src/refinement/commenter.py

# 伪代码，需接入实际 API (如 OpenAI GPT-4o)
class Commenter:
    def __init__(self, client):
        self.client = client

    def review_slide(self, image_path):
        """

        调用 VLM 分析渲染后的 PPT 图片

        """
        system_prompt = """

        You are a Presentation Design Expert. Analyze the slide screenshot for visual layout issues.

        Focus strictly on:

        1. Text Overflow: Is text going out of its container or the slide boundary?

        2. Balance: Is one side too heavy or too empty?

        3. Spacing: Are images too small or text too crowded?

        

        Output valid JSON only:

        {

            "status": "PASS" or "NEEDS_REVISION",

            "issues": ["list of specific issues..."],

            "suggestion": "Natural language suggestion for the reviser."

        }

        """
        
        # 这里模拟 VLM 调用
        # response = client.chat.completions.create(model="gpt-4o", messages=..., image=...)
        # return json.loads(response)
        pass