Skip to main content
Each agent worker is a specialized Cloudflare Worker that generates artifacts for a specific phase of software architecture.

​
Agent Structure

All agents follow a consistent structure: 
{agent-name}/
├── src/
│   ├── index.ts           # Main worker handler
│   └── systemPrompts.ts   # Agent system prompt
├── package.json
└── wrangler.toml

​
Agent Interface

All agents implement the same interface: 
export default {
  async fetch(request: Request, env: Env): Promise<Response> {
    // 1. Parse request body
    // 2. Extract context and user input
    // 3. Call LLM with system prompt
    // 4. Format response
    // 5. Return artifact
  }
};

​
System Prompts

Each agent has a specialized system prompt defining:
  • Role: Agent’s expertise (e.g., “Senior Product Manager”)
  • Goal: Primary objective
  • Instructions: Step-by-step process
  • Output Structure: Required format
  • Constraints: Rules and limitations

​
PRD Agent Prompt

Focuses on:
  • Executive summary
  • User personas
  • Functional requirements
  • MoSCoW roadmap

​
Data Agent Prompt

Focuses on:
  • Entity identification
  • Relationship modeling
  • Database schema design
  • Migration strategies

​
Design Agent Prompt

Focuses on:
  • Design tokens
  • Component library
  • UI/UX patterns
  • Accessibility

​
Logic Agent Prompt

Focuses on:
  • Business rules
  • State machines
  • Validation logic
  • Triggers HITL interrupt

​
API Agent Prompt

Focuses on:
  • RESTful API design
  • Resource modeling
  • Request/response schemas
  • OpenAPI specification

​
Frontend Agent Prompt

Focuses on:
  • Component architecture
  • State management
  • Routing structure
  • Performance optimization

​
Deployment Agent Prompt

Focuses on:
  • Infrastructure as Code
  • CI/CD pipelines
  • Monitoring and observability
  • Security configuration

​
Request Format

Agents receive requests with: 
{
  system_state: {
    current_phase: string;
    project_id: string;
  };
  context: string;        // Previous artifacts
  user_input: string;     // Product seed or phase input
}

​
Response Format

Agents return responses in: 
{
  system_state: {
    current_phase: string;
    status: 'complete' | 'interrupted';
    interrupt_signal?: boolean;
  };
  artifact: {
    type: 'markdown' | 'code';
    content: string;
  };
  trace: {
    agent: string;
    timestamp: string;
  };
}

​
LLM Integration

Agents use Gemini API for LLM calls: 
import { GoogleGenerativeAI } from '@google/generative-ai';

const genAI = new GoogleGenerativeAI(env.GEMINI_API_KEY);
const model = genAI.getGenerativeModel({ model: 'gemini-pro' });

const result = await model.generateContent(prompt);
const response = result.response.text();

​
Error Handling

Agents include error handling: 
try {
  // Agent execution
} catch (error) {
  return new Response(JSON.stringify({
    error: error.message,
    system_state: { status: 'error' }
  }), { status: 500 });
}

​
Service Bindings

Agents are called via Cloudflare service bindings: 
// In supervisor wrangler.toml
[[services]]
binding = "PRD_AGENT"
service = "parti-prd-agent"