Maestro

AI agent workflow skillpack. Plan first, code once.

Install

Universal install (skills.sh CLI):

npx skills add ReinaMacCredy/maestro

Codex install (optional runtime):

bash scripts/install-codex.sh

Codex install with optional ECC skillpack:

bash scripts/install-codex.sh --with-ecc-skillpack

Install specific skills and agents:

npx skills add ReinaMacCredy/maestro --skill planning --agent claude-code --agent amp

Legacy direct installs (still supported):

# Claude Code plugin route
/plugin install https://github.com/ReinaMacCredy/maestro

# Amp route
amp skill add https://github.com/ReinaMacCredy/maestro --global

Quick Start

graph LR
    A["/maestro:setup"] --> B["/maestro:new-track"]
    B --> C[Spec + Plan Generated]
    C --> D["/maestro:implement"]
    D --> E["/maestro:review"]

1. /maestro:setup — Scaffold project context (product, tech stack, guidelines)
2. /maestro:new-track — Create a feature/bug track with spec and plan
3. /maestro:implement — Execute tasks (single-agent or --team for parallel)
4. /maestro:review — Verify implementation correctness

Setup

Enable Agent Teams in ~/.claude/settings.json:

{
  "env": {
    "CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS": "1"
  }
}

Commands

Maestro command names are unchanged across Claude Code, Amp, and Codex compatibility paths.

Core Workflow

Command	Description
/maestro:setup	Scaffold project context (product, tech stack, guidelines)
/maestro:new-track <request>	Create a feature/bug track with spec and plan
/maestro:implement [<track>] [--team]	Execute track tasks (supports --resume)
/maestro:review	Post-execution review with auto-fix
/maestro:status	Show track progress across all tracks
/maestro:revert	Undo implementation if needed
/maestro:note [--priority <P0-P3>] <text>	Capture decisions and context to persistent notepad
/maestro:AGENTS.md	Generate AGENTS.md context file

Agents

Agent	Purpose	Model	Team Lead?
prometheus	Interview-driven planner	sonnet	Yes
orchestrator	Execution coordinator	sonnet	Yes
kraken	TDD implementation	sonnet	No
spark	Quick fixes	sonnet	No
build-fixer	Build/compile/lint error specialist	sonnet	No
oracle	Strategic advisor	sonnet	No
critic	Post-implementation reviewer	sonnet	No
security-reviewer	Security analysis specialist	sonnet	No
explore	Codebase search	haiku	No
leviathan	Deep plan reviewer	sonnet	No
wisdom-synthesizer	Knowledge consolidator	haiku	No
progress-reporter	Status tracker	haiku	No

Key Rules

• Plan before code — Run /maestro:new-track to create a spec and plan before implementing
• TDD by default — Use kraken for new features
• Track-based workflow — All work is organized into tracks with specs and plans
• Verify subagent claims — Always verify, agents can make mistakes

Recommended MCP Servers

These MCP servers enhance the Maestro experience:

Server	Purpose
Context7	Up-to-date library documentation
Sequential Thinking	Dynamic reasoning for complex planning

Documentation

Topic	Path
Skills	skills/
Agent Definitions	.claude/agents/
Agent Teams Guide	docs/AGENT-TEAMS.md
Universal Skills Format	docs/AGENT-SKILLS.md
Codex Compatibility	.codex/

Optional ECC Skillpack

Maestro includes an optional ECC-derived skillpack under skillpacks/ecc/skills/. It is intentionally not active by default.

# Enable all optional ECC skills
bash scripts/enable-ecc-skillpack.sh all

# Disable all optional ECC skills
bash scripts/disable-ecc-skillpack.sh all

ECC hook toggles:

# Disable ECC quality gates (default is enabled)
export MAESTRO_ENABLE_ECC_QUALITY_GATES=0

# Enable ECC learning observer hooks (default is disabled)
export MAESTRO_ENABLE_LEARNING_HOOKS=1

Credits

• BMAD-METHOD — Multi-agent design methodology
• conductor — Context-driven development patterns

License

MIT