Cross-Context Review: Production과 Review 세션을 분리해서 LLM 출력 품질 높이기

# CCR 리뷰 프롬프트 템플릿 (논문 Appendix A 기반)
CCR_REVIEW_PROMPT = """
Review the following {artifact_type} from a fresh perspective:

1. Factual accuracy: Are numbers, names, dates, and technical claims correct?
2. Internal consistency: Are there contradictions or terminology mismatches?
3. Contextual fitness: Would this work correctly in its intended environment?
4. Audience perspective: Could the target reader misinterpret any part?
5. Completeness: Is anything important missing?

For each issue found, provide:
- Location (line number or section)
- Description of the error
- Type (FACT/CONS/CTXT/RCVR/MISS)
- Severity (Critical/Major/Minor)
- Suggested fix

--- ARTIFACT START ---
{artifact_content}
--- ARTIFACT END ---
"""

# CCR 적용 예시 (OpenAI API)
from openai import OpenAI

def cross_context_review(artifact_content: str, artifact_type: str = "code") -> str:
    """
    새 클라이언트 인스턴스 = 새 컨텍스트.
    프로덕션 대화 히스토리를 절대 messages에 포함시키지 말 것.
    """
    client = OpenAI()  # 새 세션 — 이전 대화 히스토리 없음
    
    prompt = CCR_REVIEW_PROMPT.format(
        artifact_type=artifact_type,
        artifact_content=artifact_content
    )
    
    response = client.chat.completions.create(
        model="gpt-4o",
        messages=[
            # 중요: 프로덕션 세션 메시지 히스토리 포함 금지!
            {"role": "user", "content": prompt}
        ]
    )
    return response.choices[0].message.content

# 사용법
generated_code = """
def get_business_days(start, end):
    count = 0
    for d in range((end - start).days):
        if (start + timedelta(d)).weekday() >= 6:  # 버그: 5여야 함
            continue
        count += 1
    return count
"""

review_result = cross_context_review(generated_code, artifact_type="Python function")
print(review_result)