<subtitle> ``` 或在代码中： ```python from xhscosmoskill.scripts.build_docx import build_word build_word(md_path="/tmp/report.md", out_path="/tmp/report.docx", title="xixiCharon", subtitle="爆款选题公式") ``` --- ## Cookie 管理 ```python # 优先使用（最新） COOKIES = ".../shopify-marketing/xhs_cookies.json" # 备用 COOKIES_ALT = ".../xiaohongshu_new/xhs_cookies.json" # Cookie 健康检查 from xhscosmoskill.utils import check_cookies status = check_cookies(COOKIES) # 返回 {valid: bool, expired_keys: list} ``` **Cookie 过期标志：** notes 返回 ≤ 1 条 → 提示重新运行 `xhs_login.py` --- ## 数据库操作 ```python from xhscosmoskill.archetype_registry import ( save_blogger, # 写入/更新博主记录 get_blogger, # 按名称查询 list_bloggers, # 列出所有博主 list_archetypes, # 查看当前类型库 add_archetype, # 新增自定义类型 update_archetype_signals # 迭代更新类型信号词 ) ``` --- ## 交付物规范 | 文件 | 格式 | 说明 | |------|------|------| | `{博主名}-结构化总结报告.md/.docx` | Markdown + Word | 单账号深度分析（15节）| | `{博主名}-爆款选题公式.md/.docx` | Markdown + Word | 6模型 + 30选题方向 | | `选题公式学习-综合版.md/.docx` | Markdown + Word | 多账号对比 | --- ## 证据分级 | 级别 | 来源 | 使用方式 | |------|------|---------| | A1 | 小红书公开主页可见数据 | 直接陈述 | | A2 | 用户提供截图 | 直接陈述 | | B1 | 第三方公开资料（采访/新榜/百科）| 作为背景补充 | | C1 | 综合推断 | 明确标注为"推断" | --- ## 参考资源 - **实战案例库（hsword）：** `openclaw_cosmo/afa/hsword/` - 井越（Type A）：荒诞美学型完整报告 + 爆款公式 - 橘一橙NiceFriend（Type B）：共鸣命名型完整报告 + 爆款公式 - 丑穷女孩陈浪浪（Type C）：现实策略型完整报告 + 爆款公式 - 选题公式学习-综合版：双系统 + 混合公式 + 6种标题公式 - **分析框架参考：** `references/workflow.md` - **报告模板：** `references/templates.md` - **hsword框架：** `references/hsword-frameworks.md` - **新榜数据：** `https://www.newrank.cn/profile/xiaohongshu/{user_id}` --- ## 已分析博主档案 | 博主 | 类型 | 最高赞 | 均赞 | 分析日期 | |------|------|--------|------|---------| | xixiCharon | B+A | 10万+ | 3,148 | 2026-04-23 | *更多博主将在 `/xhsfx` 调用后自动写入 data/bloggers.json* --- *整合自 xhscosmoskill v1.0 + xhsfenxi v2.1 + hsword实战案例 · 版本 2.0.0 · 2026-04-23* --- ## Purpose & Capability **Xhsfenxi Pro** is a full-stack Xiaohongshu (小红书) blogger analysis skill. It combines automated data collection with a structured deep-analysis framework derived from real-world case studies (hsword archive). | Capability | Description | |-----------|-------------| | Data Collection | Scrape user notes via SeleniumBase XHR interception — no API key required | | Three-Archetype Classification | Classify bloggers as Type A (Absurdist Aesthetics) / B (Resonance Naming) / C (Reality Strategy) or hybrid | | Core Identity Analysis (hsword) | Three-layer persona deconstruction: surface labels / mid-layer traits / deep values | | Viral Topic Formula | 6-model formula system with reusable sentence templates and 30 ready-to-use topic directions | | Structured Reports | Full 15-section analysis report in Markdown | | Word Output | Black Heiti-font Word documents via `scripts/build_docx.py` | | Iterative Archetype DB | `data/archetypes.json` evolves as more bloggers are analyzed | **Does NOT:** - Require any Xiaohongshu API token (uses browser-based XHR interception) - Access private/protected notes or accounts - Store or transmit user credentials - Generate fake engagement data or fabricate analysis results --- ## Instruction Scope **In scope — will handle:** - "Analyze this Xiaohongshu blogger URL" - "Generate viral topic formula for this account" - "What archetype is this blogger?" - "Produce a Word report for this creator" - "Compare two bloggers" - Any `/xhsfx` command invocation **Out of scope — will not handle:** - Accessing private accounts or bypassing platform security - Publishing or posting content to Xiaohongshu on behalf of users - Real-time follower/engagement data (uses public page data only) - Non-Xiaohongshu platforms **When cookies expire:** The browser session cookie file expires approximately every 30 days. If `notes` returns ≤ 1 result, prompt the user to re-run `python3 xhs_login.py` to refresh cookies. The skill will not silently fail — it will detect and report the expired state. --- ## Credentials This skill uses **no API tokens or platform credentials for analysis**. | Action | Credential | Scope | |--------|-----------|-------| | Data collection | Xiaohongshu session cookie (browser-based) | Local file only — `xhs_cookies.json` | | Report generation | None | Local file writes only | | Word output | None | Local file writes only | Cookie file locations (in priority order): 1. `shopify-marketing/xhs_cookies.json` (preferred — most recent) 2. `xiaohongshu_new/xhs_cookies.json` (fallback) **Does NOT hardcode tokens, API keys, or account credentials.