analyze-xiaohongshu

name	analyze-xiaohongshu
description	Analyze a Xiaohongshu (小红书/RED) account by scraping its notes and covers, then generating a comprehensive methodology report across 7 dimensions. Use when the user provides a Xiaohongshu account URL or user ID and asks to analyze the account, reverse-engineer its strategy, or generate a methodology report. Triggers on phrases like "分析小红书账号", "爬取小红书", "analyze xiaohongshu account", "生成账号方法论报告".

analyze-xiaohongshu

Scrape a Xiaohongshu account's notes and cover images, then produce a structured methodology report.

Workflow

Step 1: Parse input

Accept either format:

• Full URL: https://www.xiaohongshu.com/user/profile/<ID>
• Raw user ID string

Extract the account ID using the pattern /user/profile/([a-zA-Z0-9]+).

Step 2: Environment check

python3 --version          # must be 3.8+
pip3 show playwright       # check if installed

If playwright is missing:

pip3 install playwright && playwright install chromium

Step 3: Run the scraper

The scraper is at scripts/scraper.py (relative to this skill directory). Find its absolute path first.

python3 <skill_dir>/scripts/scraper.py <account_url_or_id>

Behavior:

• Opens a visible Chromium browser
• If no saved cookie exists or cookie is expired → waits for user to manually scan QR code and log in (up to 3 minutes)
• Saves cookie to data/cookies.json for reuse
• Scrolls the profile page to collect up to 100 note cards — no detail page visits
• For each card, extracts: note ID, URL, title, cover image URL, like count, content type (video/image)
• Downloads cover images to covers/
• Supports resume: skips notes already in data/<account_id>/notes.json
• Saves data to data/<account_id>/ inside the skill directory

Output paths:

<skill_dir>/data/
├── cookies.json
└── <account_id>/
    ├── profile.json
    ├── notes.json        # fields: note_id, url, title, cover_url, likes, type, cover_local_path
    └── covers/
        └── <note_id>.jpg

Step 4: Load data

After the scraper finishes, read the JSON files:

import json
profile = json.load(open("data/<account_id>/profile.json"))
notes   = json.load(open("data/<account_id>/notes.json"))

Step 5: Analyze covers (batched visual analysis)

Read cover images in batches of 5–10 using the Read tool (images). Prioritize covers from the top 20% notes by likes first, then the rest for comparison.

Engagement score per note = likes (only metric available from profile cards)

Step 6: Generate the report

Write a Markdown report to data/<account_id>/report.md. Include all 7 dimensions below.

Step 6.5: Generate analysis.json

After the report, generate data/<account_id>/analysis.json with structured data extracted from your analysis. The dashboard (scripts/dashboard.py) requires this file.

{
  "account_id": "<account_id>",
  "generated_at": "YYYY-MM-DD",
  "topic_categories": {
    "<category_name>": ["keyword1", "keyword2", ...]
  },
  "title_formulas": [
    {
      "name": "公式名称",
      "pattern": "regex_pattern",
      "template": "模板描述",
      "trigger": "触发机制",
      "color": "#hex",
      "examples": ["示例标题1", "示例标题2"]
    }
  ],
  "cloud_keywords": ["word1", "word2", ...],
  "radar_dimensions": ["收藏潜力", "分享潜力", "时效性", "专业门槛", "变现潜力"],
  "radar_scores": {
    "<category_name>": [0-100, 0-100, 0-100, 0-100, 0-100]
  },
  "strategies": [
    {
      "title": "策略标题",
      "body": "策略详细说明",
      "action": "具体行动建议"
    }
  ],
  "top10_note_ids": ["8-char-prefix", ...],
  "top10_analysis": [
    { "note_id_prefix": "8-char-prefix", "success_factors": "该笔记的成功因素拆解" }
  ],
  "hit_patterns": [
    "爆款共性总结（如：同一公式贡献X%总赞）"
  ],
  "title_hit_vs_miss": {
    "dimensions": ["对比维度1", "对比维度2", ...],
    "hit": ["爆款特征1", "爆款特征2", ...],
    "miss": ["普通特征1", "普通特征2", ...]
  },
  "cover_analysis": {
    "styles": [
      { "name": "风格名", "count": 4, "traits": "风格特征描述", "performance": "high|medium|mixed|low" }
    ],
    "hit_cover_patterns": ["爆款封面共性1", "爆款封面共性2"],
    "template_suggestion": "推荐的封面模板描述"
  },
  "category_insights": {
    "<category>": "该类别的定性分析（优势、弱点、建议）"
  },
  "format_comparison": {
    "video_avg_likes": 0,
    "image_avg_likes": 0,
    "video_total_likes_pct": 0,
    "insight": "视频vs图文总结",
    "per_category": [
      { "category": "类别名", "best_format": "视频|图文|均可", "reason": "原因" }
    ]
  },
  "content_formulas_recap": [
    { "label": "A 公式名", "template": "公式模板" }
  ],
  "action_plan": {
    "this_week": ["本周行动1", "本周行动2"],
    "this_month": ["本月行动1"],
    "long_term": ["长期方向1"]
  }
}

