LeoYeAI/agent-master-skills / skills/anti-seo-researcher
anti-seo-researcher
VerifiedAnti-SEO deep consumer research tool. When a user wants to buy a product or make a consumer decision, use this Skill. Automatically detects user language and adapts to regional platforms and search strategies. Works with or without web_search — gracefully degrades to built-in Bing scraping when web_search is unavailable.
| File | Kind | Size |
|---|---|---|
| README.md | markdown | 21.2 KB |
| SKILL.md | skill md | 23.7 KB |
| _meta.json | config | 290 B |
| examples/category_profile_gaming_chair.json | config | 5.6 KB |
| examples/category_profile_office_chair_en.json | config | 7.2 KB |
| examples/task_config_gaming_chair.json | config | 413 B |
| examples/task_config_office_chair_en.json | config | 448 B |
| references/SKILL_REFERENCE.md | markdown | 10.2 KB |
| references/credibility_rules.md | markdown | 6.5 KB |
| scripts/ai_credibility_analyzer.py | script | 18.1 KB |
| scripts/brand_scorer.py | script | 17.7 KB |
| scripts/conflict_resolver.py | script | 20.2 KB |
| scripts/credibility_scorer.py | script | 34.8 KB |
| scripts/deep_dive_search.py | script | 40.3 KB |
| scripts/generate_report.py | script | 24.7 KB |
| scripts/platform_search.py | script | 54.1 KB |
| scripts/web_search_fallback.py | script | 14.5 KB |
scripts/ai_credibility_analyzer.py
View on GitHub#!/usr/bin/env python3
# -*- coding: utf-8 -*-
"""
AI 语义分析可信度评分器(V2 新增模块)
基于大语言模型对帖子内容进行多维度语义分析,
弥补正则表达式规则引擎在复杂场景下的识别盲区。
核心能力:
- 5 维度评分:语气真实性、论述逻辑、细节丰富度、情感一致性、利益关联判断
- 伪缺点识别:能识别"唯一缺点是包装不好看"这类高级软文手法
- 灰色地带精判:只对正则分 30-85 的帖子调用 AI,避免浪费
- 本地缓存:相同 URL 不重复分析
用法:
# 作为模块被 credibility_scorer.py 调用
from ai_credibility_analyzer import AICredibilityAnalyzer
analyzer = AICredibilityAnalyzer()
result = analyzer.analyze_single(title, content, platform)
# 独立运行:分析单条内容
python ai_credibility_analyzer.py --title "标题" --content "正文内容" --platform "知乎"
# 批量分析
python ai_credibility_analyzer.py --input results.json --output ai_scored.json
"""
import argparse
import hashlib
import json
import os
import re
import sys
from datetime import datetime
# ============================================================
# AI 评分 Prompt 模板
# ============================================================
AI_SCORING_PROMPT = """你是一个专业的互联网内容可信度分析师。你的任务是判断以下帖子是否为真实用户反馈、还是软文/水军内容。
## 待分析帖子
标题:{title}
平台:{platform}
正文:
{content}
## 评估维度
请从以下 5 个维度进行评估,每个维度 0-20 分(总分 0-100):
### 1. 语气真实性(0-20分)
- 20分:语言自然、有个人情绪波动、口语化、有犹豫和不确定性
- 10分:语言较为正式但有个人色彩
- 0分:语言高度规范化、像模板写作、措辞精准如公关稿
### 2. 论述逻辑(0-20分)
- 20分:叙事自然流畅、有时间线索、有因果逻辑、信息密度适中
- 10分:有一定逻辑但部分段落像拼凑
- 0分:行文结构化、卖点逐一罗列、像产品说明书或测评模板
### 3. 细节丰富度(0-20分)
- 20分:包含具体使用场景、精确时间、独特个人经历、非标准化的细节
- 10分:有一些细节但偏泛化
- 0分:细节缺失或全是公开信息(参数、规格)
### 4. 情感一致性(0-20分)
- 20分:情感表达与描述内容匹配、有真实的纠结和权衡
- 10分:情感较为单一但不算虚假
- 0分:一边倒的情感表达、过度正面或过度负面、缺点部分敷衍
### 5. 利益关联判断(0-20分)
- 20分:无任何商业引导、不含链接/优惠码、无引流行为
- 10分:有轻微的推荐倾向但无明显利益关联
- 0分:包含购买链接、优惠码、引流话术、明显的商业合作
## 特别注意
- "伪缺点"识别:如果缺点描述的是无关痛痒的方面(如"包装一般"、"颜色不够多"、"价格小贵"),而所有核心功能都是满分评价,这可能是高级软文的伪缺点手法,应降低情感一致性得分。
- 如果帖子过短(不足50字),在论述逻辑和细节丰富度上应保守打分。
- 注意区分"真正有用的使用反馈"和"拼凑公开信息的伪原创"。
## 输出格式
严格按以下 JSON 格式输出,不要添加任何额外文字:
{{
"tone_authenticity": <0-20>,
"logic_coherence": <0-20>,
"detail_richness": <0-20>,
"emotional_consistency": <0-20>,
