---
slug: "edgeknowledge-skill"
source_type: "readme"
source_url: "https://cdn.jsdelivr.net/gh/1596941391qq/edgeknowledge_skill@main/README.md"
repo: "https://github.com/1596941391qq/edgeknowledge_skill"
source_file: "README.md"
branch: "main"
---
[![GitHub stars](https://img.shields.io/github/stars/1596941391qq/EdgeKnowledge_Skill?style=social)](https://github.com/1596941391qq/EdgeKnowledge_Skill/stargazers) [![License: MIT](https://img.shields.io/badge/License-MIT-green.svg)](LICENSE) [![Python](https://img.shields.io/badge/Python-3.10%2B-blue.svg)](https://www.python.org/) [![Pipeline](https://img.shields.io/badge/Pipeline-3%20of%203-6e40c9.svg)](#pipeline)

# Edge Knowledge

> **[Chinese Documentation](https://github.com/1596941391qq/edgeknowledge_skill/blob/HEAD/README_CN.md)** | **English Documentation**

AI-powered edge knowledge mining system. Intelligently recommends high-value forums (Reddit, BlackHatWorld, GreyHatMafia, etc.) based on user needs, performs deep crawling with browser-use, visual recognition, and analyzes edge tactics, power users, and resources, outputting a single structured Markdown report.

## 🎯 Core Philosophy

> **💡 Knowledge Equity Initiative**
> This project encourages community contributions! You can create branches on Git to build domain-specific search expert skills (e.g., SEO Expert, Affiliate Hunter, Tool Digger), making edge knowledge accessible to more people.
>
> **🤝 How to Contribute PRs**:
> - 🌿 **Build Domain-Specific Branches** - Create specialized versions for specific domains (e.g., `seo-expert`, `affiliate`)
> - 🔧 **Optimize Existing Skill** - Improve tool versatility and robustness (anti-scraping strategies, error handling, performance optimization)
> - 📚 **Share Quality Forums** - Add deep-dive forums you use to `forum_database.json` with search strategies
> - 🤖 **Add Automation Capabilities** - Integrate scheduled tasks, push to Notion/Feishu/Slack and other collaboration tools
>
> Let's break information barriers and promote knowledge equity together! 🚀

### What is Edge Knowledge?

In the AI era, information that LLMs can easily understand or collect from the internet is considered common knowledge. Edge knowledge refers to scarce, industry know-how, and gray/black hat techniques that reach the boundaries of platform rules.

**Edge Knowledge Includes**:
- ✅ Black/White/Gray Hat SEO: PBN, Parasite SEO, Cloaking, Link Farms
- ✅ Tactical Tricks: Banned, risky, but effective methods
- ✅ Underground Tools: GSA SER, SEnuke, Xrumer, Scrapebox
- ✅ Real Data: Success rates, prices, risks, case numbers
- ✅ Controversial Tactics: High-voted replies, controversial discussions, practical sharing

**Does NOT Include**:
- ❌ White Hat Platitudes ("write good content and traffic will come")
- ❌ Official Documentation (Google SEO Guidelines)
- ❌ Mainstream Advice ("improve user experience")

### Three Characteristics of Edge Knowledge

1. **Freshness** - Recent information (post-2025)
2. **Scarcity** - Exclusive information (known in small circles)
3. **Credibility** - Trustworthy information (with data and cases)

## ✨ Core Capabilities

### V1: Foundation
- 🎯 **Intent Extraction & Query Generalization** - Generalizes query directions and recommends high-quality forums
- 🧠 **User Preference Memory** - Remembers favorite forums and search habits
- 🔐 **Credential Management** - Automatically manages forum credentials, supports session reuse
- 🛡️ **Anti-Detection System** - Fixed fingerprints, random delays, session management to reduce account risk
- 🌐 **Smart Browser Crawling** - Uses browser-use skill, supports screenshots + visual recognition
- 🔍 **Deep Content Analysis** - Edge knowledge identification, power user identification, resource extraction
- 📊 **Single Report Output** - Named by date + topic, structured display

### 🆕 V2: Hunter Mode (NEW!)

V2 adds aggressive resource acquisition capabilities:

| Feature | Description |
|---------|-------------|
| **🎯 Value Signal Detection** | 6 pattern types: Reply-unlock, Hidden content, Download links, Extract codes, Attachments, Task thresholds |
| **🤖 Auto-Reply System** | 25 random templates (15 EN + 10 CN), auto language detection |
| **📦 Resource Downloader** | Downloads ALL file types (.exe/.bat/.torrent), maintains completeness |
| **🔗 Deep Dive Tracking** | External link follow, Author tracking, Comment section mining |
| **🧵 Tool Integration** | Agent-Reach, gallery-dl, yt-dlp, Crawl4AI ready |

