Skip to content

LeoYeAI/agent-master-skills / skills/anti-seo-researcher

anti-seo-researcher

Verified

Anti-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.

# Anti-SEO Researcher

[简体中文](#简体中文)

**Anti-SEO, Anti-Astroturfing Deep Consumer Research Tool**

> In an internet where search results are heavily polluted by SEO manipulation and astroturfing, this tool helps you find genuine consumer feedback and make informed purchase decisions — in any language, for any market.

## Quick Start

### Install

Copy the entire directory into your AI coding assistant's Skill folder:

```bash
# CodeBuddy
cp -r anti-seo-researcher/ your-project/.codebuddy/skills/

# Cursor
cp -r anti-seo-researcher/ your-project/.cursor/skills/
```

Or clone via Git:

```bash
git clone https://github.com/AmosHc/ai-skill-anti-seo-researcher.git
```

Requirements: Python >= 3.6 (standard library only, no pip install needed), and an AI coding assistant that supports Skills (must provide `web_fetch` tool; `web_search` is optional — see Fallback Mode below).

### Usage

After installation, simply describe your purchase needs in natural language. The tool automatically detects your language and adapts to the appropriate regional platforms and search strategies.

**English:**
```
Research the best ergonomic office chairs under $500, I work from home and sit 8+ hours daily

Looking for baby formula for my 6-month-old, budget around $30-40, mainly concerned about safety and digestion

Best noise-cancelling headphones under $300 for commuting and office use
```

**中文:**
```
帮我调研一下3000元以内的人体工学椅,主要打游戏用,每天坐6小时以上

想给宝宝买奶粉,预算300-350一罐,6个月大,主要关注安全性和消化吸收

预算5000左右的手机,主要拍照和日常使用,不玩游戏
```

**日本語:**
```
予算5万円以内のゲーミングチェアを調べてください。毎日6時間以上座ります

赤ちゃん用のミルクを探しています。予算は2000-3000円、安全性重視
```

The tool automatically handles the entire pipeline — from searching, filtering, scoring, to generating a final research report in your language with recommend/avoid conclusions, scoring tables, real user quotes, and data sources.

---

## What Problem Does It Solve

When you want to buy a product, search engines give you mostly sponsored content: paid Q&A answers, influencer collaborations, affiliate review blogs, and brand-sponsored "reviews". The genuinely valuable consumer feedback is hidden in e-commerce follow-up reviews (how a product holds up after 6 months of real use), comment sections under promotional posts ("I bought it too, and it was a disaster"), niche enthusiast forums with detailed complaints, and rebuttals in Q&A comment threads.

This tool **automatically searches real user feedback from across the internet in your market, uses algorithms to filter out ads and astroturfing, scores candidate products on multiple dimensions, and generates a research report with recommend/avoid conclusions.**

## Multi-language & Multi-region Support

The tool adapts to any language and region automatically. When you ask a question, the AI detects your language and generates all region-specific configurations — platforms to search, search query templates, marketing signal patterns, and report labels — tailored to your market.

| Region | E-commerce Platforms | Social/Forums | Regulatory Bodies |
|--------|---------------------|---------------|-------------------|
| China | JD.com, Taobao, Pinduoduo | Zhihu, Xiaohongshu, V2EX, Chiphell, NGA | SAMR, CFDA |
| US/UK | Amazon, Best Buy, Walmart | Reddit, YouTube, Head-Fi, Wirecutter | FDA, CPSC, FTC |
| Japan | Amazon.co.jp, Rakuten, Kakaku.com | 5ch, Kakaku forums, Twitter/X | MHLW, CAA |
| Korea | Coupang, Naver Shopping | Naver Cafe, DC Inside, Clien | MFDS, KCA |
| Europe | Amazon (regional), Trustpilot | Reddit, regional forums | EFSA, national agencies |

The regional platform list, search templates, and marketing signal dictionaries are ALL dynamically generated by the AI based on your language — not hardcoded. This means the tool works for any region, including ones not listed above.