** Cookie files are local only and never transmitted. --- ## Persistence & Privilege | Path | Content | When written | |------|---------|-------------| | `data/archetypes.json` | Archetype registry — evolves as bloggers are analyzed | On each `save_blogger()` call | | `data/bloggers.json` | Analyzed blogger database | On each `save_blogger()` call | | `/tmp/{creator}-*.md` | Markdown analysis reports | On report generation | | `/tmp/{creator}-*.docx` | Word documents | On `build_word()` call | **Does NOT write to:** - Any system directories outside the skill directory and `/tmp/` - Shell configuration files (`~/.zshrc` etc.) - Xiaohongshu platform (read-only) **Uninstall:** Delete the skill directory. Cookie files at `xhs_cookies.json` paths can be deleted separately. --- ## Install Mechanism ```bash clawhub install xhsfenxi-pro ``` **Prerequisites:** ```bash pip install seleniumbase python-docx ``` **First-time cookie setup:** ```bash python3 xhs_login.py # Opens browser → log in to Xiaohongshu → cookies saved automatically ``` **Verify installation:** ```python import sys sys.path.insert(0, "/path/to/xhscosmoskill/..") import xhscosmoskill print(xhscosmoskill.__version__) # should print: 2.0.0 from xhscosmoskill import print_cookie_status print_cookie_status() # ✅ 全部有效 / ❌ 已过期 ``` **Quick analysis:** ```python from xhscosmoskill import XhsClient, analyze_account with XhsClient() as xhs: notes = xhs.get_user_notes("USER_ID_HERE", limit=50) report = xhs.analyze_account(notes, creator_name="BloggerName") print(report) ``` FILE:README.md # xhscosmoskill — 小红书数据抓取工具 > Agent 可直接调用的小红书内容抓取客户端。 > 基于 SeleniumBase UC Mode + XHR 拦截，绕过反爬检测，无需逆向签名算法。 --- ## 核心原理 ``` 用户登录一次 → 保存 Cookie ↓ SeleniumBase UC Mode 打开浏览器（绕过 headless 检测） ↓ 注入 JS 拦截器（劫持 fetch/XHR） ↓ 浏览器自动带签名请求小红书 API ↓ 拦截器捕获 API 响应 JSON → 返回结构化数据 ``` 不需要逆向 X-s/X-t 签名，因为我们让浏览器自己完成签名，然后从内存中读取响应。 --- ## 前置步骤（只需一次） ```bash # 1. 安装依赖 pip install seleniumbase # 2. 登录小红书，保存 cookie cd /Users/zezedabaobei/Desktop/cosmocloud/Deeplumen/cosmowork/xiaohongshu_new python3 xhs_login.py # → 浏览器弹出，手动登录，按 Enter 保存 # → 生成 xhs_cookies.json（有效期约 30 天） ``` --- ## 快速使用 ```python import sys sys.path.insert(0, "/Users/zezedabaobei/Desktop/cosmocloud/Deeplumen/cosmowork/xiaohongshu_new") from xhscosmoskill import XhsClient with XhsClient() as xhs: # 搜索关键词 notes = xhs.search("咖啡", limit=20) # 获取用户所有笔记 notes = xhs.get_user_notes("5b07eb49e8ac2b5fe1e7de09", limit=50) # 获取单篇详情（正文 + 评论） note = xhs.get_note_detail("https://www.xiaohongshu.com/explore/xxx") # 批量补充详情 notes = xhs.batch_get_details(notes, delay=2.0) # 保存 xhs.save(notes, "output.json") ``` --- ## API 文档 ### `XhsClient(cookies_file, headless, scroll_times)` | 参数 | 类型 | 默认 | 说明 | |------|------|------|------| | cookies_file | str | `../xhs_cookies.json` | Cookie 文件路径 | | headless | bool | False | 无头模式（稳定后可开启） | | scroll_times | int | 5 | 默认滚动次数（影响加载量） | --- ### `xhs.search(keyword, limit, scroll_times) → List[Note]` 搜索笔记。 ```python notes = xhs.search("咖啡", limit=30) # notes[0].title, notes[0].likes, notes[0].url, notes[0].type ``` --- ### `xhs.get_user_notes(user_id, limit, scroll_times) → List[Note]` 获取用户主页所有笔记。`user_id` 是 URL 中的 hex 字符串。 ```python # URL: xiaohongshu.com/user/profile/5b07eb49e8ac2b5fe1e7de09 notes = xhs.get_user_notes("5b07eb49e8ac2b5fe1e7de09", limit=100) ``` --- ### `xhs.get_note_detail(note_url) → Note | None` 获取单篇笔记详情，包含正文、评论、互动数据。 ```python note = xhs.get_note_detail("https://www.xiaohongshu.com/explore/xxx") print(note.content) # 正文全文 print(note.comments) # List[Comment] ``` --- ### `xhs.batch_get_details(notes, delay) → List[Note]` 在搜索/列表结果基础上批量补充正文和评论。 ```python notes = xhs.search("咖啡", limit=20) notes = xhs.batch_get_details(notes, delay=2.0) # delay 建议 1.5~3 秒 ``` --- ### `xhs.save(notes, filepath)` 保存为 JSON 文件。 --- ## 数据模型 ### `Note` ``` id str 笔记 ID title str 标题 content str 正文全文 url str 完整 URL author str 作者昵称 author_id str 作者 ID likes str 点赞数（字符串，如 "3万"） collects str 收藏数 comments_count str 评论数 type str "video" | "normal" images list 图片 URL 列表 tags list 话题标签列表 comments list List[Comment] ``` ### `Comment` ``` user str 评论者昵称 text str 评论内容 likes str 评论点赞数 time str 评论时间 ``` --- ## 稳定性说明 | 场景 | 稳定性 | 说明 | |------|--------|------| | 搜索列表 | ✅ 稳定 | DOM 兜底，基本必成 | | 用户主页列表 | ✅ 稳定 | 同上 | | 笔记详情正文 | ⚠️ 较好 | API 拦截优先，DOM 兜底 | | 评论数据 | ⚠️ 较好 | 依赖 API 拦截成功 | - Cookie 有效期约 30 天，过期后重跑 `xhs_login.py` - 建议单账号每次请求间隔 1.5 秒以上，避免风控 - 首次使用建议 `headless=False` 观察浏览器行为 --- ## 文件结构 ``` xhscosmoskill/ ├── __init__.py 入口，导出 XhsClient ├── client.py 主客户端类（对外 API） ├── browser.py 浏览器会话 + XHR 拦截器 ├── parser.py API JSON / DOM 解析器 ├── models.py 数据模型（Note / Comment / UserProfile） ├── README.md 本文件 └── examples/ ├── search_notes.py 搜索示例 ├── user_profile.py 用户主页示例 └── note_detail.py 单篇详情示例 ``` --- ## 常见问题 **Q: `Cookie 文件不存在` 报错** A: 运行 `python3 xhs_login.py` 先登录一次。 **Q: content 字段为空** A: 该笔记的详情页 API 未被拦截到。可尝试增加 `wait` 时间，或该笔记需要登录才可见。 **Q: 运行很慢** A: 正常，每次操作需要等待页面渲染。可在稳定后开启 `headless=True`。 **Q: 被封号怎么办** A: 降低频率（`delay=3`），换账号重新 `xhs_login.py`。 