro doctor	Health check and diagnostics
/maestro config	View/edit configuration
/maestro board	Kanban board view
/maestro brain	Second brain operations
/maestro history	Past sessions and cost analysis
/maestro plan "task"	Deep planning with codebase exploration
/maestro notify	Push notifications (Slack, Discord, Telegram)
/maestro viz	Visual dashboards and Mermaid diagrams
/maestro demo	Interactive demo — learn how Maestro works
/maestro quick-start	Pick from pre-built task templates

---

Three-Layer Architecture

User Request
    |
    v
[Classifier] --> routes to the right layer
    |
    +---> Layer 1: Vision & Strategy
    |       research, strategy, opus-loop
    |
    +---> Layer 2: Tactics & Architecture
    |       decompose, architecture, forecast
    |
    +---> Layer 3: Execution
            dev-loop, git-craft, ship, preview

Execution: 7-Phase Dev Loop

For each story in dependency order:

VALIDATE > DELEGATE > IMPLEMENT > SELF-HEAL > QA > GIT > CHECKPOINT

1. Validate — Check prerequisites and dependencies
2. Delegate — Build right-sized context, select model
3. Implement — Dispatch implementer agent (TDD)
4. Self-Heal — Run checks, auto-fix failures (up to 3x)
5. QA Review — Dispatch QA reviewer (read-only, separate agent)
6. Git Craft — Create documentation-quality commit
7. Checkpoint — User review (mode-dependent)

Execution Modes

Mode	Behavior
--yolo	Auto-approve everything, maximum speed
--checkpoint	Pause after each story for review (default)
--careful	Pause after each phase for granular control

---

Magnum Opus

Build entire products with /maestro opus "vision".

1. Deep Interview — 10-dimension adaptive vision interview
2. Research Sprint — 8 parallel research agents (competitors, tech, architecture)
3. Roadmap — Milestone-driven plan with cost estimates
4. Build Loop — Decompose, implement, evaluate, auto-fix per milestone
5. Live Chat — Talk to Maestro while it builds

---

Integrations

Kanban (Project Management)

Sync stories with your project management tool. Stories appear as cards, status updates flow bidirectionally.

/maestro config set integrations.kanban.provider github
/maestro board

Provider	Setup
GitHub Issues	gh CLI (no extra setup)
Asana	Asana MCP Server
Jira	Atlassian MCP Server
Linear	Linear MCP Server

Second Brain (Knowledge Base)

Persist decisions, learnings, and session summaries across sessions.

/maestro brain connect
/maestro brain search "authentication"
/maestro brain tldr

Provider	Setup
Obsidian	Enable CLI in Obsidian Settings
Notion	Notion MCP Server

Diagnostics

/maestro doctor

Checks: core files, config, trust metrics, git status, hooks, MCP servers, integrations.

---

Progressive Trust

Maestro tracks reliability and adapts behavior:

Level	Requirement
Novice	< 5 stories
Apprentice	5-15 stories, > 60% QA first-pass
Journeyman	15-30 stories, > 75% QA rate
Expert	30+ stories, > 85% QA rate

Higher trust unlocks more autonomy (yolo mode becomes default for Expert).

---

Cost Tracking

Maestro estimates costs before starting and tracks actual spend per story.

/maestro history cost
/maestro model

Model costs (per million tokens):

• Haiku: $0.80 / $4.00
• Sonnet: $3.00 / $15.00
• Opus: $15.00 / $75.00

---

Session Memory

Dual-sector memory persists across sessions:

• Semantic: long-term facts, preferences, architecture decisions
• Episodic: session context with salience decay (auto-prunes after ~10 sessions)

Memory is injected into agent context for continuity.

---

Agents

Agent	Model	Role
Implementer	Sonnet	TDD story implementation
QA Reviewer	Opus	Skeptical code review (read-only)
Researcher	Sonnet	Competitive analysis, web research
Strategist	Opus	Marketing, growth, positioning
Fixer	Sonnet	Targeted error fixes (self-heal)
Proactive	Haiku	Background monitoring, health checks

---

File Structure

.maestro/
  dna.md              Project DNA (tech stack, patterns)
  config.yaml         Configuration
  trust.yaml          Trust metrics
  state.md            Persistent project state
  state.local.md      Session state (gitignored)
  stories/            Story spec files
  logs/               Session logs
  research/           Research output
  memory/
    semantic.md       Long-term facts
    episodic.md       Session context (decays)

---

License

MIT