5,439,563 AI resources indexedWorking toward the world’s largest AI resource index.

PDF

打假技能 · AI 工具安利粉碎机

Evaluates AI tools against hype and practicality using a structured three-layer audit system, ideal for discerning legitimacy in Chinese contexts.

Installable skillMITBy tUpdated Nov 3, 2026
INTENDED FORClaude CodeCodexOpenClawCursorInstall in a listed agent.

instructions.md

---
name: anti-hype-filter-f3d59b93e560
description: >-
  Three-layer sandbox audit system for filtering AI marketing noise. 
  Evaluates tools on 4 dimensions: relevance, hype, practicality, and 
  delivery gap vs. claims. Use when user wants to verify if a hyped AI 
  tool is worth trying, needs a structured evaluation framework, or asks 
  "is this tool legit?" in Chinese contexts.
license: MIT
compatibility: Requires Python 3.8+, DeepSeek API key
metadata:
  author: t
  version: "1.0"
  tags: ai-tools hype-detection evaluation audit chinese
---

# 打假技能 · AI 工具安利粉碎机

> 触发词:`打假` `粉碎机` `沙盘评估` `这个工具靠谱吗` `帮我鉴定` `安利粉碎` `AI 工具测评`
>
> 不触发:一般性工具推荐、纯信息搜索、不涉及 AI 工具的提问

## 能力边界

三层沙盘审核系统,专为过滤 AI 圈营销噪音设计:

1. **多源自动抓取**:GitHub Trending / Hacker News / Product Hunt / Reddit / 短视频平台
2. **4 维第一层粗筛**(DeepSeek 自动评分):相关度、吹嘘度、可操作性、宣传落差
3. **第二层 15 分钟快测模板**:5 维度手工填表打分
4. **第三层深度评估矩阵**:6 维度加权汇总 + 决策矩阵
5. **专项突击模式**:可对特定品类定向批量评估

## 执行流程

### 标准运行
```
python run.py
→ fetcher.py 多源抓取 → distiller.py 4维评分 → reporter.py 生成报告
→ 输出 data/shattered_illusions_report.md
```

### 专项评估
```
python run.py --source <candidates.json>
→ 跳过抓取,直接评估指定文件
```

## 4 维评分标准

| 维度 | 1 分 | 10 分 | 通过阈值 |
|------|------|-------|----------|
| relevance_score | 完全不相关 | 核心 AI 工具 | ≥ 6 |
| hype_score | 零营销话术 | 纯吹嘘无实质 | ≤ 7 |
| practicality_score | 需编译+GPU | 浏览器即用 | ≥ 5 |
| delivery_gap_score | 所见即所得 | AE 特效骗局 | ≤ 6 |

## 决策矩阵

| 加权分 | 决策 | 动作 |
|--------|------|------|
| ≥ 80 | 纳入工作流 | 写文档替换旧方案 |
| ≥ 60 | 观望保留 | 每月复查一次 |
| ≥ 40 | 归档备忘 | 不主动关注 |
| ≥ 0 | 彻底丢弃 | 从收藏夹删除 |

## 前置条件

- DeepSeek API Key(环境变量 `DEEPSEEK_API_KEY` 或写入 config.py)
- Python 3.8+
- `pip install requests feedparser`

Discuss this resource

Platform guidance

Imported from the listed AgentSkill.sh category. Complete supplied package files and notices are retained. Includes source scripts; review them before use. Required external tools and credentials are not installed. No bundled code was executed during import. Individual agent task behavior has not been tested.