FILE:__init__.py from .client import XhsClient from .analyzer import analyze_account, classify_archetype, compute_stats, build_five_layers from .formula import generate_formula_report from .archetype_registry import ( save_blogger, get_blogger, list_bloggers, list_archetypes, add_archetype, update_archetype_signals, ) from .utils import check_cookies, get_best_cookies, print_cookie_status __all__ = [ "XhsClient", "analyze_account", "classify_archetype", "compute_stats", "build_five_layers", "generate_formula_report", "save_blogger", "get_blogger", "list_bloggers", "list_archetypes", "add_archetype", "update_archetype_signals", "check_cookies", "get_best_cookies", "print_cookie_status", ] __version__ = "2.0.0" FILE:_meta.json { "ownerId": "kn79bebfnwg15sb0g7cj5z5nyd83gxh0", "slug": "xhsfenxi-pro", "version": "2.0.0", "publishedAt": null } FILE:analyzer.py """ analyzer.py — 小红书博主分析引擎整合 xhsfenxi 三型博主分类体系 + 五层账号模型可与 XhsClient 采集结果直接对接用法: from xhscosmoskill import XhsClient from xhscosmoskill.analyzer import analyze_account with XhsClient() as xhs: notes = xhs.get_user_notes("user_id", limit=50) report = analyze_account(notes, creator_name="xixiCharon") print(report) """ from __future__ import annotations from typing import List, Dict, Any import re from collections import Counter # ── 三型博主分类系统 ────────────────────────────────────────── ARCHETYPE_SIGNALS = { "A": { "name": "荒诞美学型", "desc": "用荒诞幽默包裹哲学内核，高质感视觉，品牌符号统一", "title_signals": ["（劲爆）", "劲爆", "＊", "*", "·"], "content_signals": ["哲", "美学", "荒诞", "飞", "高山", "宇宙", "奖励", "永恒"], "commercial": "高端生活方式、旅行、摄影、轻奢品牌", "difficulty": "高 — 依赖长期审美积累", }, "B": { "name": "共鸣命名型", "desc": "把私人经历转化为普世命题，给模糊情绪命名", "title_signals": ["*", "为什么", "是什么", "原来", "才知道", "命题", "灯塔", "解药"], "content_signals": ["成长", "情绪", "留学", "记录", "年末", "不确定", "命题", "感受", "理解", "接纳"], "commercial": "成长/教育/创意平台、中端生活方式", "difficulty": "中 — 方法可学，需要真实观察力", }, "C": { "name": "现实策略型", "desc": "打破职场/人际/金钱潜规则，提供可执行的向上策略", "title_signals": ["骗子", "不要脸", "装", "穷", "规则", "策略", "大法"], "content_signals": ["策略", "规则", "执行", "调整", "快速", "现实", "向上", "方法"], "commercial": "大众消费、职场、电商、实用工具", "difficulty": "中 — 框架可学，切忌复制表面攻击性", }, } def classify_archetype(notes: List[Any]) -> Dict[str, Any]: """ 基于笔记标题关键词，初步判断博主类型。返回: { "type": "B", "name": "共鸣命名型", "scores": {...}, "rationale": [...] } """ scores = {"A": 0, "B": 0, "C": 0} evidence = {"A": [], "B": [], "C": []} for note in notes: title = note.title or "" if hasattr(note, "title") else note.get("title", "") content = note.content or "" if hasattr(note, "content") else note.get("content", "") text = title + " " + content for typ, cfg in ARCHETYPE_SIGNALS.items(): for sig in cfg["title_signals"]: if sig in title: scores[typ] += 2 evidence[typ].append(f"标题含「{sig}」: {title[:30]}") for sig in cfg["content_signals"]: if sig in text: scores[typ] += 1 # 归一化 total = sum(scores.values()) or 1 pct = {k: round(v / total * 100) for k, v in scores.items()} best = max(scores, key=scores.get) second = sorted(scores, key=scores.get, reverse=True)[1] # 混合判断 is_mixed = scores[second] >= scores[best] * 0.6 if is_mixed: archetype_type = f"{best}+{second}" archetype_name = f"{ARCHETYPE_SIGNALS[best]['name']} × {ARCHETYPE_SIGNALS[second]['name']}" else: archetype_type = best archetype_name = ARCHETYPE_SIGNALS[best]["name"] rationale = evidence[best][:3] + (evidence[second][:2] if is_mixed else []) return { "type": archetype_type, "name": archetype_name, "desc": ARCHETYPE_SIGNALS[best]["desc"], "scores": scores, "pct": pct, "rationale": rationale, "commercial": ARCHETYPE_SIGNALS[best]["commercial"], "difficulty": ARCHETYPE_SIGNALS[best]["difficulty"], "is_mixed": is_mixed, } # ── 数据工具 ────────────────────────────────────────────────── def _parse_likes(v) -> int: if not v: return 0 s = str(v).replace(",", "").strip() if "万" in s: return int(float(s.replace("万", "")) * 10000) try: return int(s) except ValueError: return 0 def _engagement_tier(likes: int) -> str: if likes >= 50000: return "爆款 🔥" if likes >= 10000: return "高量 ⭐" if likes >= 5000: return "中高量" if likes >= 1000: return "稳定量" return "普通" # ── 五层账号模型 ────────────────────────────────────────────── def build_five_layers(notes: List[Any], archetype: Dict) -> Dict[str, Any]: """ 提取五层账号模型: 1. Identity — 博主是谁（人设定位） 2. Contract — 用户为什么关注（情感契约） 3. Topic — 反复覆盖的问题/欲望（选题体系） 4. Expression — 标题/开头公式、品牌符号 5. Transfer — 可学习点 vs 不能硬抄 """ titles = [] likes_list = [] for n in notes: t = n.title if hasattr(n, "title") else n.get("title", "") lk = _parse_likes(n.likes if hasattr(n, "likes") else n.get("likes")) if t: titles.append((t, lk)) likes_list.append(lk) # 选题体系：基于标题词频聚类 topic_keywords = { "自然/户外": ["山", "自然", "户外", "攀岩", "跑", "溪", "飞", "森林", "海"], "留学/海外": ["留学", "巴黎", "欧洲", "外国", "服装生", "独居", "课后"], "情绪/成长": ["情绪", "成长", "失败", "不确定", "淤青", "消极", "烦恼", "困境"], "旅行/探索": ["旅行", "solotrip", "solo", "gap", "世界", "探索"], "生活哲学": ["生活", "命题", "本质", "人生", "记录", "计划", "感受", "存在", "宇宙"], "创作/记录": ["记录", "vlog", "日记", "分享", "创造力"], } topic_dist: