AI Maestro Orchestrator Agent (amoa-)

June 20, 2026 · View on GitHub

Version: 1.9.0

A Claude Code plugin that turns one Claude session into a multi-agent orchestrator: it breaks a plan into GitHub-Issue-backed modules, assigns them to sub-agents and remote agents over AI Maestro inter-agent messaging, tracks progress on a GitHub Projects V2 kanban board, and enforces a verification gate before the session is allowed to exit.

Overview

The Orchestrator Agent handles task distribution, agent coordination, and progress monitoring for multi-agent projects. It receives plans and coordinates subagents to implement them via AI Maestro inter-agent messaging.

Requirements

  • Python 3.8+ with PyYAML (pip install pyyaml)
  • GitHub CLI (gh) for issue and project management
  • AI Maestro messaging system for inter-agent communication

Core Responsibilities

  1. Task Distribution: Break plans into assignable tasks with dependency ordering
  2. Agent Coordination: Manage subagents and remote agents via AI Maestro
  3. Progress Monitoring: Track task completion via polling hooks
  4. Module Management: Organize work into GitHub Issue-backed modules
  5. Verification: Ensure instructions are followed correctly before exit
  6. Kanban Management: GitHub Projects V2 board and column management

Components

Agents (6)

AgentDescription
ai-maestro-orchestrator-agent-main-agentMain orchestrator — delegates to sub-agents
amoa-team-orchestratorCoordinates team of remote agents
amoa-task-summarizerSummarizes task progress and state
amoa-checklist-compilerCreates verification checklists
amoa-docker-container-expertDocker and container expertise
amoa-experimenterExperimentation and prototyping

Commands (15)

CommandDescription
/amoa-start-orchestrationStart orchestration phase
/amoa-orchestrator-loopStart orchestration loop
/amoa-orchestration-statusShow orchestration phase status (modules/agents)
/amoa-orchestrator-statusShow orchestrator loop status (iterations/tasks)
/amoa-cancel-orchestratorCancel orchestration
/amoa-register-agentRegister remote agent
/amoa-check-agentsCheck agent status and poll progress
/amoa-assign-moduleAssign module to agent
/amoa-reassign-moduleReassign module to different agent
/amoa-add-moduleAdd new module
/amoa-modify-moduleModify existing module
/amoa-remove-moduleRemove module
/amoa-prioritize-moduleChange module priority
/amoa-reassign-kanban-tasksReassign kanban board tasks
/amoa-generate-replacement-handoffGenerate agent replacement handoff document

Skills (16)

SkillDescription
amoa-two-phase-modePlan-then-Execute workflows with formal approval
amoa-orchestration-commandsOrchestration command reference and loop mechanics
amoa-orchestration-patternsTask breakdown patterns for human developers
amoa-remote-agent-coordinatorRemote AI agent coordination via AI Maestro
amoa-module-managementModule CRUD operations (1:1 with GitHub Issues)
amoa-verification-patternsImplementation verification and evidence collection
amoa-developer-communicationDeveloper communication patterns and templates
amoa-checklist-compilation-patternsVerification checklist generation
amoa-agent-replacementAgent replacement and handoff protocols
amoa-task-distributionTask breakdown, assignment, and load balancing
amoa-progress-monitoringProgress tracking via state-based polling
amoa-messaging-templatesAI Maestro message format templates
amoa-label-taxonomyGitHub label system for multi-agent coordination
amoa-implementer-interview-protocolInterview-based task verification
amoa-github-action-integrationGitHub Actions CI/CD patterns
amoa-kanban-managementGitHub Projects V2 kanban board management

Hooks (4)

HookEventDescription
amoa-orchestrator-stopStopBlocks exit until all tasks complete (120s timeout)
amoa-instruction-verification-checkPreToolUseBlocks agent work if verification incomplete
amoa-polling-reminderUserPromptSubmitReminds to poll agent progress
amoa-file-trackerPostToolUseTracks Edit/Write file modifications

Project Structure

ai-maestro-orchestrator-agent/
├── .claude-plugin/      # Plugin manifest (plugin.json)
├── agents/              # Agent definitions (6 agents)
├── commands/            # Slash command definitions (15 commands)
├── docs/                # Architecture docs (role boundaries, workflow, registry)
├── git-hooks/           # Git hooks (pre-push validation)
├── hooks/               # Plugin hooks (hooks.json + 4 hook scripts)
├── scripts/             # Python scripts (56 scripts + validation suite)
│   └── amoa_stop_check/ # Stop hook package (lock, tasks, phase checks)
├── shared/              # Shared resources across skills
├── skills/              # Skill definitions (16 skills with references)
├── tests/               # Test suite
└── requirements.txt     # Python dependencies (PyYAML)

Workflow

  1. Plan Phase: Define modules and requirements (two-phase mode)
  2. Assignment: Assign modules to agents via kanban board
  3. Execution: Agents implement modules, report via AI Maestro
  4. Monitoring: Poll agent progress, detect stalls, handle failures
  5. Verification: 4-loop verification ensures all tasks complete
  6. Completion: Stop hook enforces completion before session exit

Usage

Drive the orchestrator from its slash commands once the plugin is installed (run /help to list the full amoa- command set). The common entry points:

/amoa-start-orchestration   # begin a new orchestration run from a plan
/amoa-orchestration-status  # show the live module/kanban/agent state
/amoa-cancel-orchestrator   # stop the active orchestrator loop

Underlying tooling is also runnable directly for scripting:

uv run python scripts/amoa_kanban_manager.py <command>   # kanban board ops
uv run python scripts/amoa_orchestration_status.py        # status snapshot

Installation

Install using --plugin-dir for local development:

claude --plugin-dir /path/to/ai-maestro-orchestrator-agent

Start a session with the main orchestrator agent:

claude --agent ai-maestro-orchestrator-agent-main-agent

After modifying plugin files, reload without restarting:

/reload-plugins

Validation

Run the CPV (Claude Plugins Validation) suite. Validation is delegated entirely to CPV — this plugin ships no local validators:

cd ai-maestro-orchestrator-agent
uvx --from git+https://github.com/Emasoft/claude-plugins-validation --with pyyaml \
    cpv-remote-validate plugin . --strict --verbose

scripts/publish.py re-runs this exact --strict gate (plus lint and the test suite) immediately before every push. The pre-push hook only enforces that pushes go through publish.py (a process-ancestry check); it runs no validation of its own.

Development

  • Kanban management: uv run python scripts/amoa_kanban_manager.py <command>
  • GitHub project columns: uv run python scripts/gh-project-add-columns.py

License

MIT