## Core Capabilities

### Three-tier Data Source Search

| Tier | Source | Value |
|------|--------|-------|
| E-commerce Reviews | Real buyer reviews from regional e-commerce platforms (indirect search) | Buyers who spent real money, long-term follow-up reviews |
| Comment Sections | Comment sections on social platforms (indirect search) | Real "debunking" feedback under promotional content |
| Forum Posts | Regional community forums and enthusiast sites | In-depth experience posts and comparisons |

E-commerce reviews and social comment sections are dynamically loaded content that search engines cannot directly index. The tool uses **indirect search strategies** (searching for "review compilations", "negative review roundups", "follow-up review summaries") to access this high-value data.

### Dual-layer Credibility Filtering

The first layer is a regex rule engine with 100+ rules identifying marketing signals (promotional language, zero-defect descriptions, affiliate links, incentivized reviews) and authenticity signals (usage duration, defect descriptions, after-sales experience). The second layer is AI semantic analysis for content in the gray zone, evaluating 5 dimensions: tone authenticity, narrative logic, detail richness, emotional consistency, and interest disclosure. This dual-layer approach maintains filtering precision while saving approximately 40-50% of AI call costs.

### Category Adaptation

Different product categories have completely different consumer decision factors. The tool automatically generates a category-specific configuration based on your needs, driving all differentiated behavior in subsequent steps. For example, baby formula focuses on ingredient safety and recall events with the strictest safety thresholds; phones focus on SoC performance, battery life, and thermals; office chairs focus on long-session comfort and mesh durability.

### Safety Event Screening

Every candidate brand undergoes a dedicated safety event search, covering recalls, fatalities, bans, lawsuits, regulatory penalties (using the appropriate regulatory bodies for your region), and category-specific safety risks. Products with serious safety issues are force-downgraded regardless of other dimension scores.

## Workflow

| Step | What It Does |
|------|-------------|
| Requirements Confirmation | Confirm category, budget, use case |
| Category Analysis | Auto-generate category + region config, driving all subsequent behavior |
| Three-tier Search | E-commerce reviews + comment sections + forum posts, covering regional platforms |
| Content Fetching | Fetch full text of valuable posts |
| Parameter Extraction | Extract structured product parameters from content |
| Dynamic Deep Dive | Negative long-tail search + e-commerce review deep dive for high-frequency models |
| Safety Search | Search safety events for candidate brands |
| Credibility Assessment | Regex pre-filter + AI deep analysis, filtering ads and astroturfing |
| Conflict Arbitration | Weighted resolution of contradictory reviews from different sources |
| Multi-dimensional Scoring | Score by category dimensions + safety capping |
| Report Generation | Output research report in user's language |

## Fallback Mode (No `web_search` Required)

Not all AI coding assistants have `web_search` available (it may require an API key you haven't configured). This skill **works without it** by automatically falling back to built-in Bing HTML scraping.

**How it works**: The skill includes `scripts/web_search_fallback.py`, which performs web searches by scraping DuckDuckGo (primary) and Bing (fallback) search results directly — zero API keys, zero configuration. When the AI detects that `web_search` is unavailable, it automatically uses this script via `execute_command` instead. Additionally, the internal search scripts (`platform_search.py`, `deep_dive_search.py`) have been upgraded with automatic DuckDuckGo fallback — if Bing returns a captcha, they seamlessly switch to DuckDuckGo.

**What you need**:
- `web_fetch` tool (for fetching specific page content) — this is usually always available
- `execute_command` tool (for running the fallback search scripts) — this is usually always available
- Python >= 3.6 with internet access

**What changes in fallback mode**: All search operations are routed through the Python scripts instead of the `web_search` tool. The rest of the workflow (credibility scoring, conflict resolution, brand scoring, report generation) works identically. The search quality is comparable since the scripts use the same Bing search engine that many `web_search` implementations use internally.

## Scoring System

### Credibility Scoring

Each search result receives a 0-100 credibility score. 85+ is highly credible (long-term feedback with specific defects), 60-84 is generally credible, 30-59 enters the gray zone triggering AI analysis, below 30 is suspected astroturfing and filtered out.

### Product Overall Scoring

| Score | Verdict |
|-------|---------|
| >= 70 | Recommend |
| 55-69 | Conditional Recommend |
| 40-54 | Caution |
| < 40 | Avoid |

### Safety Capping

Even if a product scores highly on other dimensions, safety issues force-cap the overall score. Food/baby products are capped at 30 (strictest), personal care at 25, electronics at 20, durable goods at 15.

## Project Structure

```
anti-seo-researcher/
├── SKILL.md                          # Skill definition (workflow, rules, schema)
├── references/
│   ├── SKILL_REFERENCE.md            # Detailed reference manual
│   └── credibility_rules.md          # Credibility assessment rules
├── scripts/
│   ├── platform_search.py            # Multi-platform search engine (3-tier data sources)
│   ├── web_search_fallback.py        # Fallback search (replaces web_search when unavailable)
│   ├── credibility_scorer.py         # Credibility scoring engine (regex + AI fusion)
│   ├── ai_credibility_analyzer.py    # AI semantic analysis scorer
│   ├── deep_dive_search.py           # Dynamic deep dive search (negative long-tail + safety)
│   ├── conflict_resolver.py          # Evidence conflict arbitrator
│   ├── brand_scorer.py               # Multi-dimensional scoring engine (safety capping)
│   └── generate_report.py            # Report generator
├── examples/
│   ├── task_config_gaming_chair.json          # Example: gaming chair (zh-CN)
│   ├── category_profile_gaming_chair.json     # Example: gaming chair profile (zh-CN)
│   ├── task_config_office_chair_en.json       # Example: office chair (en-US)
│   └── category_profile_office_chair_en.json  # Example: office chair profile (en-US)
├── README.md
├── LICENSE
└── .gitignore
```

## Contributing

Issues and Pull Requests are welcome. Especially welcome contributions in these areas: new category profile examples for different regions, platform search adapters for additional markets, credibility rule optimization for specific languages, and new region/language support.

## License

[MIT](LICENSE)

---

<a id="简体中文"></a>

# Anti-SEO Researcher

[English](#anti-seo-researcher)

**反 SEO、反软文深度消费调研工具**

> 在搜索结果被 SEO 操控和软文营销严重污染的互联网环境下,这个工具帮你找到真实的消费者反馈,做出明智的购买决策——支持任何语言、任何市场。

## 快速开始

### 安装

将整个目录复制到你的 AI 编程助手的 Skill 目录中:

```bash
# CodeBuddy
cp -r anti-seo-researcher/ your-project/.codebuddy/skills/

# Cursor
cp -r anti-seo-researcher/ your-project/.cursor/skills/
```

或者通过 Git 克隆:

```bash
git clone https://github.com/AmosHc/ai-skill-anti-seo-researcher.git
```

环境要求:Python >= 3.6(仅使用标准库,无需 pip 安装任何依赖),以及一个支持 Skills 的 AI 编程助手(必须提供 `web_fetch` 工具;`web_search` 为可选——见下方"降级模式")。

### 使用方式

安装完成后,直接用自然语言描述你的购买需求即可。工具会自动检测你的语言,并适配对应地区的平台和搜索策略。

**中文:**
```
帮我调研一下3000元以内的人体工学椅,主要打游戏用,每天坐6小时以上

想给宝宝买奶粉,预算300-350一罐,6个月大,主要关注安全性和消化吸收

预算5000左右的手机,主要拍照和日常使用,不玩游戏
```

**English:**
```
Research the best ergonomic office chairs under $500, I work from home and sit 8+ hours daily

Looking for baby formula for my 6-month-old, budget around $30-40, mainly concerned about safety and digestion

Best noise-cancelling headphones under $300 for commuting and office use
```

**日本語:**
```
予算5万円以内のゲーミングチェアを調べてください。毎日6時間以上座ります

赤ちゃん用のミルクを探しています。予算は2000-3000円、安全性重視
```

工具会自动完成整个流程——从搜索、过滤、评分到生成最终调研报告,报告以你使用的语言输出,包含推荐/避坑结论、评分表格、真实用户引用和数据来源。

---

## 解决什么问题

当你想买一个产品时,搜索引擎给你的大部分是广告内容:付费问答、网红合作、联盟营销测评博客、品牌赞助的"测评"。真正有价值的消费者反馈隐藏在电商追评(使用半年后产品真实表现)、推广帖评论区("我也买了,翻车了")、垂直论坛的详细吐槽帖、以及问答评论区的反驳中。