Field guidelines:

• topic_categories: 3-6 categories with 5-20 keywords each, derived from the account's actual content themes
• title_formulas: 4-8 formulas with valid regex patterns, derived from the account's title patterns
• cloud_keywords: 20-40 high-frequency keywords from the account's titles
• radar_scores: One entry per topic category, scores 0-100 for each of the 5 radar dimensions
• strategies: 3-5 actionable strategies from the report's dimension 7
• top10_note_ids: First 8 characters of the top 10 notes' IDs (by likes)
• top10_analysis: Per-note success factor breakdown for each top 10 note, matched by note_id_prefix
• hit_patterns: 1-3 bullet points summarizing the key patterns shared by hit content
• title_hit_vs_miss: Multi-dimension comparison table of hit vs miss title characteristics
• cover_analysis: Cover style distribution, hit cover patterns, and template suggestion
• category_insights: One qualitative paragraph per topic category (strengths, weaknesses, recommendations)
• format_comparison: Video vs image performance comparison with per-category recommendations
• content_formulas_recap: 3-5 quick-reference content formulas (label + template)
• action_plan: Time-bound action items split into this_week, this_month, long_term

Step 7: Generate the dashboard

Run the dashboard generator to produce an interactive HTML dashboard:

python3 <skill_dir>/scripts/dashboard.py <account_id>

This reads data/<account_id>/analysis.json (required) along with profile.json and notes.json, and outputs data/<account_id>/dashboard.html.

Note on data availability: the scraper collects card-level data only (title, likes, cover, type). Hashtags and publish dates are not visible on profile cards and are excluded.

---

Report Structure (7 Dimensions)

1. 账号概览

• 昵称、简介、粉丝量级、内容类型定位
• 核心统计：总笔记数、平均点赞、点赞中位数
• 视频 vs 图文比例

2. 爆款笔记分析（Top 10）

• 按点赞数降序排列
• 每篇列出：标题、点赞数、内容类型、成功因素拆解
• 爆款定义：点赞量排名前 20% 的笔记

3. 标题分析

• 标题长度分布与最佳长度区间
• 高频词和句式模式（如"xx个方法"、"千万别xx"、提问式等）
• 爆款 vs 普通标题的差异对比
• 可复用的标题公式总结（至少 3 个）

4. 封面图分析（视觉分析）

• 风格类型分布（实拍 / 设计图 / 截图 / 对比图等）
• 配色特征、文字使用规律、构图规律
• 爆款封面的共性特征
• 可复用的封面模板建议

5. 内容选题分析

• 主题分类及各类占比（基于标题文本聚类）
• 各类别的平均点赞表现
• 最受欢迎的选题方向

6. 内容形式分析

• 图文 vs 视频的比例与点赞表现对比
• 哪种形式更受该账号受众欢迎

7. 可复制的方法论总结

• 该账号的核心增长策略（3-5条）
• 可直接复用的内容公式
• 具体行动建议清单（可执行的 to-do list）

---

Error Handling

• Scraper gets no notes: Check that section.note-item elements exist on the profile page. If not, the scraper falls back to a[href*="/explore/"] links.
• Cookie expired: Delete data/cookies.json and rerun the scraper to trigger a fresh login.
• Partial data: If the scraper was interrupted, rerun it — it will skip already-downloaded covers.
• Missing cover images: Proceed with analysis using available covers; note which notes lack covers.
• Anti-bot detection: The scraper stays on the profile page only. If detection still occurs, increase DELAY_MIN/DELAY_MAX in scraper.py.