Dict[str, List] = {k: [] for k in topic_keywords} for title, lk in titles: for topic, kws in topic_keywords.items(): if any(kw in title for kw in kws): topic_dist[topic].append((title, lk)) best_topic = max(topic_dist, key=lambda k: sum(x[1] for x in topic_dist[k]) / max(len(topic_dist[k]), 1)) # 标题模式分析 has_asterisk = sum(1 for t, _ in titles if "*" in t or "＊" in t) has_question = sum(1 for t, _ in titles if "为什么" in t or "如何" in t or "？" in t) has_contrast = sum(1 for t, _ in titles if "但" in t or "，" in t) has_emotion_label = sum(1 for t, _ in titles if any(w in t for w in ["模糊", "流动", "感受", "治愈", "珍贵"])) # 爆款标题 top_titles = sorted(titles, key=lambda x: -x[1])[:5] return { "identity": _infer_identity(titles, archetype), "contract": _infer_contract(archetype), "topic_system": {k: len(v) for k, v in topic_dist.items() if v}, "best_topic": best_topic, "best_topic_avg": round( sum(x[1] for x in topic_dist[best_topic]) / max(len(topic_dist[best_topic]), 1) ), "expression": { "asterisk_ratio": f"{has_asterisk}/{len(titles)}", "question_titles": has_question, "contrast_titles": has_contrast, "emotion_labels": has_emotion_label, "top_titles": top_titles, }, "transfer": _infer_transfer(archetype), } def _infer_identity(titles, archetype) -> str: typ = archetype["type"][0] if typ == "A": return "用荒诞美学构建独特视角的创作者，以品牌符号建立强识别度" if typ == "B": return "把私人经历转化为公共命题、擅长给模糊情绪命名的思考者" return "洞察现实规则、提供可执行策略的实用派创作者" def _infer_contract(archetype) -> str: typ = archetype["type"][0] if typ == "A": return "看了就愉快 + 被美学震撼 + 看到了另一种看待生活的方式" if typ == "B": return "被理解 + 获得命名 + 对自己的状态有了新的视角" return "被验证 + 被激活 + 获得可执行的向上移动路径" def _infer_transfer(archetype) -> Dict[str, List[str]]: typ = archetype["type"][0] if typ == "A": return { "可学": ["建立自己的品牌符号", "练习反差命名而非描述地点", "荒诞框架选题"], "不能硬抄": ["视觉审美需长期积累", "语气独特性不可复制"], } if typ == "B": return { "可学": ["命题化思维：私人经历→公共命题", "建立个人概念词汇库", "五步公式：经历→命题→命名→判断→共鸣"], "不能硬抄": ["真实性是基础，无法凭空虚构", "语气温度需与真实人设一致"], } return { "可学": ["找到自己的现实母题", "冲突词标题练习", "公式：困境→说破→规则→策略→爽感"], "不能硬抄": ["表面攻击性语气", "特定人设的羞耻感先夺法"], } # ── 数据统计 ────────────────────────────────────────────────── def compute_stats(notes: List[Any]) -> Dict[str, Any]: likes = [_parse_likes(n.likes if hasattr(n, "likes") else n.get("likes")) for n in notes] types = Counter( (n.type if hasattr(n, "type") else n.get("type", "unknown")) for n in notes ) brackets = {"1万+": 0, "5000-9999": 0, "1000-4999": 0, "500-999": 0, "<500": 0} for lk in likes: if lk >= 10000: brackets["1万+"] += 1 elif lk >= 5000: brackets["5000-9999"] += 1 elif lk >= 1000: brackets["1000-4999"] += 1 elif lk >= 500: brackets["500-999"] += 1 else: brackets["<500"] += 1 top = sorted( [(n.title if hasattr(n, "title") else n.get("title", ""), _parse_likes(n.likes if hasattr(n, "likes") else n.get("likes"))) for n in notes], key=lambda x: -x[1], )[:10] return { "total": len(notes), "video_count": types.get("video", 0), "normal_count": types.get("normal", 0), "total_likes": sum(likes), "avg_likes": round(sum(likes) / len(likes)) if likes else 0, "max_likes": max(likes) if likes else 0, "brackets": brackets, "top10": top, } # ── 主入口 ──────────────────────────────────────────────────── def analyze_account( notes: List[Any], creator_name: str = "未知博主", mode: str = "full", ) -> str: """ 对采集到的笔记列表做完整分析，返回 Markdown 报告。参数: notes — XhsClient.get_user_notes() 的返回值 creator_name — 博主名称 mode — 'full' 完整报告 | 'formula' 只输出选题公式 | 'snapshot' 快速摘要 """ stats = compute_stats(notes) archetype = classify_archetype(notes) five = build_five_layers(notes, archetype) if mode == "snapshot": return _render_snapshot(creator_name, stats, archetype) if mode == "formula": return _render_formula(creator_name, stats, archetype, five) return _render_full_report(creator_name, stats, archetype, five) # ── 渲染函数 ────────────────────────────────────────────────── def _render_snapshot(name, stats, archetype) -> str: return f"""# {name} — 快速摘要 | 指标 | 数值 | |------|------| | 分析笔记 | {stats['total']} 篇 | | 视频占比 | {stats['video_count']}/{stats['total']} | | 平均点赞 | {stats['avg_likes']:,} | | 最高单篇 | {stats['max_likes']:,} | | 博主类型 | **{archetype['name']}** ({archetype['type']}) | | 商业适配 | {archetype['commercial']} | """ def _render_formula(name, stats, archetype, five) -> str: top_topic = five["best_topic"] top_avg = five["best_topic_avg"] transfer = five["transfer"] top_titles_md = "\n".join( f"- 👍{lk:,} {t[:50]}" for t, lk in five["expression"]["top_titles"] ) can_learn = "\n".join(f"- ✅ {x}" for x in transfer["可学"]) cant_copy = "\n".join(f"- ❌ {x}" for x in transfer["不能硬抄"]) return f"""# {name} — 爆款选题公式 > 类型：{archetype['name']} | 平均点赞：{stats['avg_likes']:,} | 最高：{stats['max_likes']:,} ## 一、为什么能做出爆款？