"interest_independence": <0-20>,
"total": <0-100>,
"verdict": "<真实反馈|疑似软文|难以判断>",
"reasoning": "<一句话解释判断理由>"
}}"""
# ============================================================
# 缓存管理
# ============================================================
class AnalysisCache:
"""本地 JSON 文件缓存,避免对相同内容重复调用 AI"""
def __init__(self, cache_dir=None):
if cache_dir is None:
# 默认缓存目录在脚本同级的 .cache 下
script_dir = os.path.dirname(os.path.abspath(__file__))
cache_dir = os.path.join(script_dir, ".cache")
self.cache_dir = cache_dir
self.cache_file = os.path.join(cache_dir, "ai_analysis_cache.json")
self._cache = self._load()
def _load(self):
"""加载缓存文件"""
if os.path.exists(self.cache_file):
try:
with open(self.cache_file, "r", encoding="utf-8") as f:
return json.load(f)
except (json.JSONDecodeError, IOError):
return {}
return {}
def _save(self):
"""保存缓存到文件"""
os.makedirs(self.cache_dir, exist_ok=True)
with open(self.cache_file, "w", encoding="utf-8") as f:
json.dump(self._cache, f, ensure_ascii=False, indent=2)
def _make_key(self, url, content):
"""生成缓存键(基于 URL 或内容哈希)"""
if url:
return f"url:{url}"
return f"hash:{hashlib.md5(content.encode()).hexdigest()}"
def get(self, url="", content=""):
"""获取缓存的分析结果"""
key = self._make_key(url, content)
return self._cache.get(key)
def set(self, result, url="", content=""):
"""设置缓存"""
key = self._make_key(url, content)
self._cache[key] = {
"result": result,
"cached_at": datetime.now().isoformat(),
}
self._save()
@property
def size(self):
return len(self._cache)
# ============================================================
# AI 分析核心
# ============================================================
def _get_default_result():
"""AI 分析失败时返回的中性默认结果"""
return {
"tone_authenticity": 10,
"logic_coherence": 10,
"detail_richness": 10,
"emotional_consistency": 10,
"interest_independence": 10,
"total": 50,
"verdict": "难以判断",
"reasoning": "AI分析响应解析失败,使用默认分数",
"ai_error": True,
}
SCORE_DIMENSIONS = [
"tone_authenticity",
"logic_coherence",
"detail_richness",
"emotional_consistency",
"interest_independence",
]
def parse_ai_response(response_text):
"""
解析 AI 模型的 JSON 响应
Args:
response_text: AI 返回的文本
Returns:
解析后的评分字典,或失败时的默认结果
"""
try:
# 尝试直接解析
result = json.loads(response_text.strip())
except json.JSONDecodeError:
# 尝试从文本中提取 JSON 块
json_match = re.search(r'\{[^{}]*"tone_authenticity"[^{}]*\}', response_text, re.DOTALL)
if json_match:
try:
result = json.loads(json_match.group())
except json.JSONDecodeError:
return _get_default_result()
else:
return _get_default_result()
# 校验并修正分数范围
for key in SCORE_DIMENSIONS:
if key not in result:
result[key] = 10 # 缺失维度给中性分
else:
result[key] = max(0, min(20, int(result[key])))
# 重新计算总分(确保一致性)
result["total"] = sum(result[k] for k in SCORE_DIMENSIONS)
# 确保 verdict 字段存在
if "verdict" not in result:
if result["total"] >= 70:
result["verdict"] = "真实反馈"
elif result["total"] <= 35:
result["verdict"] = "疑似软文"
else:
result["verdict"] = "难以判断"
# 确保 reasoning 字段存在
if "reasoning" not in result:
result["reasoning"] = ""
result["ai_error"] = False
return result
class AICredibilityAnalyzer:
"""
AI 语义分析可信度评分器
使用大语言模型从语气真实性、论述逻辑、细节丰富度、
情感一致性、利益关联判断五个维度评估内容可信度。
"""
def __init__(self, max_batch=20, cache_dir=None):
"""
Args:
max_batch: 单次调查最多分析的帖子数量(成本控制)
cache_dir: 缓存目录路径
"""