### 🆕 V3: Freshness & Credibility (NEW!)

V2 could grab resources. V3 makes sure what you grab is **fresh** and **trustworthy** — because edge knowledge decays, and an opinion is not a tactic.

| Feature | Description |
|---------|-------------|
| **🌐 GEO/AEO Edge Category** | The frontier moved. New `GEO_Edge_Search` category + `GEO_AI_Visibility` strategy target the freshest, least-saturated edge: getting cited by ChatGPT/Perplexity/Gemini, intercepting AI Overview traffic, engineering LLM visibility — before the playbooks go mainstream |
| **⏳ Freshness Contract** | Every source now carries `last_verified` / `activity_status` / `freshness_score`. Stale pools get down-ranked, declining ones flagged for re-verification (`reverify_after_days: 90`). "Freshness" stops being a slogan and becomes a mechanism |
| **🎯 Credibility Scoring** | `credibilityScoring` turns the *trustworthy* characteristic into a computable score. Concrete numbers, real pricing, and case evidence score high; pure opinion gets penalized. Extracted knowledge is ranked by evidence strength before it lands in a report |

**Why V3 matters:** A forum that was gold in 2024 may be dead in 2026. A reply with "$30 CPA, 14-day ROI, ran it on 3 accounts" is worth more than "I think this should work." V3 encodes both judgments so the system stops treating all signals as equal.


## 🚀 Quick Install

```bash
git clone https://github.com/1596941391qq/EdgeKnowledge_Skill.git
cd EdgeKnowledge_Skill
chmod +x install.sh
./install.sh
```

Or copy this skill to Claude Code's skills directory:

```bash
cp -r edge-knowledge ~/.claude/skills/
```


## 🔄 MCP Tool Routing (V2)

V2 includes an **intelligent three-tier routing engine** that automatically selects the best tool for each task — cost-first with fallback guarantees.

### Tool Tiers

| Tier | Tool | Cost | Best For |
|------|------|------|----------|
| **Tier 1** | `browser-use` | Free (local Playwright) | Screenshot + visual recognition, clicks/scrolls/forms, JS lazy-load, post-login access |
| **Tier 2** | `agent-browser` | Free (Vercel CLI) | Repetitive structured extraction, @e1/@e2 element selection, scripted multi-step operations |
| **Tier 3** | `google-gemini-mcp` | API key (per-token) | Bypassing anti-bot blocks, batch URL analysis (>10 pages), complex multimodal understanding |

### Routing Rules

```
IF captcha_detected AND captcha_type == "recaptcha_v2":
    → ai-captcha-bypass (GPT-4o or Gemini 2.5) → retry

IF cloudflare_blocked AND browser_use_failed:
    → google-gemini-mcp (Tier 3)

IF batch_analysis AND urls > 10:
    → google-gemini-mcp (concurrent)

IF visual_heavy AND needs_screenshot:
    → browser-use (Tier 1)

IF download_only:
    → gallery-dl / yt-dlp (bypasses browser)
```

### MCP Server Config

All MCP server configurations are in `mcp_config.json`:
- **`google-gemini-mcp`** — Gemini 2.5 depth search, URL fetch, multimodal analysis
- **`ai-captcha-bypass`** — GPT-4o / Gemini-driven CAPTCHA solving (Selenium + Firefox)



## ⚙️ Configuration Files

### 1. forum_database.json

The forum knowledge base containing forum information and search strategies.

**Structure**:
```json
{
  "categories": {
    "Q&A_Search": {
      "description": "Suitable for mining deep discussions and real user feedback in comment sections",
      "forums": [...]
    },
    "Edge_Knowledge_Search": {
      "description": "Suitable for mining gray/black hat techniques not found in mainstream channels",
      "forums": [...]
    },
    "Deep_Dive_Forums": {
      "description": "Deep content that others don't know about",
      "forums": [...]
    }
  },
  "search_strategies": {
    "Instagram_Growth": {
      "keywords": [...],
      "recommended_forums": [...],
      "focus": "Real feedback in comment sections and gray techniques"
    }
  }
}
```

**Usage**:
- The system automatically reads this file to recommend forums
- You can add new forums or search strategies
- Each forum includes: name, URL, rating, cost, target audience, tags

### 2. memory.json.template

Template for user preferences and crawling history. Copy to `memory.json` for first use:

```bash
cp memory.json.template memory.json
```

**Structure**:
```json
{
  "userPreferences": {
    "favoriteForums": ["BestBlackHatForum"],
    "domains": ["SEO", "Black Hat Techniques", "Traffic Arbitrage"],
    "lastUsedDomain": "Black Hat SEO"
  },
  "forumCredentials": {
    "bestblackhatforum.com": {
      "username": "",
      "password": "",
      "lastLogin": "",
      "loginCount": 1,
      "cookies": null,
      "localStorage": null,
      "sessionValid": true
    }
  },
  "crawledResources": [],
  "antiDetection": {
    "viewport": {"width": 1920, "height": 1080},
    "userAgent": "",
    "timezone": "",
    "locale": "",
    "randomDelayRange": [5000, 30000],
    "maxLoginPerDay": 3,
    "reuseSession": true,
    "sessionExpiryHours": 24
  }
}
```

**Fields Explanation**:
- `userPreferences`: Your favorite forums and domains
- `forumCredentials`: Forum login credentials (auto-saved when you provide them)
- `crawledResources`: History of crawled URLs (prevents duplicate crawling)
- `antiDetection`: Anti-detection configuration (viewport, delays, login limits)

**Privacy Note**: `memory.json` is in `.gitignore` and won't be committed to Git. Your credentials are safe locally.

### 3. value_patterns.json (V2)

Value signal detection patterns for Hunter Mode.

**Patterns Include**:
- `reply_unlock` - "Reply to unlock hidden content"
- `hidden_content` - "Hidden content / Spoiler blocks"
- `download_link` - Mega/Mediafire/Baidu pan links
- `extract_code` - "Password: xxx" / "Code: xxx"
- `attachment` - "Download attachment"
- `task_threshold` - "Need X posts / X likes to view"

### 4. platforms.json (V2)

Platform-specific download configurations.

**Structure**:
```json
{
  "mega.nz": {
    "tool": "gallery-dl",
    "args": ["--no-mtime"],
    "maxConcurrent": 2
  },
  "youtube.com": {
    "tool": "yt-dlp",
    "args": ["-f", "best"]
  }
}
```

### 5. resources/ Directory (V2)

Organized resource storage:
```
resources/
├── downloads/      # Downloaded files (by date)
├── links/
│   ├── mega.json   # Mega links index
│   ├── baidu.json  # Baidu pan links
│   └── gdrive.json # Google Drive links
├── codes/
│   └── passwords.json  # Extract codes
└── index.json      # Unified resource index
```

## 🚀 Usage

### Basic Usage

```
Use edge-knowledge to mine cost-effective Instagram growth services
```

### Real-World Examples

Demonstrate the skill's edge knowledge tracking capabilities:

#### Example 1: Black Hat SEO Techniques
```
Use edge-knowledge to find the latest black hat SEO techniques for 2026
```
**What you'll get**:
- Latest PBN (Private Blog Network) strategies
- Parasite SEO tactics that still work
- Cloaking techniques to bypass Google detection
- Link farming methods and automation tools

#### Example 2: Affiliate Marketing Arbitrage
```
Use edge-knowledge to discover profitable affiliate traffic sources
```
**What you'll get**:
- Underground traffic sources with high ROI
- CPA networks that accept gray hat methods
- Media buying strategies from top affiliates
- Real case studies with actual numbers

#### Example 3: Social Media Growth Hacks
```
Use edge-knowledge to find Instagram automation tools that bypass detection
```
**What you'll get**:
- Automation bots that work in 2026
- SMM panels with real engagement
- Growth hacking scripts and techniques
- Risk assessment and detection avoidance

#### Example 4: Tool Discovery
```
Use edge-knowledge to find cracked SEO tools and automation software
```
**What you'll get**:
- Working cracks for premium SEO tools
- Automation scripts for scraping and posting
- Nulled WordPress plugins and themes
- Community reviews and safety ratings

### Workflow

1. **Stage 1: Forum Recommendation** - System recommends relevant forums based on your needs
2. **Stage 2: Smart Crawling** - Deep crawls forum content using browser-use
3. **Stage 3: Content Analysis** - Identifies edge knowledge, power users, resources
4. **Stage 4: Report Generation** - Outputs structured Markdown report



## 🛠️ Bundled Tool Modules

The following modules are included for edge-case automation needs:

| Module | Purpose |
|--------|---------|
| `rules/access-control.md` | Access control & authorization boundary definition |
| `mcp_config.json` | Central MCP server & tool routing configuration |
| `cdp-scripts/` | CDP-based forum crawlers (see CDP Scripts section below) |

All tools are called automatically by the routing engine when triggered.

