CHT-Agent Repository

System Architecture Overview

CLI Interface
    ↓
Master Supervisor (Orchestrator)
├── Research Supervisor
│   ├── Documentation Search Agent
│   └── Context Analysis Agent
│   
├── [HUMAN VALIDATION CHECKPOINT #1]
│   • Validate research findings
│   • Approve orchestration plan
│   
├── Development Supervisor
│   ├── Code Generation Agent
│   └── Test Environment Agent
│   
├── QA Supervisor
│   ├── Code Validation Agent
│   └── Test Orchestration Agent
│   
└── [HUMAN VALIDATION CHECKPOINT #2]
    • Review generated code
    • Verify test results
    • Approve for completion

Core Design Principles

1. Run with any model: Easily changeable to any model
2. Tool Orchestration Over Reinvention: Agents act as intelligent coordinators of existing CHT tools (cht-conf, cht-toolbox, cht-datasource, cht-docs, npm scripts) rather than implementing custom validation
3. Context-Based Learning: Lightweight file-based memory system that mirrors CHT repository structure
4. Incremental Value Delivery: Each agent provides immediate value while building knowledge over time
5. Developer-Friendly Integration: Seamless integration with existing npm scripts and development workflows

For WIP, you can find more design details here

Current Status: Research Supervisor POC

The Research Supervisor and its underlying agents have been implemented:

✅ Implemented Components

• Documentation Search Agent: Searches CHT documentation (currently mocked, will integrate with Kapa.AI MCP server)
• Context Analysis Agent: Analyzes past implementations and identifies reusable patterns
• Research Supervisor: Orchestrates the research phase using LangGraph workflow

📁 Project Structure

cht-agent/
├── src/
│   ├── agents/
│   │   ├── documentation-search-agent.ts    # Doc search with mocked MCP
│   │   └── context-analysis-agent.ts        # Context analysis
│   ├── supervisors/
│   │   └── research-supervisor.ts           # LangGraph orchestration
│   ├── types/
│   │   └── index.ts                         # TypeScript interfaces
│   ├── utils/
│   │   └── context-loader.ts                # Context file utilities
│   └── examples/
│       └── research-supervisor-example.ts   # Demo script
├── agent-memory/                            # Context storage (auto-created)
│   ├── domains/                             # Domain contexts
│   ├── workflows/                           # Workflow contexts
│   ├── knowledge-base/                      # Resolved issues
│   └── indices/                             # Lookup tables
└── tickets/                                 # Issue tickets to process

Getting Started

Prerequisites

• Node.js >= 20.0.0
• Anthropic API key

Installation

# Install dependencies
npm install

# Create .env file with your API key
cp .env.example .env
# Edit .env and add your ANTHROPIC_API_KEY

# Build the project
npm run build

Running the Example

# Run the Research Supervisor demo
npm run example:research

This will demonstrate the complete research workflow:

1. Documentation Search (using mocked Kapa.AI responses)
2. Context Analysis (analyzing past implementations)
3. Orchestration Plan Generation (creating implementation plan)

Example Output

The example processes a sample issue and outputs:

• Documentation references found
• Similar past implementations
• Reusable patterns identified
• Implementation phases
• Risk factors
• Effort estimates

Development

# Watch mode for development
npm run dev

# Lint code
npm run lint

# Build
npm run build

Next Steps

1. MCP Integration: Connect Documentation Search Agent to Kapa.AI MCP server
2. Context Population: Populate agent-memory with actual CHT domain contexts
3. Development Supervisor: Implement code generation and test environment agents
4. QA Supervisor: Implement validation and test orchestration agents
5. CLI Interface: Build command-line interface for issue processing