博主类型：**{archetype['name']}** {archetype['desc']} 核心公式：{_get_core_formula(archetype['type'][0])} ## 二、最强选题方向最高互动主题：**{top_topic}**（均赞 {top_avg:,}） ## 三、爆款标题参考 {top_titles_md} ## 四、标题模式 - 含 * 符号：{five['expression']['asterisk_ratio']}（情感标记） - 疑问句式：{five['expression']['question_titles']} 篇 - 情绪词命名：{five['expression']['emotion_labels']} 篇 ## 五、可学 / 不能硬抄 {can_learn} {cant_copy} """ def _render_full_report(name, stats, archetype, five) -> str: topic_table = "\n".join( f"| {t} | {c}篇 |" for t, c in sorted(five["topic_system"].items(), key=lambda x: -x[1]) ) brackets_md = "\n".join( f"| {k} | {'█' * v} {v}篇 |" for k, v in stats["brackets"].items() ) top10_md = "\n".join( f"| {i+1} | 👍{lk:,} | {_engagement_tier(lk)} | {t[:45]} |" for i, (t, lk) in enumerate(stats["top10"]) ) rationale_md = "\n".join(f"- {r}" for r in archetype["rationale"]) or "- 基于内容风格综合判断" can_learn = "\n".join(f"- ✅ {x}" for x in five["transfer"]["可学"]) cant_copy = "\n".join(f"- ❌ {x}" for x in five["transfer"]["不能硬抄"]) return f"""# {name} — 结构化分析报告 > 数据来源：小红书公开主页，样本量 {stats['total']} 篇 --- ## 一、账号快照 | 指标 | 数值 | |------|------| | 分析样本 | {stats['total']} 篇 | | 视频/图文 | {stats['video_count']} / {stats['normal_count']} | | 总点赞 | {stats['total_likes']:,} | | 平均点赞 | {stats['avg_likes']:,} | | 最高单篇 | {stats['max_likes']:,} | | 爆款（1万+） | {stats['brackets']['1万+']} 篇 | --- ## 二、账号类型判断 **类型：** {archetype['name']} ({archetype['type']}) **核心内核：** {archetype['desc']} **判断依据：** {rationale_md} --- ## 三、人设定位（Identity） {five['identity']} --- ## 四、用户情感契约（Audience Contract）用户关注这个博主，是为了： {five['contract']} --- ## 五、选题体系（Topic System） | 主题方向 | 篇数 | |----------|------| {topic_table} **最强方向：** {five['best_topic']}（均赞 {five['best_topic_avg']:,}） --- ## 六、表达体系（Expression System） | 维度 | 数据 | |------|------| | 含 * 情感符号 | {five['expression']['asterisk_ratio']} 篇 | | 疑问句标题 | {five['expression']['question_titles']} 篇 | | 情绪词命名 | {five['expression']['emotion_labels']} 篇 | **核心公式：** {_get_core_formula(archetype['type'][0])} --- ## 七、高互动内容 Top 10 | # | 点赞 | 量级 | 标题 | |---|------|------|------| {top10_md} --- ## 八、点赞分布 | 段位 | 分布 | |------|------| {brackets_md} --- ## 九、商业化判断 - **适合方向：** {archetype['commercial']} - **复制难度：** {archetype['difficulty']} --- ## 十、可学 / 不能硬抄 **可以学习：** {can_learn} **不能硬抄：** {cant_copy} --- ## 十一、后续建议 - 路径 A：深入拆解 3-5 篇高互动笔记的正文结构 - 路径 B：制作专项爆款选题公式（`mode='formula'`） - 路径 C：引入对标账号做多账号综合对比 """ def _get_core_formula(typ: str) -> str: formulas = { "A": "荒诞场景 × 品牌符号 × 哲思轻量化 → 审美共鸣", "B": "私人经历 → 命题化 → 给模糊状态命名 → 普世共鸣", "C": "困境 → 说破规则 → 提供策略 → 爽感执行", } return formulas.get(typ, "内容积累 → 风格稳定 → 用户粘性") FILE:archetype_registry.py """ archetype_registry.py — 可迭代博主类型注册表随分析案例积累，自动更新类型数据库，支持添加新类型 """ from __future__ import annotations import json import os from datetime import date from typing import Dict, Any, Optional _BASE = os.path.dirname(__file__) ARCHETYPES_FILE = os.path.join(_BASE, "data", "archetypes.json") BLOGGERS_FILE = os.path.join(_BASE, "data", "bloggers.json") # ── 读写工具 ────────────────────────────────────────────────── def _load(path: str) -> dict: with open(path, encoding="utf-8") as f: return json.load(f) def _save(path: str, data: dict): with open(path, "w", encoding="utf-8") as f: json.dump(data, f, ensure_ascii=False, indent=2) # ── 类型注册表操作 ──────────────────────────────────────────── def load_archetypes() -> Dict[str, Any]: """加载当前所有类型（内置 + 自定义）""" db = _load(ARCHETYPES_FILE) merged = {**db["archetypes"], **db.get("custom_archetypes", {})} return merged def add_archetype(key: str, name: str, desc: str, formula: str, title_signals: list, content_signals: list, commercial: str, difficulty: str = "中") -> str: """ 添加新博主类型（当现有三型无法覆盖时）参数: key — 类型代码，如 "D"、"E" 或自定义字符串 name — 类型名，如 "知识科普型" ... 返回: 成功提示 """ db = _load(ARCHETYPES_FILE) if key in db["archetypes"]: return f"⚠️ 类型 {key} 已存在于内置类型，请用其他 key" db["custom_archetypes"][key] = { "name": name, "desc": desc, "formula": formula, "title_signals": title_signals, "content_signals": content_signals, "commercial": commercial, "difficulty": difficulty, "examples": [], "confirmed_by": 0, } db["_meta"]["last_updated"] = str(date.today()) _save(ARCHETYPES_FILE, db) return f"✅ 新增类型 {key}「{name}」" def update_archetype_signals(key: str, new_title_signals: list = None, new_content_signals: list = None) -> str: """从新分析的博主身上迭代更新信号词""" db = _load(ARCHETYPES_FILE) target = db["archetypes"].get(key) or db["custom_archetypes"].get(key) if not target: return f"❌ 类型 {key} 不存在" if new_title_signals: existing = set(target["title_signals"]) added = [s for s in new_title_signals if s not in existing] target["title_signals"].extend(added) if new_content_signals: existing = set(target["content_signals"]) added = [s for s in new_content_signals if s not in existing] target["content_signals"].extend(added) target["confirmed_by"] = target.get("confirmed_by", 