## CDP Scripts

Browser automation crawlers using Chrome DevTools Protocol for targeted forum mining:

| Script | Target | What it does |
|---|---|---|
| `bhw-crawler.mjs` | BlackHatWorld | Thread listing crawler |
| `bhw-crawler-v2.mjs` | BlackHatWorld | V2 with value signal detection |
| `bhw-detail-crawler.mjs` | BlackHatWorld | Deep thread content extraction |
| `bhw-backlink-scrape.mjs` | BlackHatWorld | Backlink opportunity scraper |
| `bhw-vendor-verify.mjs` | BlackHatWorld | Vendor/service verification |
| `cdp-batch-bbhf.mjs` | BestBlackHatForum | Batch forum crawler |
| `cdp-bbhf-threads.mjs` | BestBlackHatForum | Thread detail extractor |
| `cdp-buildersociety.mjs` | BuilderSociety | Forum crawler |
| `cdp-buildersociety-v2.mjs` | BuilderSociety | V2 enhanced extraction |
| `scrape-onehack.mjs` | OneHack | Content scraper |

## 📚 Supported Forums (Partial List)

| Rank | Forum | Rating | Cost | Target Audience |
|------|-------|--------|------|-----------------|
| 1 | GreyHatMafia | 9.5/10 | Free | Everyone |
| 4 | SEO Isn't Dead | 9/10 | Free | SEO Practitioners |
| 6 | BlackHatWorld | 8.5/10 | Free | General Marketing |
| 7 | BestBlackHatForum | 9.5/10 | Free | Recommended: slenderman's posts |
| 8 | Reddit | 10/10 | Free | Value in comment sections |

## 📋 Report Example

Generated reports contain three-layer analysis:

### 1. Edge Knowledge Identification
```markdown
### Edge Knowledge #1: [Knowledge Title]
**Compressed Expression**: [One-sentence summary]
**Easy Explanation**: [Detailed explanation]
**Viewpoints**: @username: "viewpoint content"
**Risk**: [Potential risks]
**Cost**: [Time/money/learning cost]
**Source Link**: [Original link]
```

### 2. Power User Identification
```markdown
### Power User #1: @username
**Username**: username (forum name)
**High-Energy Viewpoints**: "viewpoint1", "viewpoint2"
**Link**: [User profile link]
```

### 3. Resource Extraction
```markdown
### Resource #1: [Tool/Service Name]
**Name**: Tool name
**Link**: [Tool link]
**Description**: [Feature description]
**Price**: [Price information]
**Review**: [User review summary]
```

## 🔧 Technical Architecture

### Dependencies
- Claude Code CLI
- browser-use skill / agent-browser skill
- Python 3.8+

### Cross-Platform Installer
`install.sh` auto-detects **macOS**, **Ubuntu/Debian**, **CentOS/RHEL**, and **Arch Linux**, installs system packages, configures Playwright, and sets up MCP servers.

### V2 Additional Dependencies
```bash
pip install gallery-dl yt-dlp crawl4ai
```

### Data Flow
```
User Need → Read memory.json → Recommend Forums → User Confirms →
Check Credentials → Apply Anti-Detection Config → browser-use Crawl →
Claude Analysis → Generate Report → Update memory.json
```

### Reference Screenshots

![](https://cdn.nlark.com/yuque/0/2026/png/65208130/1770384664181-3508fcfa-012e-45b9-a3b3-43f343c4ad21.png)
![](https://cdn.nlark.com/yuque/0/2026/png/65208130/1770384830572-2db0ae2f-2b82-4077-bff1-8e04658d708a.png)
![](https://cdn.nlark.com/yuque/0/2026/png/65208130/1770384935540-1488fc21-351a-4515-ad6f-0431e20ba6ec.png)

## ⚠️ Important Notes

1. **Legal Use** - For educational and research purposes only, comply with forum rules and local laws
2. **Account Security** - Use dedicated accounts, avoid using your main accounts
3. **Anti-Detection** - System automatically applies anti-detection strategies, but use cautiously
4. **Content Risk** - Edge knowledge may contain risky operations, use your own judgment

## 📄 License

MIT License

## 👤 Author

黑咖啡和冰月亮 / Black Coffee & Ice Moon (@weihackings)

## 🤝 Contributing

Issues and Pull Requests are welcome!

## ⭐ Star History

[![Star History Chart](https://api.star-history.com/svg?repos=1596941391qq/EdgeKnowledge_Skill&type=Date)](https://star-history.com/#1596941391qq/EdgeKnowledge_Skill&Date)



## ❓ FAQ

### Q1: Is this tool legal to use?
**A**: This tool is for educational and research purposes only. Always comply with forum terms of service and local laws. Use at your own risk.

