Skip to content

Core Concepts

Understanding these core concepts will help you get the most out of Orchestre V3. Let's explore what makes Orchestre different from traditional development tools.

Dynamic Prompt Engineering

Traditional vs Dynamic Approach

❌ Traditional Tools

  • Follow rigid scripts
  • One-size-fits-all solutions
  • Ignore project context
  • Generate boilerplate blindly

✅ Orchestre Dynamic Prompts

  • Adapt to your project
  • Discover and learn context
  • Make intelligent decisions
  • Generate contextual solutions

How Dynamic Prompts Work

Each Orchestre command is a sophisticated prompt that:

  1. Discovers your project's structure and patterns
  2. Analyzes requirements in context
  3. Adapts its approach to fit your needs
  4. Executes with awareness of existing code
  5. Learns from the results

Multi-LLM Orchestration

Orchestre leverages different AI models for their strengths:

🧠 Gemini

Role: Analysis & Planning

Strengths:

  • Deep reasoning
  • Complex analysis
  • Strategic planning

Used in: /orchestrate, /analyze

🔍 GPT-4

Role: Review & Security

Strengths:

  • Code review
  • Security analysis
  • Best practices

Used in: /review, /security-audit

🤖 Claude

Role: Implementation

Strengths:

  • Code generation
  • Context awareness
  • Natural interaction

Used in: All execution tasks

Consensus Building

When multiple models review code, Orchestre builds consensus:

typescript
// Example consensus from /review command
{
  security: {
    gpt4: { score: 8, issues: ["Missing rate limiting"] },
    claude: { score: 9, issues: [] }
  },
  consensus: {
    score: 8.5,
    priority_issues: ["Missing rate limiting"],
    confidence: "high"
  }
}

The MCP Integration

What is MCP?

The Model Context Protocol (MCP) enables Claude Code to interact with external tools and services. Orchestre is built as an MCP server, providing:

  • Tools: Functions Claude can call
  • Context: Project understanding
  • State: Persistent memory
  • Integration: Seamless workflow

MCP Tools in Orchestre

typescript
// MCP tool definition example
{
  name: "analyze_project",
  description: "Analyze project requirements and complexity",
  input_schema: {
    type: "object",
    properties: {
      requirements: { type: "string" },
      context: { type: "object" }
    }
  }
}

Distributed Memory System

Traditional State Management ❌

Central State File → Single Point of Failure
                  → Version Control Conflicts
                  → Lost Context

Orchestre's Approach ✅

CLAUDE.md Files → Distributed Across Codebase
                → Git-Friendly
                → Context Where Needed
                → Natural Documentation

Memory Structure Example

my-project/
├── CLAUDE.md                    # Project overview
├── src/
│   ├── auth/
│   │   ├── CLAUDE.md           # Auth-specific context
│   │   └── login.ts
│   └── api/
│       ├── CLAUDE.md           # API patterns
│       └── routes.ts
└── .orchestre/
    └── memory-templates/        # Reusable patterns

Command Types

1. Core Commands

Direct project operations:

  • /create - Initialize projects
  • /orchestrate - Analyze and plan
  • /execute-task - Implement features
  • /review - Quality assurance

2. Meta Commands

Commands that create commands:

  • /compose-prompt - Build complex workflows
  • /extract-patterns - Learn from code
  • /interpret-state - Understand context

3. Template Commands

Template-specific operations:

  • MakerKit: /setup-stripe, /add-team-feature
  • Cloudflare: /add-worker-cron, /implement-queue
  • React Native: /add-offline-sync, /setup-push-notifications

Workflow Patterns

Sequential Development

Traditional step-by-step approach:

/orchestrate → /execute-task → /review → /deploy

Parallel Development

Distribute work across multiple tasks:

/setup-parallel → /distribute-tasks → /coordinate-parallel → /merge-work

Iterative Refinement

Continuous improvement cycle:

/execute-task → /review → /suggest-improvements → /execute-task

Adaptive Intelligence

Context Awareness

Orchestre commands adapt based on:

  • Project Structure: Monorepo vs single package
  • Technology Stack: React vs Vue, Node vs Deno
  • Conventions: Existing patterns in your code
  • Requirements: Specific needs you express

Example: Same Command, Different Results

bash
# In a Next.js project:
/execute-task "Add user authentication"
# → Creates pages/api/auth, uses NextAuth

# In a FastAPI project:
/execute-task "Add user authentication"  
# → Creates auth router, uses JWT + OAuth

# In a Rails project:
/execute-task "Add user authentication"
# → Uses Devise gem, follows Rails conventions

Best Practices

1. Start with Clear Requirements

markdown
# Good Requirements
- User stories with acceptance criteria
- Technical constraints clearly stated
- Success metrics defined

# Poor Requirements
- "Make it work"
- Vague descriptions
- No clear goals

2. Use Natural Language

bash
# ✅ Good
/execute-task "Add a way for users to reset their forgotten passwords via email"

# ❌ Overly Technical
/execute-task "Implement JWT-based password reset flow with Redis token storage"

3. Leverage Reviews Early

bash
# After each major feature
/review

# For specific concerns
/security-audit
/performance-check

4. Document Decisions

bash
# Record important context
/update-state "Decided to use server-side sessions for better security"

# Document complex features
/document-feature "Authentication system"

Understanding Output

Analysis Results

json
{
  "complexity": {
    "score": 7.5,
    "factors": {
      "authentication": 2.5,
      "data_modeling": 3.0,
      "integrations": 2.0
    }
  },
  "recommendations": [
    "Start with authentication",
    "Use established patterns",
    "Plan for scalability"
  ]
}

Plan Structure

yaml
Phase 1: Foundation (2-3 hours)
  - Set up project structure
  - Configure database
  - Create base models

Phase 2: Core Features (3-4 hours)
  - Implement authentication
  - Build main CRUD operations
  - Add validation

Phase 3: Polish (1-2 hours)
  - Add error handling
  - Optimize performance
  - Write tests

Next Steps

Now that you understand the core concepts:

Built with ❤️ for the AI Coding community, by Praney Behl