self.max_batch = max_batch
self.cache = AnalysisCache(cache_dir)
self._analysis_count = 0 # 本次会话的分析计数
def analyze_single(self, title, content, platform, url=""):
"""
分析单条帖子的可信度
Args:
title: 帖子标题
content: 帖子正文
platform: 发布平台名称
url: 帖子 URL(用于缓存)
Returns:
评分结果字典,包含五个维度分数、总分、判定和理由
"""
# 检查缓存
cached = self.cache.get(url=url, content=content)
if cached:
print(f"[AI] 命中缓存: {url[:60] if url else title[:30]}", file=sys.stderr)
return cached["result"]
# 检查批量上限
if self._analysis_count >= self.max_batch:
print(f"[AI] 已达批量上限 ({self.max_batch}),跳过分析", file=sys.stderr)
return _get_default_result()
# 构建 Prompt
# 截断过长内容以控制 token 消耗
truncated_content = content[:3000] if content else ""
prompt = AI_SCORING_PROMPT.format(
title=title or "无标题",
content=truncated_content or "无正文内容",
platform=platform or "未知平台",
)
# 调用 AI 模型
# 注意:此处的 AI 调用由宿主环境(SKILL.md 工作流中的 AI)代理执行
# 在独立运行模式下,通过 stdin/stdout 模拟
print(f"[AI] 分析中: {title[:50]}...", file=sys.stderr)
self._analysis_count += 1
# 返回 prompt 供宿主 AI 执行
# 实际调用由 credibility_scorer.py 的融合层发起
return {
"_prompt": prompt,
"_needs_ai_call": True,
"_url": url,
"_title": title,
}
def process_ai_response(self, response_text, url="", content=""):
"""
处理 AI 模型返回的响应文本
在宿主 AI 执行了 prompt 后,将响应传回此方法进行解析和缓存。
Args:
response_text: AI 模型返回的文本
url: 原始帖子 URL(用于缓存键)
content: 原始内容(用于缓存键备用)
Returns:
解析后的评分结果字典
"""
result = parse_ai_response(response_text)
# 写入缓存(非错误结果才缓存)
if not result.get("ai_error"):
self.cache.set(result, url=url, content=content)
return result
def analyze_batch(self, items):
"""
批量分析多条帖子
Args:
items: 帖子列表,每条应包含 title, content/snippet, platform_name, url 字段
Returns:
分析结果列表(与输入一一对应)
"""
results = []
batch = items[:self.max_batch]
for i, item in enumerate(batch):
content = item.get("content", item.get("snippet", ""))
result = self.analyze_single(
title=item.get("title", ""),
content=content,
platform=item.get("platform_name", "未知"),
url=item.get("url", ""),
)
results.append(result)
print(f"[AI] 批量进度: {i+1}/{len(batch)}", file=sys.stderr)
return results
def should_analyze(self, regex_score):
"""
判断是否需要 AI 深度分析(灰色地带过滤)
不是所有帖子都需要过 AI:
- 正则分 < 30:明显的软文,不需要 AI 确认
- 正则分 > 85:明显的真实反馈,不需要 AI 确认
- 30-85 分:灰色地带,需要 AI 深度判别
Args:
regex_score: 正则评分引擎给出的分数
Returns:
True 表示需要 AI 分析,False 表示不需要
"""
return 30 <= regex_score <= 85
def should_use_advanced_model(self, regex_score):
"""
判断是否使用高级模型(对高价值灰色帖子)
对正则分 40-75 的"核心灰色地带"使用更精准的高级模型。
Args:
regex_score: 正则评分分数
Returns:
True 表示应使用高级模型
"""
return 40 <= regex_score <= 75
@property
def analysis_count(self):
"""本次会话已分析的帖子数"""
return self._analysis_count
@property
def cache_size(self):
"""缓存中的条目数"""
return self.cache.size
def get_stats(self):
"""获取分析统计信息"""
return {
"session_analysis_count": self._analysis_count,
"max_batch": self.max_batch,
"cache_size": self.cache.size,
"remaining_quota": max(0, self.max_batch - self._analysis_count),
}
# ============================================================
# 融合评分辅助函数
# ============================================================
def compute_fusion_score(regex_score, ai_score, content_length=0):
"""
计算正则分数与 AI 分数的融合分
Args:
regex_score: 正则评分引擎分数 (0-100)
ai_score: AI 语义分析分数 (0-100)
content_length: 分析内容的长度
Returns:
(final_score, alpha, beta) 元组
alpha: 正则权重
beta: AI 权重
"""
if ai_score is None:
# AI 未参与评分,回退到纯正则
return regex_score, 1.0, 0.0
