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 |
references/credibility_rules.md
View on GitHub# Credibility Assessment Rules Reference
> This document describes universal credibility rules. Region-specific keywords, marketing signals, and platform weights are loaded from `category_profile` at runtime. The patterns below are EXAMPLES — the actual patterns used depend on the user's locale.
## 1. Sponsored/Ad Content Identification
The following patterns lower credibility scores. Specific keywords vary by language and region (loaded from `category_profile.marketing_signals`).
### High-weight Negative Signals (each match: -20 to -30 points)
1. **Contains purchase links or promo codes**: Embedded e-commerce links, coupon codes, affiliate links in body text
2. **Opens with full brand name + model**: Real users rarely write the complete "Brand + Model + Product Line" in the first sentence
3. **Zero defects throughout**: Genuine long-term use reviews almost always mention at least one negative
4. **Heavy marketing language**: Excessive superlatives, buzzwords, hype terms (specific terms vary by language/region)
5. **Uses official promotional images**: Studio product photos instead of real-life photos
6. **Trending marketing buzzwords**: Region-specific viral marketing phrases
### Medium-weight Negative Signals (each match: -10 to -15 points)
1. **Overly polished formatting**: Clean paragraphs, neat subheadings, professional images — resembles commercial content
2. **Clustered posting times**: Multiple positive reviews for the same product appearing within 1-3 days
3. **Repetitive comment patterns**: Many comments with identical phrases (astroturfing indicators)
4. **Author history concentrated in one category**: 80%+ of the account's content is recommendations in the same category
5. **Content mirrors brand's official messaging**: Wording, feature order matches the official product page
### Low-weight Negative Signals (each match: -5 points)
1. **Title contains clickbait/review keywords**: Review, must-buy, guide, top picks (in local language)
2. **Call-to-action at the end**: "Follow for more recommendations" type prompts
3. **Cross-posted with identical content**: Same text posted on multiple platforms
## 2. Genuine User Feedback Signals
The following patterns increase credibility scores. Specific expressions vary by language (loaded from `category_profile.authenticity_signals`).
### High-weight Positive Signals (each match: +20 to +30 points)
1. **Clear usage duration**: "Used for 8 months", "Bought 2 years ago", "Came back to update after 6 months"
2. **Specific defect descriptions**: Concrete physical issues observed over time
3. **Before/after comparisons**: "Previously used Brand X, switched to this one and..."
4. **Real-life photos (not official)**: Cluttered background, natural lighting, signs of daily life
5. **After-sales experience**: Describes returns, customer service interactions, warranty claims
### Medium-weight Positive Signals (each match: +10 to +15 points)
1. **Casual, imperfect language**: Typos, filler words, incomplete sentences
2. **Responds to follow-up questions**: Answers specific usage questions in comments
3. **Diverse posting history**: Account posts about many different topics
4. **Contains purchase details**: Mentions specific price paid, where purchased, whether they waited for a sale
5. **Category-specific experience signals**: Loaded dynamically from `category_profile.category_positive_signals`
6. **Long-term change narrative**: Contains "at first... but later..." temporal progression
### Low-weight Positive Signals (each match: +5 points)
1. **Reasonable time gap between purchase and posting**
2. **Posted in niche forum or sub-community** (not category hot zone)
3. **Low follower count / ordinary user**
4. **Recent content (recency bonus)**: Content from the 90-day instant search window gets extra points
## 3. Platform Credibility Weights
Platform weights are defined in `category_profile.platform_relevance` and `category_profile.regional_platforms`. Below are REFERENCE values for common regions:
### China (zh-CN) Reference Weights
| Platform | Base Weight | Notes |
|----------|-------------|-------|
| V2EX | 0.9 | Tech community, low ad density |
| Chiphell | 0.9 | Hardcore tech forum, expert users |
| NGA | 0.85 | Vertical community, lower astroturfing |
| Douban | 0.8 | Cultural community, relatively authentic |
| Baidu Tieba (niche) | 0.75 | Niche sub-forums have real discussions |
| SMZDM (original) | 0.65 | Original long-form is decent, but has affiliate mechanisms |
| Zhihu (overall) | 0.55 | High-upvote answers often commercial |
| Zhihu (low-vote) | 0.7 | Low-vote but detailed answers tend to be genuine |
| Zhihu (high-vote) | 0.4 | High-vote answers are heavily commercialized |
| Xiaohongshu | 0.3 | Social commerce, high baseline ad probability |
### US/English (en-US) Reference Weights
| Platform | Base Weight | Notes |
|----------|-------------|-------|
| Reddit (niche subreddits) | 0.85 | Community moderation, lower ad density |
| Head-Fi / AVSForum | 0.9 | Enthusiast forums, expert users |
| Amazon (verified purchase) | 0.7 | Verified buyers, but incentivized reviews exist |
| YouTube (comments) | 0.5 | Mixed quality, some astroturfing |
| Wirecutter (comments) | 0.75 | Quality audience |
| Slickdeals | 0.65 | Deal-focused but real user feedback |
### Japan (ja-JP) Reference Weights
| Platform | Base Weight | Notes |
|----------|-------------|-------|
| Kakaku.com | 0.8 | Price comparison + reviews, relatively trusted |
| 5ch | 0.75 | Anonymous forum, raw opinions |
| Amazon.co.jp (verified) | 0.7 | Verified purchase reviews |
## 4. Scoring Calculation
### Layer 1: Regex Pre-filter
```
Regex Score = Base platform weight x 100 + positive signal score - negative signal score
```
### Layer 2: AI Semantic Deep Analysis (only for gray zone 30-85 scores)
```
AI Score = Tone authenticity (0-20) + Narrative logic (0-20) + Detail richness (0-20)
+ Emotional consistency (0-20) + Interest disclosure (0-20)
```
### Fusion Layer
```
With full text (≥200 chars): Final = Regex x 0.3 + AI x 0.7
With snippet only (<200 chars): Final = Regex x 0.6 + AI x 0.4
AI analysis failed/skipped: Final = Regex x 1.0 (fallback)
```
### Score Levels
Labels are read from `category_profile.report_labels`:
- 80+: High credibility — core reference
- 60-79: Medium credibility — supplementary reference
- 40-59: Low credibility — only when no better source available
- Below 40: Suspected ad/astroturfing — excluded
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: 323k | 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: 323k | 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: 323k | Updated | Claude CodeCodex | Verified | Install |