Xiaohongshu Public Monitor

小红书舆论监控 - 自动搜索小红书帖子、筛选需要舆论引导的内容、生成多人设评论话术、输出到飞书多维表格。适用于任何品牌/产品的小红书舆情监控。Use when: (1) 需要监控小红书上的品牌/产品讨论, (2) 需要生成多角色评论话术, (3) 需要将监控结果输出到飞书表格, (4) 用户提到舆论监控、舆情分析...

Security Scan

The skill mostly matches its stated goal of searching Xiaohongshu, but there are mismatches between claims and implementation (Feishu upload / comment-generation pieces are described but not present or wired up), and it stores persistent browser login state and plaintext tokens in config — both deserve caution.

Purpose & Capability

The README/metadata promise a full workflow: search → filter → generate multi‑persona comments → write to Feishu Bitable. The included Python script implements searching/extraction using Playwright and persisting login state, but the repository files shown do not contain code to generate persona comments or to write results to Feishu. config.yaml expects a Feishu app_token/table_id, but the runtime code does not (in the visible files) use them. This inconsistency could be sloppy engineering or an incomplete package; it is not coherent as-is.

Instruction Scope

SKILL.md instructs the agent/user to login via a persistent Playwright browser session, run batch searches, and run the full workflow including comment generation and writing to Feishu. The script clearly instructs a user to scan to create a persistent login and to run searches. There are no instructions to read unrelated system files, nor evidence of exfiltration, but SKILL.md also includes operational guidance for coordinating multiple commenter personas and an explicit '建议水军数量' (suggest water‑army/account numbers), which enables coordinated influence operations and is ethically concerning. Additionally, the agent-facing instructions promise Feishu output and comment-generation steps that the provided script does not implement.

Install Mechanism

No opaque download or archive-extraction is used. The skill is instruction‑only plus a Python script; dependencies are standard (Playwright). Installation steps shown use git clone and pip install playwright / playwright install chromium — these are normal. The script hardcodes a macOS Chrome path which is brittle but not a supply‑chain red flag.

Credentials

config.yaml asks for Feishu app_token/table_id (sensible if writing to Feishu). However the skill registry metadata lists no required environment variables or primary credential, which is inconsistent with the config and SKILL.md metadata (which lists 'feishu-bitable' as a requirement). The script creates a local browser_data directory to persist login cookies — this stores sensitive session tokens on disk. Storing Feishu tokens in plaintext config.yaml (as instructed) is also sensitive. No other unrelated credentials are requested.

Persistence & Privilege

always is false and the skill does not request elevated platform privileges. It persists Playwright login state under its own browser_data directory and mentions an agent .learnings folder — this is confined to the skill's workspace and does not alter other skills or system settings. The ability to run autonomously is default and present, but not combined with other high privileges here.

What to consider before installing

This skill partly does what it says (search Xiaohongshu using Playwright) but the package appears incomplete: it promises comment generation and writing to Feishu, yet the visible script only performs searches and persists a browser session. Before installing, verify the following with the author or by inspecting the repo: (1) where and how the Feishu upload and comment-generation are implemented (search for feishu API calls and persona‑generation code), (2) that no code silently posts comments or exfiltrates data to unexpected endpoints, and (3) how login/session data and Feishu tokens are stored. If you proceed, do not place app tokens in plaintext config.yaml on a shared machine — use a secrets manager or environment variables, and run first in an isolated environment. Also consider the ethical and legal implications: the skill explicitly guides coordinating multiple personas and suggests 'water‑army' quantities, which can facilitate deceptive influence operations — ensure your use complies with laws and platform terms. If you cannot confirm the missing pieces or trust the author, treat this skill as untrusted.