0) + 1 db["_meta"]["last_updated"] = str(date.today()) _save(ARCHETYPES_FILE, db) return f"✅ 类型 {key} 信号词已更新" def list_archetypes() -> str: """打印当前所有类型（用于 /xhsfx 开头展示）""" db = _load(ARCHETYPES_FILE) lines = [f"📊 当前博主类型库（{db['_meta']['last_updated']} 更新）\n"] for key, cfg in db["archetypes"].items(): ex = f" — 案例: {', '.join(cfg['examples'][:2])}" if cfg["examples"] else "" lines.append(f" **{key}** {cfg['name']} × {cfg['confirmed_by']}个案例{ex}") custom = db.get("custom_archetypes", {}) if custom: lines.append("\n **自定义类型：**") for key, cfg in custom.items(): lines.append(f" **{key}** {cfg['name']} × {cfg['confirmed_by']}个案例") return "\n".join(lines) # ── 博主数据库操作 ──────────────────────────────────────────── def save_blogger(creator_name: str, user_id: str, archetype: dict, stats: dict, best_topic: str, best_topic_avg: int, tags: list = None, formula: str = "") -> str: """ 分析完成后写入博主数据库，并更新对应类型的案例数 """ db = _load(BLOGGERS_FILE) # 检查是否已存在 existing = next((b for b in db["bloggers"] if b["user_id"] == user_id), None) if existing: existing.update({ "analyzed_at": str(date.today()), "archetype": archetype["type"], "archetype_name": archetype["name"], "stats": stats, "best_topic": best_topic, "best_topic_avg": best_topic_avg, }) action = "更新" else: db["bloggers"].append({ "creator_name": creator_name, "user_id": user_id, "analyzed_at": str(date.today()), "archetype": archetype["type"], "archetype_name": archetype["name"], "stats": stats, "best_topic": best_topic, "best_topic_avg": best_topic_avg, "tags": tags or [], "formula": formula, }) db["_meta"]["total"] = len(db["bloggers"]) action = "新增" _save(BLOGGERS_FILE, db) # 同步更新类型案例 _update_archetype_example(archetype["type"][0], creator_name) return f"✅ 博主「{creator_name}」已{action}到数据库（共 {db['_meta']['total']} 位）" def _update_archetype_example(type_key: str, creator_name: str): """把博主名写入对应类型的 examples 列表""" db = _load(ARCHETYPES_FILE) target = db["archetypes"].get(type_key) or db["custom_archetypes"].get(type_key) if target and creator_name not in target["examples"]: target["examples"].append(creator_name) target["confirmed_by"] = len(target["examples"]) db["_meta"]["total_analyzed"] = sum( len(v["examples"]) for v in {**db["archetypes"], **db.get("custom_archetypes", {})}.values() ) _save(ARCHETYPES_FILE, db) def list_bloggers(archetype_filter: str = None) -> str: """列出已分析博主，可按类型过滤""" db = _load(BLOGGERS_FILE) bloggers = db["bloggers"] if archetype_filter: bloggers = [b for b in bloggers if archetype_filter in b["archetype"]] lines = [f"📋 已分析博主（共 {len(bloggers)} 位）\n"] for b in bloggers: lines.append( f" · **{b['creator_name']}** [{b['archetype_name']}] " f"均赞{b['stats'].get('avg_likes',0):,} 最高{b['stats'].get('max_likes',0):,} " f"| {b['analyzed_at']}" ) return "\n".join(lines) def get_blogger(creator_name: str) -> Optional[dict]: """按名称查找已分析博主记录""" db = _load(BLOGGERS_FILE) return next((b for b in db["bloggers"] if b["creator_name"] == creator_name), None) FILE:browser.py """ 浏览器引擎：SeleniumBase UC Mode + XHR 拦截核心思路：注入 JS 拦截 fetch/XHR，捕获小红书真实 API 响应 JSON 不需要逆向签名算法，直接从浏览器内存中读取已签名的响应 """ import json import time import os from seleniumbase import SB # 注入到页面的 JS 拦截器 _INTERCEPT_JS = """ (function() { if (window.__xhs_interceptor_installed) return; window.__xhs_interceptor_installed = true; window.__xhs_api_cache = {}; // 拦截 fetch const _fetch = window.fetch; window.fetch = async function(...args) { const url = typeof args[0] === 'string' ? args[0] : args[0]?.url || ''; const res = await _fetch.apply(this, args); if (url.includes('/api/sns/') || url.includes('edith.xiaohongshu')) { try { const clone = res.clone(); clone.json().then(data => { window.__xhs_api_cache[url] = data; }).catch(() => {}); } catch(e) {} } return res; }; // 拦截 XHR const _open = XMLHttpRequest.prototype.open; const _send = XMLHttpRequest.prototype.send; XMLHttpRequest.prototype.open = function(method, url) { this.__url = url; return _open.apply(this, arguments); }; XMLHttpRequest.prototype.send = function() { this.addEventListener('load', function() { if (this.__url && (this.__url.includes('/api/sns/') || this.__url.includes('edith.xiaohongshu'))) { try { window.__xhs_api_cache[this.__url] = JSON.parse(this.responseText); } catch(e) {} } }); return _send.apply(this, arguments); }; })(); """ class BrowserSession: def __init__(self, cookies_file: str, headless: bool = False): self.cookies_file = cookies_file self.headless = headless self._sb = None self._ctx = None def __enter__(self): self._ctx = SB(uc=True, headless2=self.headless) self._sb = self._ctx.__enter__() self._init() return self def __exit__(self, *args): if self._ctx: self._ctx.__exit__(*args) def _init(self): self._sb.uc_open_with_reconnect("https://www.xiaohongshu.com", 4) time.sleep(2) self._inject_cookies() # Register CDP script BEFORE refresh so it runs on reload self._register_cdp_interceptor() self._sb.refresh() time.sleep(2) self._sb.execute_script(_INTERCEPT_JS) def _inject_cookies(self): if not os.path.exists(self.cookies_file): raise FileNotFoundError(f"Cookie 文件不存在: {self.cookies_file}\n请先运行 xhs_login.py") with open(self.cookies_file, encoding="utf-8") as f: cookies = json.load(f) for c in cookies: c.pop("sameSite", None) c.pop("httpOnly", None) try: self._sb.driver.add_cookie(c) except Exception: pass def _register_cdp_interceptor(self): """注册 CDP onNewDocument 脚本，每次页面加载前执行（捕获 page-load API）""" try: self._sb.driver.execute_cdp_cmd("Page.addScriptToEvaluateOnNewDocument", { "source": _INTERCEPT_JS }) except Exception: pass def _install_interceptor(self): self._sb.execute_script(_INTERCEPT_JS) def navigate(self, url: str, wait: float = 3.0): """导航到页面：用 driver.get 保留 CDP 会话（使拦截器在 page-load 前运行）""" # Re-register in case CDP session was reset self._register_cdp_interceptor() # Use driver.get to keep CDP alive (interceptor runs before page JS) self._sb.driver.get(url) time.sleep(wait) # Also inject immediately for any late-arriving race conditions self._install_interceptor() def scroll_and_collect(self, times: int = 3, gap: float = 2.0): """滚动页面触发懒加载 API""" for _ in range(times): self._sb.execute_script("window.scrollBy(0, 1500)") time.sleep(gap) def get_api_cache(self) -> dict: """读取所有已拦截的 API 响应""" try: return self._sb.execute_script("return window.__xhs_api_cache || {}") or {} except Exception: return {} def clear_cache(self): self._sb.execute_script("window.__xhs_api_cache = {}") def get_page_source(self) -> str: return self._sb.get_page_source() def find_elements(self, selector: str): return self._sb.find_elements(selector) def execute_script(self, script: str): return self._sb.execute_script(script) @property def driver(self): return self._sb.driver FILE:client.py """ XhsClient — 小红书数据抓取 + 分析客户端基于 SeleniumBase UC Mode + XHR 拦截，无需逆向签名算法整合 xhsfenxi 三型博主分类体系 + 五层账号模型用法: from xhscosmoskill import XhsClient with XhsClient() as xhs: notes = xhs.search("咖啡", limit=30) user_notes = xhs.get_user_notes("5b07eb49e8ac2b5fe1e7de09") report = xhs.analyze_account(user_notes, creator_name="博主名") detail = xhs.get_note_detail("https://www.xiaohongshu.com/explore/xxx") """ import json import time import os from typing import List, Optional from .browser import BrowserSession from .models import Note, UserProfile from .parser import ( parse_search_response, parse_user_notes_response, parse_note_detail_response, parse_comment_page_response, parse_dom_notes, ) from .analyzer import analyze_account as _analyze_account DEFAULT_COOKIES = os.path.join(os.path.dirname(__file__), "..", "xhs_cookies.json") class XhsClient: """ 小红书数据抓取客户端参数: cookies_file: cookie 文件路径（运行 xhs_login.py 生成） headless: 是否无头模式（默认 False，首次建议 False 观察行为） scroll_times: 每次操作滚动次数（影响加载数量）支持 context manager（with 语句）或手动 open()/close() """ def __init__( self, cookies_file: str = DEFAULT_COOKIES, headless: bool = True, scroll_times: int = 5, ): self.cookies_file = os.path.abspath(cookies_file) self.headless = headless self.scroll_times = scroll_times self._session: Optional[BrowserSession] = None def open(self): self._session = BrowserSession(self.cookies_file, self.headless) self._session.__enter__() return self def close(self): if self._session: self._session.__exit__(None, None, None) self._session = None def __enter__(self): return self.open() def __exit__(self, *args): self.close() # ─── 公开接口 ────────────────────────────────────────────── def search(self, keyword: str, limit: int = 20, scroll_times: int = None) -> List[Note]: """ 搜索笔记参数: keyword: 搜索关键词 limit: 最大返回数量 scroll_times: 滚动次数（越多加载越多，默认用初始化值）返回: List[Note] """ self._ensure_session() url = f"https://www.xiaohongshu.com/search_result?keyword={keyword}&source=web_explore_feed" self._session.clear_cache() self._session.navigate(url, wait=3) self._session.scroll_and_collect(scroll_times or self.scroll_times) cache = self._session.get_api_cache() notes = parse_search_response(cache) # API 拦截失败时 fallback 到 DOM if not notes: notes = parse_dom_notes(self._session) return notes[:limit] def get_user_notes(self, user_id: str, limit: int = 50, scroll_times: int = None) -> List[Note]: """ 获取用户主页所有笔记参数: user_id: 小红书用户 ID（URL 中的那串 hex） limit: 最大返回数量 scroll_times: 滚动次数返回: List[Note] """ self._ensure_session() url = f"https://www.xiaohongshu.com/user/profile/{user_id}" self._session.clear_cache() self._session.navigate(url, wait=4) seen_ids = set() notes = [] n_scroll = scroll_times or self.scroll_times for _ in range(n_scroll): self._session.execute_script("window.scrollBy(0, 1500)") time.sleep(2) cache = self._session.get_api_cache() batch = parse_user_notes_response(cache) # DOM fallback if API empty if