### Q2: Will my account get banned?
**A**: The tool includes anti-detection features (random delays, session reuse, login limits), but there's always a risk. We recommend:
- Use dedicated accounts, not your main accounts
- Respect the `maxLoginPerDay` limit (default: 3)
- Don't crawl too aggressively

### Q3: Do I need to provide forum credentials?
**A**: Only for forums that require login (like BestBlackHatForum). For public forums (like Reddit), no credentials needed. Your credentials are stored locally in `memory.json` and never committed to Git.

### Q4: How do I add my own forums?
**A**: Edit `forum_database.json` and add your forum to the appropriate category:
```json
{
  "name": "YourForum",
  "url": "https://yourforum.com",
  "rating": 9.0,
  "cost": "Free",
  "target_audience": "Your Target Audience",
  "tags": ["tag1", "tag2"]
}
```

### Q5: Can I create domain-specific branches?
**A**: Yes! Fork the repo and create a branch like `seo-expert` or `affiliate`. Customize `forum_database.json` and `skill.md` for your domain. See our [contribution guidelines](#-knowledge-equity-initiative).

### Q6: How do I integrate with Notion/Feishu?
**A**: This is a community contribution opportunity! You can:
1. Fork the repo
2. Add integration code to push reports to your collaboration tool
3. Submit a PR with your integration

### Q7: What's the difference between edge knowledge and common knowledge?
**A**:
- **Common Knowledge**: Information easily found via Google/ChatGPT (e.g., "write good content")
- **Edge Knowledge**: Scarce, risky, or controversial tactics from underground communities (e.g., "PBN networks that bypass Google penalties in 2026")

### Q8: How often is the forum database updated?
**A**: Community-driven! Submit PRs to add new forums or update ratings. We review and merge quality contributions regularly.

### Q9: Can I use this for white hat SEO?
**A**: While the tool focuses on "edge" knowledge, you can customize `forum_database.json` to include white hat forums and adjust search strategies accordingly.

### Q10: How do I report bugs or request features?
**A**: Open an issue on [GitHub Issues](https://github.com/1596941391qq/EdgeKnowledge_Skill/issues) with:
- Clear description of the bug/feature
- Steps to reproduce (for bugs)
- Expected vs actual behavior
- Your environment (OS, Claude Code version)
## OpenCLI Integration

This local skill now ships with a Windows PowerShell OpenCLI layer:

- `scripts/setup-opencli.ps1`
- `scripts/test-opencli.ps1`
- `scripts/invoke-opencli.ps1`

Recommended order:

1. Install and enable the Chrome extension named `Browser Bridge`
2. Run `scripts/setup-opencli.ps1`
3. Use `scripts/invoke-opencli.ps1` for browser-backed OpenCLI commands
4. Fall back to the existing `browser-use`, MCP, or custom forum flows when OpenCLI does not cover the target workflow

## 🗺️ V3 Roadmap — Community Track

V3 shipped the **mechanism layer** (freshness fields, GEO category, credibility scoring). The **refinement layer** below is where the real depth lives — and where community contributions are most welcome. These are hard, open problems:

| Track | Problem | Contribution opportunity |
|-------|---------|--------------------------|
| **Source-first mining** | Keyword-driven search misses knowledge you didn't know to look for. Can the crawler read a high-signal source *first*, then let topics emerge from what it finds? | A mode that ingests a source fully and surfaces emergent topics, instead of matching pre-set keywords |
| **Gap-driven scheduling** | What should you mine next? Today it's manual. Could the system detect where it already has dense raw material but no synthesized output, and prioritize that? | A `gap-assessment` pass that ranks "ready-to-synthesize" topics by raw-material density |
| **Incremental de-dup** | Re-crawling a source re-reports everything. It should only surface the *net new* since last run. | A knowledge-signature store so repeat crawls emit deltas, not full dumps |
| **Provenance & contradiction** | Credibility scoring is a start. Production-grade verification means tagging every claim's provenance, cross-checking freshness, and flagging contradictions without inflating the report. | A provenance layer that tags each extracted claim (first-hand / cross-verified / unverified) and surfaces conflicts |
| **Reader-grade output** | Raw reports read like crawl logs. The endgame is wiki-grade knowledge: synthesized, cross-referenced, source-anonymized, no outline-stubs. | A post-processing pass that turns extraction dumps into standalone knowledge articles |

> 🤝 Pick a track, open a branch (e.g. `source-first`, `incremental-dedup`), and PR it. These are the problems that separate a scraper from a knowledge engine.