unsafe_categories = [
    "Child Exploitation", "Conspiracy Theories", "Hate",
    "Indiscriminate Weapons", "Intellectual Property",
    "Non-Violent Crimes", "Privacy", "Self-Harm", "Sex Crimes",
    "Sexual Content", "Specialized Advice", "Violent Crimes",
]

import json
import anthropic
 
client = anthropic.Anthropic()
 
def moderate_message(message, unsafe_categories):
    category_str = "\n".join(unsafe_categories)
    prompt = f"""
Determine whether the following message warrants moderation,
based on the unsafe categories outlined below.
 
Message:
<message>{message}</message>
 
Unsafe Categories:
<categories>
{category_str}
</categories>
 
Respond with ONLY a JSON object, using the format below:
{{
"violation": <Boolean field denoting whether the message should be moderated>,
"categories": [Comma-separated list of violated categories],
"explanation": [Optional. Only include if there is a violation.]
}}
Do not include markdown formatting or code fences in your response."""
 
    response = client.messages.create(
        model="claude-haiku-4-5-20251001",
        max_tokens=200,
        messages=[{"role": "user", "content": prompt}],
    )
    text = next(b for b in response.content if b.type == "text").text
    result = json.loads(text)
    violation = result["violation"]
    cats = result.get("categories", []) if violation else []
    reason = result.get("explanation") if violation else None
    return violation, cats, reason

def moderate_safely(message, unsafe_categories, retries=1):
    for _ in range(retries + 1):
        try:
            return moderate_message(message, unsafe_categories)
        except (json.JSONDecodeError, KeyError):
            continue
    # 判定できない投稿は通さず、人のレビュー待ちに回す
    return None, [], "needs_human_review"

Assign a risk level based on your confidence that the user's message
should be moderated based on the defined unsafe categories:
 
0 - No risk
1 - Low risk
2 - Medium risk
3 - High risk
 
Respond with ONLY a JSON object, using the format below:
{
  "risk_level": <Numerical field denoting the risk level>,
  "categories": [Comma-separated list of violated categories],
  "explanation": <Optional. Only include if risk level is greater than 0>
}

BLOCK_AT = 3
REVIEW_AT = 2
 
def decide(result):
    level = result["risk_level"]
    if level >= BLOCK_AT:
        return "block"
    if level >= REVIEW_AT:
        return "review"
    return "pass"

labeled = [(c, False) for c in allowed_user_comments] + \
          [(c, True) for c in disallowed_user_comments]
 
tp = fp = fn = 0
for text, expected in labeled:
    predicted, _, _ = moderate_message(text, unsafe_categories)
    tp += predicted and expected
    fp += predicted and not expected
    fn += (not predicted) and expected
 
precision = tp / (tp + fp) if tp + fp else 0.0
recall = tp / (tp + fn) if tp + fn else 0.0
print(f"precision={precision:.2f} recall={recall:.2f}")

unsafe_category_definitions = {
    "Hate": "Content that is hateful toward people on the basis of "
            "their protected characteristics ...",
    "Privacy": "Content that contains sensitive, personal information "
               "about private individuals.",
    "Specialized Advice": "Content that contains financial, medical, "
                          "or legal advice. Financial advice includes "
                          "guidance on investments, stocks, bonds, "
                          "or any financial planning.",
    # ... 残りのカテゴリも同様に定義を書く
}
 
category_str = "\n".join(
    f"{name}: {definition}"
    for name, definition in unsafe_category_definitions.items()
)

unsafe_category_definitions_ja = {
    "Privacy": "個人の氏名・住所・電話番号・勤務先など、"
               "私人を特定できる情報を含む投稿。",
    "Specialized Advice": "投資・医療・法律に関する具体的な助言を"
                          "含む投稿。一般的な感想や体験談は含まない。",
}

messages_str = "\n".join(
    f"<message id={i}>{m}</message>" for i, m in enumerate(messages)
)
# 出力形式は違反した投稿だけを返す配列
# {"violations": [{"id": <message id>, "categories": [...],
#                  "explanation": "..."}]}
response = client.messages.create(
    model="claude-haiku-4-5-20251001",
    max_tokens=2048,  # バッチ用に出力上限を増やす
    messages=[{"role": "user", "content": assessment_prompt}],
)