if content_length >= 200:
# 全文场景:更信任 AI(AI 能理解语境)
alpha, beta = 0.3, 0.7
else:
# 摘要场景:更信任正则(AI 信息不足)
alpha, beta = 0.6, 0.4
final = regex_score * alpha + ai_score * beta
final = max(0, min(100, round(final, 1)))
return final, alpha, beta
def classify_credibility_level(score):
"""
根据分数判定可信度等级
Args:
score: 最终可信度分数
Returns:
等级字符串
"""
if score >= 80:
return "高可信度"
elif score >= 60:
return "中可信度"
elif score >= 40:
return "低可信度"
else:
return "疑似软文"
# ============================================================
# 主程序(独立运行模式)
# ============================================================
def main():
parser = argparse.ArgumentParser(
description="AI 语义分析可信度评分器",
formatter_class=argparse.RawDescriptionHelpFormatter,
epilog="""
示例:
# 分析单条内容
python ai_credibility_analyzer.py --title "用了半年的真实感受" --content "买了半年了..." --platform "知乎"
# 批量分析(输入为 credibility_scorer.py 输出的 JSON)
python ai_credibility_analyzer.py --input scored_results.json --output ai_scored.json
# 查看缓存统计
python ai_credibility_analyzer.py --cache-stats
""",
)
parser.add_argument("--title", default=None, help="帖子标题(单条分析模式)")
parser.add_argument("--content", default=None, help="帖子正文(单条分析模式)")
parser.add_argument("--platform", default="未知", help="发布平台")
parser.add_argument("--input", "-i", default=None, help="批量分析输入文件(JSON)")
parser.add_argument("--output", "-o", default=None, help="输出文件路径")
parser.add_argument("--max-batch", type=int, default=20, help="最大分析条数(默认20)")
parser.add_argument("--cache-stats", action="store_true", help="显示缓存统计信息")
parser.add_argument(
"--gray-zone-only",
action="store_true",
help="只分析正则分 30-85 的灰色地带帖子",
)
args = parser.parse_args()
analyzer = AICredibilityAnalyzer(max_batch=args.max_batch)
if args.cache_stats:
stats = analyzer.get_stats()
print(json.dumps(stats, ensure_ascii=False, indent=2))
return
if args.title and args.content:
# 单条分析模式
result = analyzer.analyze_single(
title=args.title,
content=args.content,
platform=args.platform,
)
print(json.dumps(result, ensure_ascii=False, indent=2))
elif args.input:
# 批量分析模式
with open(args.input, "r", encoding="utf-8") as f:
data = json.load(f)
results_list = data.get("results", [])
if args.gray_zone_only:
# 只分析灰色地带
to_analyze = [
r for r in results_list
if analyzer.should_analyze(r.get("credibility_score", 50))
]
print(f"[INFO] 灰色地带过滤: {len(results_list)} -> {len(to_analyze)} 条", file=sys.stderr)
else:
to_analyze = results_list
# 执行批量分析
ai_results = analyzer.analyze_batch(to_analyze)
# 输出
output_data = {
"analysis_time": datetime.now().isoformat(),
"stats": analyzer.get_stats(),
"results": ai_results,
}
json_str = json.dumps(output_data, ensure_ascii=False, indent=2)
if args.output:
with open(args.output, "w", encoding="utf-8") as f:
f.write(json_str)
print(f"[DONE] 结果已保存到: {args.output}", file=sys.stderr)
else:
print(json_str)
else:
parser.print_help()
if __name__ == "__main__":
main()
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis. | Stars: 472k | Updated | Claude CodeCodex | Verified | Install |
| diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness. | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependencies | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is available | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |