from agent import PromptToSignatureAgent, PromptToSignatureConfig

agent = PromptToSignatureAgent(PromptToSignatureConfig())
PROMPT = """**INPUT FORMAT**: Customer product reviews in natural language text, ranging from 50-1000 words. May include informal language, slang, misspellings, emojis, and mixed sentiments. Reviews may be from various e-commerce platforms (Amazon, eBay, specialized retail sites).

**TASK DESCRIPTION**: You are an advanced sentiment analysis and feature extraction system for e-commerce platforms. Your goal is to parse customer reviews and extract granular insights about both overall sentiment and feature-specific opinions to help businesses understand customer satisfaction at a detailed level.

**EXTRACTION REQUIREMENTS**:
- **Overall Sentiment**: Determine the general sentiment of the entire review using a 4-category classification (positive/negative/mixed/neutral). Mixed indicates both positive and negative sentiments present; neutral indicates factual statements without emotional valence.
- **Product Features Mentioned**: Identify which of these standard e-commerce features are discussed: product quality, value for money, delivery/shipping experience, customer service interactions, packaging quality, product description accuracy, ease of use, durability, aesthetics/appearance, size/fit accuracy
- **Feature-Specific Sentiment**: For each feature mentioned, assign a sentiment score (-2 very negative, -1 negative, 0 neutral, +1 positive, +2 very positive)
- **Star Rating Prediction**: Based on the text sentiment and feature analysis, predict what star rating (1-5 stars) the customer likely gave, with confidence score (0-100%)
- **Purchase Verification**: Detect phrases indicating verified purchase ("verified purchase", "bought this", "received this product", "purchased from") vs. uncertain authenticity
- **Competitor Comparison**: Identify if the review compares the product to competitors, extract competitor names/products mentioned, and determine if comparison is favorable or unfavorable
- **Specific Issues**: Extract concrete complaints (e.g., "broke after 2 weeks", "arrived damaged", "wrong color sent", "missing parts")
- **Specific Praise**: Extract concrete positive mentions (e.g., "excellent battery life", "fast shipping", "exceeded expectations")
- **Review Helpfulness Indicators**: Classify review as "likely helpful" or "not helpful" based on linguistic markers (specificity, length, balanced perspective, constructive criticism vs. pure emotion)

**CLASSIFICATION REQUIREMENTS**:
- **Review Quality Score**: Rate the review quality from 1-5 where:
  - 1 = Unhelpful (pure emotion, no details, spam-like)
  - 2 = Minimal info (very brief, vague)
  - 3 = Moderate (some useful details)
  - 4 = Good (specific, balanced, informative)
  - 5 = Excellent (detailed, comprehensive, fair assessment)

**OUTPUT FORMAT**: Return the following schema:
{
  "overall_sentiment": {
    "classification": "positive|negative|mixed|neutral",
    "confidence_score": "float (0-1)"
  },
  "predicted_star_rating": {
    "stars": "integer (1-5)",
    "confidence": "integer (0-100)"
  },
  "features_analyzed": [
    {
      "feature_name": "string",
      "mentioned": "boolean",
      "sentiment_score": "integer (-2 to +2)",
      "supporting_quote": "string (relevant excerpt from review)"
    }
  ],
  "purchase_verification": {
    "appears_verified": "boolean",
    "indicators": ["array of strings showing evidence"]
  },
  "competitor_comparison": {
    "present": "boolean",
    "competitors_mentioned": ["array of strings"],
    "comparison_favorability": "favorable|unfavorable|neutral"
  },
  "specific_issues": [
    {
      "issue": "string",
      "severity": "minor|moderate|major",
      "quote": "string"
    }
  ],
  "specific_praise": [
    {
      "praise_point": "string",
      "quote": "string"
    }
  ],
  "review_helpfulness": {
    "classification": "likely_helpful|not_helpful",
    "quality_score": "integer (1-5)",
    "reasoning": "string"
  }
}
"""
def main():
    agent.push_to_hub("fadeleke/prompt-to-signature", with_code=True)
    #result = agent(PROMPT)
    #code = agent.generate_code(result)
    #print(code)
    
if __name__ == "__main__":
    main()