小红书舆论监控 Skill

自动监控小红书帖子 → 筛选内容 → 生成评论话术 → 写入飞书多维表格。

适用于任何行业：AI、电商、美妆、餐饮、教育……只需修改 config.yaml 即可适配。

快速开始

1. 安装

# Via ClawHub
clawdhub install xiaohongshu-public-monitor

# 手动安装
git clone https://github.com/Lisayinyy/xiaohongshu-public-monitor.git ~/.openclaw/skills/xiaohongshu-public-monitor

# 安装依赖
pip install playwright
playwright install chromium

2. 登录小红书（仅首次）

python3 scripts/xhs_search.py login

弹出浏览器，扫码登录一次，后续自动复用。

3. 配置

编辑 config.yaml，填入：

• 你的品牌名和产品
• 搜索关键词
• 飞书多维表格 token
• 3个评论人设（根据你的行业定义）

4. 运行

# 搜索
python3 scripts/xhs_search.py batch "关键词1" "关键词2" --scroll 5

或让 Agent 执行完整工作流（搜索 → 筛选 → 生成评论 → 写入表格）。

---

工作流程（4步）

Step 1：搜索

• 按 config.yaml 中的关键词搜索小红书
• 筛选「一天内」+ 排序「最多点赞」
• Playwright 持久化登录，headless 模式，支持滚动翻页

Step 2：去重

• 查飞书表格已有链接，相同链接跳过

Step 3：筛选

基础门槛（3条必满足）：

1. 链接能打开，标题内容对得上
2. 作者不为空
3. 互动数据完整（👍点赞 ⭐收藏 💬评论）

内容质量（满足任一入表）：

• 直接提及你的品牌或产品
• 同行业多个品牌/产品对比
• 榜单、排名、测评类内容
• 价格/性价比讨论
• 高互动且有争议讨论

不入表：

• 与你的业务完全无关
• 纯招聘/面试
• 互动极低且无对比内容
• 评论区已有足够正面声音

Step 4：生成评论 + 写入表格

• 为每篇帖子生成 3人设 × 2条 = 6条评论
• 判定紧急程度（🔴高/🟡中/🟢低）
• 按「发现日期从新到旧 → 紧急程度高到低」排序写入飞书表格

---

评论人设

在 config.yaml 中定义 3 个人设。根据你的行业选择最合适的角色组合。

选人设的原则：选你的目标用户群里最有说服力的 3 种人。

示例：

行业	人设1	人设2	人设3
AI/科技	技术专家	普通用户	产品经理
美妆	资深化妆师	学生党	成分党博主
餐饮	美食博主	附近居民	餐饮同行
电商	老买家	第一次买的人	行业从业者
教育	在读学员	家长	教育行业人士

每个人设需要定义：

• 特征：这个人是谁
• 风格：怎么说话（口语化/专业/情绪化）
• 目的：为什么要用这个人设评论

---

评论红线

• ❌ 不写广告语
• ❌ 不贬低竞品（只客观对比）
• ❌ 不虚构数据
• ❌ 多条评论风格不能雷同
• ❌ 负面帖下不强行洗白

---

紧急程度判定

🔴 高：发布<24h + 有互动 + 对比/榜单类 + 负面倾向 🟡 中：有传播潜力 + 缺正面声音 + 评价中性 🟢 低：正面帖 + 互动稳定 + 已有正面声音

---

表格字段

字段	说明
文章标题	帖子标题
文章链接	原文链接
作者	作者昵称（不能为空）
发表时间	文章发布时间
发现日期	抓取日期
互动数据	👍⭐💬 三项齐全
关键词命中	命中的搜索词
提及竞品	提到的竞品
文章倾向	正面/中性/负面
紧急程度	🔴高/🟡中/🟢低
干预理由	为什么干预
人设评论	每人设2条
建议水军数量	投入账号数
执行状态	待处理/进行中/已完成

---

文件结构

xiaohongshu-public-monitor/
├── SKILL.md              # 使用说明（本文件）
├── config.yaml           # 配置文件（用户修改这个）
├── scripts/
│   └── xhs_search.py     # 小红书搜索脚本
├── references/
│   └── workflow.md        # 工作流说明
├── assets/
└── .learnings/            # Agent 学习记录