这个工具**自动搜索你所在市场的全网真实用户反馈,用算法过滤广告和软文,从多个维度为候选产品评分,并生成包含推荐/避坑结论的调研报告。**

## 多语言与多地区支持

工具自动适配任何语言和地区。当你提问时,AI 会检测你的语言并动态生成所有地区专属配置——要搜索的平台、搜索查询模板、营销信号词典和报告标签——全部针对你的市场定制。

| 地区 | 电商平台 | 社交/论坛 | 监管机构 |
|------|---------|----------|---------|
| 中国 | 京东、淘宝、拼多多 | 知乎、小红书、V2EX、Chiphell、NGA | 国家市场监管总局、国家药监局 |
| 美国/英国 | Amazon、Best Buy、Walmart | Reddit、YouTube、Head-Fi、Wirecutter | FDA、CPSC、FTC |
| 日本 | Amazon.co.jp、乐天、价格.com | 5ch、价格.com 论坛、Twitter/X | 厚生劳动省、消费者厅 |
| 韩国 | Coupang、Naver Shopping | Naver Cafe、DC Inside、Clien | 食药处、消费者院 |
| 欧洲 | Amazon(各国站)、Trustpilot | Reddit、各国论坛 | EFSA、各国机构 |

地区平台列表、搜索模板和营销信号词典全部由 AI 根据你的语言**动态生成**,而非硬编码。这意味着工具支持任何地区,包括上面没有列出的地区。

## 核心能力

### 三层数据源搜索

| 层级 | 数据源 | 价值 |
|------|-------|------|
| 电商评论 | 各地区电商平台的真实买家评论(间接搜索) | 花了真金白银的买家,长期追评反馈 |
| 评论区 | 社交平台的评论区内容(间接搜索) | 推广内容下的真实"打脸"反馈 |
| 论坛帖子 | 各地区社区论坛和爱好者网站 | 深度使用体验帖和对比分析 |

电商评论和社交评论区是动态加载内容,搜索引擎无法直接索引。工具使用**间接搜索策略**(搜索"评论汇总"、"差评合集"、"追评总结"等)来获取这些高价值数据。

### 双层可信度过滤

第一层是正则规则引擎,包含 100+ 条规则识别营销信号(推广用语、零缺陷描述、联盟链接、利益诱导评价)和真实性信号(使用时长、缺陷描述、售后经历)。第二层是 AI 语义分析,针对灰色地带内容从 5 个维度评估:语气真实性、叙事逻辑、细节丰富度、情感一致性、利益披露。双层方案在保持过滤精度的同时,节省约 40-50% 的 AI 调用成本。

### 品类自适应

不同产品品类有完全不同的消费决策因素。工具根据你的需求自动生成品类专属配置,驱动后续所有差异化行为。例如,婴儿奶粉聚焦成分安全性和召回事件,安全阈值最严格;手机聚焦 SoC 性能、续航和散热;办公椅聚焦久坐舒适度和网面耐久性。

### 安全事件筛查

每个候选品牌都会经过专门的安全事件搜索,覆盖召回、致死事件、禁令、诉讼、监管处罚(使用你所在地区对应的监管机构),以及品类特定的安全风险。存在严重安全问题的产品,无论其他维度评分多高,都会被强制降级。

## 工作流程

| 步骤 | 具体操作 |
|------|---------|
| 需求确认 | 确认品类、预算、使用场景 |
| 品类分析 | 自动生成品类 + 地区配置,驱动后续所有行为 |
| 三层搜索 | 电商评论 + 评论区 + 论坛帖子,覆盖地区平台 |
| 内容抓取 | 抓取有价值帖子的全文内容 |
| 参数提取 | 从内容中提取结构化产品参数 |
| 动态深挖 | 负面长尾搜索 + 高频型号的电商评论深度挖掘 |
| 安全搜索 | 搜索候选品牌的安全事件 |
| 可信度评估 | 正则预过滤 + AI 深度分析,过滤广告和软文 |
| 冲突仲裁 | 加权解决不同来源的矛盾评价 |
| 多维评分 | 按品类维度评分 + 安全封顶 |
| 报告生成 | 以用户语言输出调研报告 |

