expert-finder

TotalClaw 作者 totalclaw

查找任何主题的领域专家、思想领袖和主题权威。在 Twitter 和 Reddit 中搜索在特定领域拥有深厚知识、频繁讨论和高于平均水平的专业知识的人员。专家发现、人才搜寻、研究人员识别和 KOL（关键意见领袖）映射。

安装 / 下载方式

TotalClaw CLI推荐

totalclaw skills install shop:totalclaw:totalclaw~atyachin-expert-finder

cURL直接下载，无需登录

curl -fsSL https://skills.taituai.com/api/skills/totalclaw%3Atotalclaw~atyachin-expert-finder/file -o atyachin-expert-finder.md

# Expert Finder

Find domain experts by analyzing social media activity. Expands topics into search terms, searches Twitter/Reddit, classifies by type, and ranks.

## Setup

Run `xpoz-setup` skill. Verify: `mcporter call xpoz.checkAccessKeyStatus`

## 4-Phase Process

### Phase 1: Query Expansion

Research domain with `web_search`/`web_fetch`. Generate tiered queries:

| Tier | Purpose | Example (RLHF) |
|------|---------|----------------|
| Tier 1: Core | Exact terms | `"RLHF"` |
| Tier 2: Technical | Deep jargon (strongest signal) | `"reward model overfitting"` |
| Tier 3: Adjacent | Related | `"preference optimization"` |
| Tier 4: Discussion | Opinion | `"RLHF vs"` |

### Phase 2: Search & Aggregate

```bash
mcporter call xpoz.getTwitterPostsByKeywords query='"RLHF"' startDate="<6mo>"
mcporter call xpoz.checkOperationStatus operationId="op_..." # Poll every 5s
```

Download CSVs via `dataDumpExportOperationId` (64K rows). Build author frequency: ≥3 posts, ≥2 tiers. Weight Tier 2 highest.

### Phase 3: Classify & Score

Fetch profiles for top 20-30:
```bash
mcporter call xpoz.getTwitterUser identifier="user" identifierType="username"
```

**Types:** 🔬 Deep Expert (uses Tier 2 naturally) | 💡 Thought Leader (trends, large audience) | 🛠️ Practitioner ("I built") | 📣 Evangelist (aggregates) | 🎓 Educator (explains)

**Score (0-100):** Domain depth 30%, consistency 20%, peer recognition 20%, breadth 15%, credentials 15%.

### Phase 4: Report

```markdown
## Expert Report: [Domain] — X,XXX posts analyzed

#### 🥇 @username — 🔬 Deep Expert (92/100)
**Followers:** 12.4K | **Why:** 23 posts on reward optimization, advanced terminology
**Key:** "[quote]" — ❤️ 342
```

## Tips

Narrow > broad | Tier 2 jargon = gold | Reddit comments reveal depth | 6mo window ideal

---

## 中文说明

# Expert Finder

通过分析社交媒体活动来查找领域专家。将主题扩展为搜索词，搜索 Twitter/Reddit，按类型分类并排名。

## 设置

运行 `xpoz-setup` 技能。验证：`mcporter call xpoz.checkAccessKeyStatus`

## 四阶段流程

### 第 1 阶段：查询扩展

使用 `web_search`/`web_fetch` 研究领域。生成分层查询：

| 层级 | 目的 | 示例（RLHF） |
|------|---------|----------------|
| 第 1 层：核心 | 精确术语 | `"RLHF"` |
| 第 2 层：技术 | 深度行话（信号最强） | `"reward model overfitting"` |
| 第 3 层：相邻 | 相关 | `"preference optimization"` |
| 第 4 层：讨论 | 观点 | `"RLHF vs"` |

### 第 2 阶段：搜索与聚合

```bash
mcporter call xpoz.getTwitterPostsByKeywords query='"RLHF"' startDate="<6mo>"
mcporter call xpoz.checkOperationStatus operationId="op_..." # Poll every 5s
```

通过 `dataDumpExportOperationId` 下载 CSV（64K 行）。构建作者频率：≥3 篇帖子，≥2 个层级。第 2 层权重最高。

### 第 3 阶段：分类与评分

为前 20-30 名抓取个人资料：
```bash
mcporter call xpoz.getTwitterUser identifier="user" identifierType="username"
```

**类型：** 🔬 深度专家（自然使用第 2 层术语）| 💡 思想领袖（趋势、受众庞大）| 🛠️ 实践者（"I built"）| 📣 布道者（聚合）| 🎓 教育者（讲解）

**评分（0-100）：** 领域深度 30%，一致性 20%，同行认可 20%，广度 15%，资历 15%。

### 第 4 阶段：报告

```markdown
## Expert Report: [Domain] — X,XXX posts analyzed

#### 🥇 @username — 🔬 Deep Expert (92/100)
**Followers:** 12.4K | **Why:** 23 posts on reward optimization, advanced terminology
**Key:** "[quote]" — ❤️ 342
```

## 提示

窄 > 宽 | 第 2 层行话 = 黄金 | Reddit 评论揭示深度 | 6 个月窗口最理想