Skip to main content

​
Purpose and Responsibilities

  • Combine the original idea and answers to clarifying questions.
  • Produce structured refined_requirements: title, description, features, target_audience, technical_requirements.
  • Output must be ready for PlatformSelector and Execution.

​
Input/Output

​
Input

  • idea: string — From WorkflowContext (or fallback in request).
  • answers: string[] — User answers to clarifying_questions.
  • sessionId: string — For logging.

​
Output (IntentRefinement in lib/openai.ts)

{
  status: "success";
  stage: "intent_analysis";
  refined_requirements: {
    title: string;
    description: string;
    features: string[];
    target_audience: string;
    technical_requirements: string[];
  };
  next_stage: "platform_selection";
}

​
Prompt Engineering

  • System: Defines the “expert requirements analyst (IntentAnalyzer)” and the exact JSON shape. No extra fields.
  • User: Session ID: ${sessionId}\nOriginal Idea: ${idea}\nUser Answers: ${answers.join("\n")}
  • Options: response_format: { type: "json_object" }, temperature: 0.7. Model: DEFAULT_MODEL_TYPE.
Implementation: refineIntent(idea, answers, sessionId, modelType?) in lib/openai.ts.

​
Examples

​
Input (conceptual)

  • Idea: “A habit tracker with streaks and reminders.”
  • Answers: ["Web first, mobile later", "No social", "Push and in-app reminders"]

​
Output (refined_requirements)

{
  "status": "success",
  "stage": "intent_analysis",
  "refined_requirements": {
    "title": "Habit Tracker",
    "description": "Web-first habit tracker with streaks and push plus in-app reminders; mobile later.",
    "features": ["habit logging", "streaks", "push reminders", "in-app reminders"],
    "target_audience": "Individuals focusing on personal habits",
    "technical_requirements": ["Web (React/Next.js) first", "Push notifications", "Mobile in a later phase"]
  },
  "next_stage": "platform_selection"
}

​
Customization Options

  • Model: Pass modelType to refineIntent.
  • System prompt: Adjust in lib/openai.ts to enforce different structure (e.g. more fields, stricter technical_requirements).
  • Stricter parsing: Add a schema/validator on the parsed JSON before returning.