Agent Skill
2/7/2026

multi-agent-coordination

This skill should be used when the user asks about "multi-agent coordination", "agent spawning", "parallel agents", "agent communication", "team coordination", "sub-agent management", "agent limits", "agent layers", or needs guidance on coordinating multiple agents for complex tasks.

L
lobbi
2GitHub Stars
2Views
npx skills add Lobbi-Docs/claude

SKILL.md

Namemulti-agent-coordination
DescriptionThis skill should be used when the user asks about "multi-agent coordination", "agent spawning", "parallel agents", "agent communication", "team coordination", "sub-agent management", "agent limits", "agent layers", or needs guidance on coordinating multiple agents for complex tasks.

Neural Orchestration Platform (Golden Armada)

TypeScript React Vite License

A plugin-based AI agent orchestration platform that combines a visual workflow builder with a deep Claude Code configuration system. Fourteen domain plugins extend the platform across cloud infrastructure, enterprise SaaS, home automation, and more. The root plugin (claude-orchestration v4.0.1) ships 137 agents, 55 skills, and 100+ slash commands that activate automatically through a registry-backed keyword system.

The frontend (@accos/frontend v1.0.0) is the control surface: a React 18 application where you design, visualize, and manage multi-agent workflows and the plugin marketplace.

Architecture

┌─────────────────────────────────────────────────────────────────┐
│                    Visual Workflow Builder                       │
│          ReactFlow canvas · Zustand state · WebSocket           │
└────────────────────────┬────────────────────────────────────────┘
                         │
┌────────────────────────▼────────────────────────────────────────┐
│                      Plugin System Core                         │
│   Discovery · Installation · Validation · Dependency resolution │
│   Health checking · Sandboxing · Federated registry             │
└───────────┬────────────────────────────────┬────────────────────┘
            │                                │
┌───────────▼──────────┐        ┌────────────▼───────────────────┐
│    14 Domain Plugins │        │       Agent Fleet              │
│  jira-orchestrator   │        │  137 agents · 30+ categories   │
│  tvs-microsoft-deploy│        │  55 skills · 100+ commands     │
│  home-assistant-arch │        │  Activated via registry index  │
│  fastapi-backend     │        └────────────────────────────────┘
│  ... and 10 more     │
└───────────┬──────────┘
            │
┌───────────▼──────────────────────────────────────────────────────┐
│                        MCP Servers                               │
│  deploy-intelligence · lessons-learned · project-metrics        │
│  code-quality-gate · workflow-bridge · perplexity · firecrawl   │
└──────────────────────────────────────────────────────────────────┘

Features

Visual Workflow Builder

  • Drag-and-drop ReactFlow canvas with 27 node types (Trigger, Agent, Action, Phase, Control, Terminator)
  • Real-time execution tracking over WebSocket
  • JSON Schema-driven property forms with Zod validation
  • Pre-built workflow templates (Jira dev cycle, member management, theme deployment)

Claude Code Configuration System

  • 137 agents across 30+ domain categories, invoked via the Task tool
  • 55 skills providing reusable procedures activated by keyword triggers
  • 100+ slash commands mapped through a registry quickLookup index
  • 11 lifecycle hooks for safety validation, quality gates, and self-healing
  • 5 custom MCP servers exposing deploy state, code quality, metrics, and workflow pipelines

Plugin System

  • Install plugins from registry, Git URL, or local path
  • Automatic semver dependency resolution
  • Comprehensive manifest validation against JSON schema
  • Supply chain security via Sigstore signing and composite trust scoring
  • Sandboxed execution with configurable permission grants
  • Plugin CLI for scaffolding, linting, bundling, and diagnosing plugins

Quick Start

Prerequisites: Node.js 18+, pnpm

# Install dependencies
pnpm install

# Start development server
pnpm dev
# Opens at http://localhost:5173

# Build for production
pnpm build

Project Structure

/
├── src/                        # React frontend application
│   ├── components/             # UI components (Canvas, Nodes, Palette, Properties)
│   ├── stores/                 # Zustand state (workflow, palette)
│   ├── lib/                    # API client, WebSocket, utilities
│   ├── types/                  # TypeScript types (workflow, nodes, plugins)
│   └── hooks/                  # React hooks (usePlugins, useNodeTypes)
├── plugins/                    # 14 installed domain plugins
│   ├── jira-orchestrator/      # Enterprise Jira + Harness orchestration
│   ├── tvs-microsoft-deploy/   # Microsoft Power Platform deployment
│   ├── home-assistant-architect/  # Smart home automation platform
│   └── ...                     # 11 more plugins
├── .claude/                    # Claude Code configuration hub
│   ├── agents/                 # 137 specialized subagents
│   ├── skills/                 # 55 reusable skill procedures
│   ├── rules/                  # 8 behavioral rule files
│   ├── hooks/                  # 11 lifecycle hook scripts
│   ├── mcp-servers/            # 5 custom MCP server implementations
│   ├── plugins/                # 6 available marketplace plugins
│   └── tools/plugin-cli/       # Plugin development CLI (TypeScript)
├── docs/                       # Extended documentation
│   ├── architecture/           # Architecture reviews and roadmaps
│   ├── security/               # Security review reports
│   ├── testing/                # Test strategy and guides
│   └── planning/               # Implementation plans
├── schemas/                    # JSON Schema for plugin manifests
├── .mcp.json                   # MCP server registrations
└── .claude-plugin/plugin.json  # Root plugin manifest (claude-orchestration v4.0.1)

Installed Plugins

These 14 plugins are active and ready to use. Each contributes agents, skills, and slash commands to the platform.

PluginVersionPurpose
jira-orchestrator7.5.0Enterprise Jira with 77 agents, 16 teams, 45 commands, Atlassian MCP OAuth, Temporal workflows
tvs-microsoft-deploy1.1.0Microsoft ecosystem orchestrator: Power Platform, Dataverse, Fabric, Azure, 19 agents, 18 commands
home-assistant-architect2.0.0Complete Home Assistant platform with energy, cameras, local LLM (Ollama), Ubuntu deployment
marketplace-pro1.0.0Advanced marketplace with intent-based composition, supply chain security, federated registry
exec-automator1.0.0AI-powered executive director automation for nonprofits and trade associations (LangGraph)
fullstack-iac1.0.0Full-stack IaC: FastAPI + React/Vite + Ansible + Terraform + Kubernetes
fastapi-backend0.1.0FastAPI development with MongoDB/Beanie, Keycloak auth, Redis, Celery, observability
aws-eks-helm-keycloak1.0.0AWS EKS deployments with Helm, Keycloak OIDC, and Harness CI/CD pipelines
claude-code-templating1.0.0Universal templating: Handlebars, Cookiecutter, Copier, Maven archetypes, Harness pipelines
frontend-design-system1.0.0263+ design styles with multi-tenant Keycloak theming
react-animation-studio1.1.0Web animations: Framer Motion, GSAP, 3D transforms, scroll, text, particles
team-accelerator1.0.0Enterprise DevOps toolkit: code quality, multi-cloud deployments, Playwright, Prometheus
lobbi-platform-manager1.0.0Keycloak management, MERN service orchestration, and test generation for keycloak-alpha
deployment-pipeline1.0.0Harness CD integration with state-machine workflow orchestration

Available Marketplace Plugins

These 6 plugins are ready to install from .claude/plugins/.

PluginVersionPurpose
langgraph-architect1.0.0LangGraph/LangChain agent builder with MCP integration and CLI accessibility
api-integration-helper1.0.0Production-grade API integration: typed clients, auth flows, rate limiting from OpenAPI/GraphQL specs
code-quality-orchestrator1.0.0Five quality gates (static analysis, coverage, security, complexity, dependency health)
migration-wizard1.0.0Framework migrations with codemods, strangler fig patterns, rollback strategies
testforge1.0.0Intelligent test generation for Jest, Pytest, Vitest — catches real bugs, not just coverage numbers
dev-environment-bootstrap1.0.0Developer onboarding: Docker configs, .env templates, IDE setup, dependency detection

Install a plugin:

/plugin-install langgraph-architect
/plugin-install ./path/to/local-plugin --dev

Agent Categories

The 137 agents in .claude/agents/ are organized into 30+ domain categories.

CategoryAgentsExamples
business-sales8b2b-project-shipper, customer-success-manager, finance-tracker
cloud1pulumi-specialist
data-ai4analytics-reporter, experiment-tracker, trend-researcher
design-ux6ux-researcher, brand-guardian, visual-storyteller
development11code-architect, ai-engineer, graphql-specialist, redis-specialist
devops2ansible-specialist, infrastructure-specialist
documentation4generate-api-docs, analyze-codebase, context7-docs-fetcher
frontend-theming3theme-system-architect, theme-builder, white-label-specialist
github7create-pull-request, bug-fix, analyze-issue, create-worktrees
helm3helm-chart-developer, helm-release-manager, helm-values-manager
keycloak5keycloak-realm-admin, keycloak-auth-flow-designer, keycloak-security-auditor
kubernetes4k8s-architect, k8s-debugger, k8s-security-specialist, k8s-resource-optimizer
marketing-growth7growth-hacker, content-creator, app-store-optimizer
membership-domain4membership-specialist, directory-manager, member-engagement-agent
messaging3kafka-specialist, rabbitmq-specialist, event-streaming-architect
migration2code-migration-specialist, refactoring-guru
mobile4ios-specialist, android-specialist, flutter-specialist, react-native-specialist
mongodb-atlas4mongodb-schema-designer, mongodb-query-optimizer, mongodb-aggregation-specialist
multi-tenant2multi-tenant-architect, tenant-provisioning-specialist
orchestration7+lyra, agent-sdk-dev, model-context-protocol-mcp-expert, feature-dev
security3+security-hardening-specialist, compliance-auditor
testing2+test-automation-engineer, performance-tester
Top-level6code-reviewer, debugger, doc-writer, docker-ops, k8s-image-auditor

Skills

Skills are reusable procedures in .claude/skills/. A keyword match in the registry activates the corresponding SKILL.md, which defines tools, steps, and constraints for that domain.

Reasoning & AI complex-reasoning, deep-research, deep-analysis, extended-thinking, llm-integration, prompt-caching, streaming, tool-use, vision-multimodal

Infrastructure & Cloud aws, gcp, terraform, kubernetes, helm-deploy, docker-build, k8s-image-audit, keycloak, authentication

Development react, nextjs, fastapi, flask-api, graphql, rest-api, database, redis, mongodb-atlas, testing, debugging, git-workflows, batch-processing

Project & Team jira, jira-branch, jira-commit, jira-pr, jira-work, jira-sync, jira-triage, jira-status, jira-review, jira-prepare, scrum, kanban, atlassian-api, confluence

Design & Research design-system, citations-retrieval, deep-research, scrape-docs, web-research

Payments & Integrations stripe-payments, member-management, multi-tenant, vector-db

MCP Servers

Custom (5)

ServerPurpose
deploy-intelligenceDocker build history, deployment state, rollback tracking
lessons-learnedRead/write access to the self-healing lessons-learned database
project-metricsLine counts, test coverage stats, plugin inventory
code-quality-gateESLint scores, TypeScript errors, quality gate pass/fail
workflow-bridgePipeline state and workflow orchestration data

External (2)

ServerPurpose
perplexityWeb research and current-events queries via Perplexity API
firecrawlStructured scraping of specific URLs

Configured in .mcp.json. API keys are injected via environment variables at startup.

Hooks and Lifecycle

Eleven hooks in .claude/hooks/ enforce safety, quality, and self-healing across the session lifecycle.

Hook EventScriptPurpose
SessionStartsession-init.shLoad rules, print tool reminders on startup
PreToolUse (Bash)bash-safety-validator.shBlock destructive commands before execution
PreToolUse (Edit/Write)protect-critical-files.shGuard manifests and secrets from accidental overwrite
PreToolUse (Bash)helm-deploy-validator.shValidate Helm deployment commands before running
PostToolUse (Edit/Write)post-edit-lint.shAuto-lint changed TypeScript/JavaScript files
PostToolUse (Bash)docker-build-tracker.shRecord Docker build outcomes to MCP server
PostToolUseFailurelessons-learned-capture.shAppend failures to rules/lessons-learned.md
Stopprompt hookVerify task completion; surface uncommitted changes
TeammateIdleteammate-idle-check.shMonitor and report on idle subagent teammates
TaskCompletedtask-quality-gate.shRun TypeScript check and merge-conflict scan on completion
SubagentStopsubagent-results-log.shAsync log of subagent output for audit trail

Configuration

FilePurpose
.mcp.jsonMCP server registrations (command, args, env vars)
.claude/settings.jsonModel selection, teammate mode, permission allowlist/denylist, hook bindings
.claude/CLAUDE.mdPrimary project instructions: workflow protocol, model routing, key paths
.claude/rules/Eight modular rule files (git-workflow, architecture, research, lessons-learned, etc.)
.claude-plugin/plugin.jsonRoot plugin manifest (claude-orchestration v4.0.1)
schemas/plugin.schema.jsonJSON Schema all plugin manifests validate against

Permission model

settings.json maintains an explicit allowlist for Bash subcommands (npm, pnpm, git, docker, kubectl, helm, az, terraform, etc.) and a denylist that blocks destructive operations (rm -rf /) and prevents reading .env files or secrets directories.

Development

Scripts

pnpm dev                        # Start Vite dev server (http://localhost:5173)
pnpm build                      # Type-check + generate plugin indexes + Vite build
pnpm test                       # Run Vitest test suite
pnpm test:ui                    # Vitest with browser UI
pnpm test:e2e                   # Playwright end-to-end tests
pnpm lint                       # ESLint with zero-warnings policy
pnpm type-check                 # tsc --noEmit
pnpm generate:plugin-indexes    # Regenerate plugin index files from manifests
pnpm check:plugin-schema        # Validate all plugin.json files against schema
pnpm check:plugin-context       # Validate plugin context budgets
pnpm check:hooks                # Lint hook scripts for correctness

Plugin CLI

The plugin CLI lives at .claude/tools/plugin-cli/ and is used to create and maintain plugins.

# Scaffold a new plugin
node .claude/tools/plugin-cli/bin/plugin-cli.js init my-plugin --type full

# Validate a plugin's structure and manifest
node .claude/tools/plugin-cli/bin/plugin-cli.js validate plugins/my-plugin

# Lint for best practices
node .claude/tools/plugin-cli/bin/plugin-cli.js lint plugins/my-plugin

# Bundle for distribution
node .claude/tools/plugin-cli/bin/plugin-cli.js build plugins/my-plugin -o dist/

# Diagnose common issues
node .claude/tools/plugin-cli/bin/plugin-cli.js doctor plugins/my-plugin

Plugin templates: full, agent-pack, skill-pack, workflow-pack.

Tech Stack

LayerTechnology
FrameworkReact 18, TypeScript 5.3
BuildVite 5
StateZustand 4
Workflow canvasReactFlow 11
Server stateTanStack Query 5
ValidationZod 3
AnimationFramer Motion 11
StylingTailwind CSS 3
Real-timesocket.io-client 4
Code editorMonaco Editor
TestingVitest 1, Playwright, Testing Library

Documentation

DirectoryContents
docs/architecture/Architecture quality review, improvement roadmap
docs/security/Security review report, implementation guide, quick reference
docs/testing/Test strategy, implementation guide, execution quick reference
docs/planning/Decomposition plans, resource allocation, execution timelines
docs/reference/Commands reference, dependency graph, plugin dependency analysis
docs/governance/Control matrix
.claude/README.mdClaude configuration hub — component relationships and data flow
.claude/PLUGIN_SYSTEM.mdPlugin installation and management reference

Contributing

  1. Follow the EXPLORE → PLAN → CODE → TEST → FIX → DOCUMENT workflow defined in .claude/CLAUDE.md.
  2. Use pnpm lint and pnpm type-check before committing.
  3. Plugin changes must pass pnpm check:plugin-schema and pnpm check:plugin-context.
  4. Commit messages use imperative mood: type(scope): description (max 72 chars).
  5. Never commit .env files, credentials, or secrets.

License

MIT — see individual plugin manifests for per-plugin license terms.

Skills Info
Original Name:multi-agent-coordinationAuthor:lobbi