## 降级模式(无需 `web_search`)

并非所有 AI 编程助手都提供 `web_search` 工具(可能需要额外配置 API key)。本工具**无需 `web_search` 也能正常工作**,会自动降级为内置的 Bing 网页抓取方案。

**工作原理**:工具内置了 `scripts/web_search_fallback.py` 脚本,通过直接抓取 DuckDuckGo(主)和 Bing(备选)搜索结果页面来执行搜索——零 API key、零额外配置。当 AI 检测到 `web_search` 不可用时,会自动通过 `execute_command` 调用此脚本替代。此外,内部搜索脚本(`platform_search.py`、`deep_dive_search.py`)也已升级为自动 DuckDuckGo 降级——当 Bing 返回 captcha 时,会无缝切换到 DuckDuckGo。

**你需要的最低工具集**:
- `web_fetch` 工具(用于抓取特定页面内容)——通常都可用
- `execute_command` 工具(用于运行降级搜索脚本)——通常都可用
- Python >= 3.6 + 可访问互联网

**降级模式的变化**:所有搜索操作改为通过 Python 脚本执行,而非 `web_search` 工具。其余流程(可信度评分、冲突仲裁、品牌评分、报告生成)完全相同。搜索质量相当,因为脚本底层使用的也是 Bing 搜索引擎。

## 评分体系

### 可信度评分

每条搜索结果获得 0-100 的可信度评分。85 分以上为高度可信(带具体缺陷描述的长期反馈),60-84 分为一般可信,30-59 分进入灰色地带触发 AI 分析,30 分以下疑似软文直接过滤。

### 产品综合评分

| 分数 | 结论 |
|------|------|
| >= 70 | 推荐 |
| 55-69 | 有条件推荐 |
| 40-54 | 谨慎 |
| < 40 | 避坑 |

### 安全封顶机制

即使产品在其他维度评分很高,安全问题也会强制封顶综合评分。食品/婴儿用品封顶 30 分(最严格),个人护理品封顶 25 分,电子产品封顶 20 分,耐用品封顶 15 分。

## 项目结构

```
anti-seo-researcher/
├── SKILL.md                          # Skill 定义(工作流、规则、Schema)
├── references/
│   ├── SKILL_REFERENCE.md            # 详细参考手册
│   └── credibility_rules.md          # 可信度评估规则
├── scripts/
│   ├── platform_search.py            # 多平台搜索引擎(三层数据源)
│   ├── web_search_fallback.py        # 降级搜索脚本(web_search 不可用时替代)
│   ├── credibility_scorer.py         # 可信度评分引擎(正则 + AI 融合)
│   ├── ai_credibility_analyzer.py    # AI 语义分析评分器
│   ├── deep_dive_search.py           # 动态深度搜索(负面长尾 + 安全)
│   ├── conflict_resolver.py          # 证据冲突仲裁器
│   ├── brand_scorer.py               # 多维评分引擎(安全封顶)
│   └── generate_report.py            # 报告生成器
├── examples/
│   ├── task_config_gaming_chair.json          # 示例:电竞椅(中文)
│   ├── category_profile_gaming_chair.json     # 示例:电竞椅配置(中文)
│   ├── task_config_office_chair_en.json       # 示例:办公椅(英文)
│   └── category_profile_office_chair_en.json  # 示例:办公椅配置(英文)
├── README.md
├── LICENSE
└── .gitignore
```

## 贡献

欢迎提交 Issue 和 Pull Request。特别欢迎以下方向的贡献:新地区的品类配置示例、新市场的平台搜索适配器、特定语言的可信度规则优化、以及新地区/语言支持。

## 许可证

[MIT](LICENSE)
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: 472kUpdated Claude CodeCodexVerifiedInstall
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: 323kUpdated Claude CodeCodexVerifiedInstall
dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependenciesStars: 323kUpdated Claude CodeCodexVerifiedInstall
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 availableStars: 323kUpdated Claude CodeCodexVerifiedInstall