LM Orchestra is an intelligent multi-agent orchestration system for Claude Code that automatically selects, coordinates, and manages specialized AI agents to deliver complete solutions for complex software development tasks.
When you give LM Orchestra a task, it:
- Analyzes your request to understand requirements and capabilities needed
- Plans the optimal sequence of specialized agents to invoke
- Executes the workflow with intelligent parallel/sequential coordination
- Validates outputs against domain-specific quality gates
- Synthesizes coherent, production-ready deliverables
User Request: "Build a FastAPI authentication service with JWT and OAuth2"
↓
┌─────────────┐
│ task-router │ → Analyzes intent & extracts capabilities
└──────┬──────┘
↓
┌──────────────┐
│ agent-planner│ → Selects agents & creates execution plan
└──────┬───────┘
↓
[User Approval Gate]
↓
┌──────────────────┐
│workflow-executor │ → Spawns & coordinates agents
└────────┬─────────┘
↓
┌────────────────┐
│output-validator│ → Applies quality gates
└───────┬────────┘
↓
[Pass] → Final Deliverable
[Fail] → Refinement Loop (max 2x)
- 108+ Specialized Agents across 23 domains (backend, frontend, security, DevOps, ML, etc.)
- 72 Focused Plugins with minimal token footprint (avg 3.4 components/plugin)
- 129+ Skills providing domain-specific knowledge on demand
- Intelligent Routing - Maps natural language to capabilities to agents
- Quality Gates - Domain-specific validation (code, security, architecture, testing)
- Refinement Loops - Automatic retry with targeted instructions on failures
- Parallel Execution - Identifies opportunities to run independent agents simultaneously
- User Approval - Full visibility and control before execution begins
- Claude Code CLI installed
- Python 3.8+ (for registry scripts)
-
Clone the repository with submodules
git clone --recursive https://github.com/yourusername/lmorchestra.git cd lmorchestraIf you already cloned without
--recursive, initialize the submodule:git submodule update --init --recursive
-
Install the Meta-Orchestrator plugin
cd meta-orchestrator claude plugin install .
-
Install the Agents ecosystem (optional, for full capabilities)
cd ../agents claude plugin install .
Add these aliases to your shell configuration (~/.bashrc, ~/.zshrc, etc.):
# Full orchestration pipeline
orchestrate() {
claude "/meta-orchestrator:orchestrate $*" \
--allowedTools "Bash,Read,Write,Edit,Grep,Glob,WebFetch,WebSearch,Task" \
--plugin-dir /path/to/lmorchestra/meta-orchestrator
}
# Preview execution plan only
plan() {
claude "/meta-orchestrator:plan $*" \
--allowedTools "Bash,Read,Grep,Glob,Task" \
--plugin-dir /path/to/lmorchestra/meta-orchestrator
}
# Validate existing code
validate() {
claude "/meta-orchestrator:validate $*" \
--allowedTools "Bash,Read,Grep,Glob" \
--plugin-dir /path/to/lmorchestra/meta-orchestrator
}Then run directly from your terminal:
orchestrate "Build a FastAPI CRUD API for inventory management with PostgreSQL"
plan "Migrate Django monolith to microservices"
validate --path src/ --gates securityStart Claude Code with the plugin directory:
claude --plugin-dir /path/to/lmorchestra/meta-orchestratorThen use the commands:
/orchestrate "Build a FastAPI CRUD API for inventory management with PostgreSQL"
/plan "Create a real-time collaborative editor with WebSocket support"
/validate --path src/api/ --gates security,testing --severity highRun a single orchestration task:
claude "/meta-orchestrator:orchestrate Build a user auth system with JWT" \
--allowedTools "Bash,Read,Write,Edit,Grep,Glob,WebFetch,WebSearch,Task" \
--plugin-dir /path/to/lmorchestra/meta-orchestrator| Command | Description | Example |
|---|---|---|
/orchestrate |
Full pipeline: analyze → plan → execute → validate → deliver | /orchestrate "Build user auth with JWT" |
/plan |
Preview execution plan without running | /plan "Migrate Django to FastAPI" |
/validate |
Run quality gates on existing code | /validate --path src/ --gates security |
lmorchestra/
├── agents/ # Git submodule → github.com/1fc0nfig/agents
│ ├── plugins/ # 72 specialized plugin directories
│ │ ├── backend-development/
│ │ ├── frontend-mobile-development/
│ │ ├── kubernetes-operations/
│ │ ├── python-development/
│ │ ├── security-scanning/
│ │ └── ... (67 more)
│ └── docs/
│
├── meta-orchestrator/ # Orchestration layer
│ ├── agents/ # 5 orchestration agents
│ ├── commands/ # 3 user-facing commands
│ ├── skills/ # 3 knowledge skills
│ ├── registry/ # Generated agent/capability indexes
│ └── scripts/ # Registry build tools
│
└── README.md # This file
| Agent | Model | Purpose |
|---|---|---|
task-router |
Sonnet | Analyzes user intent and extracts required capabilities |
agent-planner |
Opus | Creates optimal multi-agent execution plans |
workflow-executor |
Opus | Coordinates agent invocation and context passing |
output-validator |
Sonnet | Applies domain-specific quality gates |
result-synthesizer |
Sonnet | Merges multi-agent outputs into unified deliverables |
| Skill | Purpose |
|---|---|
agent-registry |
Complete knowledge of all 108+ agents and selection criteria |
capability-mapping |
Maps keywords → capabilities → agents |
quality-gates |
Domain-specific validation criteria by deliverable type |
The agents are organized into 23 categories:
| Category | Plugins | Example Agents |
|---|---|---|
| Backend | 4 | backend-architect, fastapi-pro, graphql-architect |
| Frontend | 4 | frontend-developer, react-pro, mobile-developer |
| Infrastructure | 5 | cloud-architect, kubernetes-operator, terraform-pro |
| Security | 4 | security-auditor, threat-modeling-expert, compliance-checker |
| Data & ML | 6 | database-architect, ml-engineer, llm-app-builder |
| Quality | 3 | code-reviewer, test-automator, performance-engineer |
| DevOps | 4 | cicd-architect, deployment-specialist, observability-expert |
| Languages | 7 | python-pro, typescript-pro, rust-pro, go-pro |
| Documentation | 3 | api-documenter, c4-architect, technical-writer |
| ... | ... | ... |
See agents/README.md for the complete catalog.
/orchestrate "Build a FastAPI REST API for user management with PostgreSQL"Generated Plan:
database-architect→ Design schema for users tablebackend-architect→ Define API contract and endpointsfastapi-pro→ Implement the APItest-automator→ Generate tests (parallel)security-auditor→ Security review (parallel with tests)
/orchestrate "Create a dashboard with React frontend, FastAPI backend, and PostgreSQL"Generated Plan:
database-architect→ Schema designbackend-architect+frontend-developer→ API & UI design (parallel)fastapi-pro→ Backend implementationreact-pro→ Frontend implementation (parallel with backend)test-automator→ E2E testssecurity-auditor→ Final security review
/orchestrate "Perform a comprehensive security audit of our authentication system"Generated Plan:
security-auditor→ OWASP vulnerability scanthreat-modeling-expert→ Threat analysisbackend-security-coder→ Generate fix recommendationsresult-synthesizer→ Produce security report
The registry indexes all agents, skills, and capabilities for intelligent selection:
registry/
├── agents.json # 108+ agents with capabilities, triggers, paths
├── capabilities.json # Capability-to-agent reverse mappings
├── skills.json # 129+ skills indexed
└── plugins.json # 72 plugins with components
When the agents repository is updated:
cd meta-orchestrator
python scripts/build-registry.py --source ../agents --output ./registry
python scripts/validate-registry.py --registry ./registryThe output-validator applies these domain-specific checks:
| Gate | Checks |
|---|---|
| Code Quality | Syntax validity, error handling, type safety, documentation |
| Security | Input validation, auth patterns, injection prevention, OWASP compliance |
| Architecture | Separation of concerns, design patterns, dependency management |
| Testing | Test coverage, edge cases, mocking patterns |
| Database | Schema design, indexing, migration safety, query performance |
Issues are categorized by severity (critical/high/medium/low) with specific remediation instructions.
LM Orchestra uses a cost-optimized three-tier model strategy:
| Tier | Model | Use Case | Cost |
|---|---|---|---|
| Heavy | Opus 4.5 | Complex planning, critical decisions | $5/$25 per 1M tokens |
| Balanced | Sonnet 4.5 | Routing, validation, synthesis | $3/$15 per 1M tokens |
| Fast | Haiku 4.5 | Simple operations, quick checks | $1/$5 per 1M tokens |
The plugin can be configured via .claude-plugin/marketplace.json:
{
"metadata": {
"model_recommendations": {
"task-router": "sonnet",
"agent-planner": "opus",
"workflow-executor": "opus",
"output-validator": "sonnet",
"result-synthesizer": "sonnet"
}
}
}We welcome contributions! See CONTRIBUTING.md for guidelines.
- Clone the repository
- Create a Python virtual environment:
python -m venv venv source venv/bin/activate # or `venv\Scripts\activate` on Windows
- Make your changes
- Test with
/orchestratecommands - Submit a pull request
This project is licensed under the MIT License - see the LICENSE file for details.
- Built on the wshobson/agents ecosystem
- Designed for Claude Code by Anthropic
- 1fc0nfig/agents - Fork of the agent ecosystem (included as submodule)
- wshobson/agents - Original agent ecosystem
- Claude Code - The AI-powered CLI this plugin extends