not batch: batch = parse_dom_notes(self._session) for n in batch: key = n.id or n.url or n.title if key and key not in seen_ids: seen_ids.add(key) notes.append(n) if len(notes) >= limit: break return notes[:limit] def get_note_detail(self, note_url: str) -> Optional[Note]: """ 获取单篇笔记详情（正文、评论、互动数据）参数: note_url: 笔记完整 URL 返回: Note 或 None """ self._ensure_session() self._session.clear_cache() self._session.navigate(note_url, wait=4) cache = self._session.get_api_cache() note = parse_note_detail_response(cache) # API 失败时从 DOM 提取正文和标题 if not note: note = Note() note.url = note_url for sel in ["#detail-desc", "span#detail-desc", ".desc.expand", "[class*='desc']"]: try: el = self._session.find_elements(sel) if el and el[0].text.strip(): note.content = el[0].text.strip() break except Exception: pass for sel in ["#detail-title", ".note-content .title", "h1", "[class*='title']"]: try: el = self._session.find_elements(sel) if el and el[0].text.strip(): note.title = el[0].text.strip() break except Exception: pass # 无论 API 是否成功，都尝试从 comment/page 补充评论 if not note.comments: note.comments = parse_comment_page_response(cache) return note if (note and (note.content or note.title)) else None def post_comment(self, note_url: str, text: str) -> bool: """ 在笔记下发表评论参数: note_url: 笔记完整 URL text: 评论内容返回: True 成功 / False 失败 """ self._ensure_session() self._session.navigate(note_url, wait=4) try: from selenium.webdriver.common.action_chains import ActionChains from selenium.webdriver.common.keys import Keys driver = self._session.driver # 1. JS 点击评论区占位符，激活输入框 activate_selectors = [ ".comment-input-inner-container", "[class*='comment-input']", ".input-inner-container", "[class*='commentInput']", "[class*='input-box']", ] activated = False for sel in activate_selectors: els = self._session.find_elements(sel) if els: driver.execute_script("arguments[0].scrollIntoView({block:'center'})", els[0]) time.sleep(0.5) driver.execute_script("arguments[0].click()", els[0]) activated = True time.sleep(1) break if not activated: print(" ✗ 找不到评论输入框") return False # 2. 找激活后的 contenteditable 或 textarea editable_selectors = [ "[contenteditable='true']", "textarea", "[class*='ql-editor']", "[class*='editor']", ] editable = None for sel in editable_selectors: els = self._session.find_elements(sel) if els: editable = els[0] break if editable: ActionChains(driver).move_to_element(editable).click().send_keys(text).perform() else: # fallback: 直接用 active element ActionChains(driver).send_keys(text).perform() time.sleep(1) # 3. 点击发送按钮 submit_selectors = [ "[class*='submit']", "[class*='send']", "button.submit", "[class*='confirm']", ] submitted = False for sel in submit_selectors: els = self._session.find_elements(sel) # 过滤不可见元素 for el in els: try: if el.is_displayed() and el.is_enabled(): driver.execute_script("arguments[0].click()", el) submitted = True break except Exception: continue if submitted: break if not submitted: # 最后 fallback: Enter ActionChains(driver).send_keys(Keys.ENTER).perform() time.sleep(2) print(f" ✓ 评论已发送: {text[:30]}") return True except Exception as e: print(f" ✗ 评论失败: {e}") return False def get_note_comments(self, note_url: str, limit: int = 20) -> List[dict]: """ 单独获取笔记评论参数: note_url: 笔记 URL limit: 最大评论数返回: List[dict] 含 user/text/likes 字段 """ note = self.get_note_detail(note_url) if note: return [c.__dict__ for c in note.comments[:limit]] return [] def batch_get_details(self, notes: List[Note], delay: float = 1.5) -> List[Note]: """ 批量获取笔记详情（在已有列表基础上补充正文和评论）参数: notes: search() 或 get_user_notes() 返回的列表 delay: 每篇请求间隔秒数（建议 1.5~3）返回: 补充了 content/comments 的 Notes 列表 """ enriched = [] for i, note in enumerate(notes): if not note.url: enriched.append(note) continue print(f" [{i+1}/{len(notes)}] {note.title[:30] or note.url[:40]}") detail = self.get_note_detail(note.url) if detail: detail.title = detail.title or note.title detail.likes = detail.likes or note.likes detail.type = detail.type or note.type enriched.append(detail) else: enriched.append(note) time.sleep(delay) return enriched # ─── 工具方法 ─────────────────────────────────────────────── def save(self, notes: List[Note], filepath: str): """保存结果到 JSON 文件""" data = [n.to_dict() for n in notes] with open(filepath, "w", encoding="utf-8") as f: json.dump(data, f, ensure_ascii=False, indent=2) print(f"已保存 {len(data)} 条到 {filepath}") def analyze_account( self, notes: List[Note], creator_name: str = "未知博主", mode: str = "full", save_to: str = None, ) -> str: """ 对已采集的笔记列表做完整分析，返回 Markdown 报告。参数: notes — get_user_notes() 的返回值 creator_name — 博主昵称（用于报告标题） mode — 'full' 完整报告 | 'formula' 选题公式 | 'snapshot' 快速摘要 save_to — 可选，保存报告到文件路径（.md）返回: Markdown 字符串 """ report = _analyze_account(notes, creator_name=creator_name, mode=mode) if save_to: with open(save_to, "w", encoding="utf-8") as f: f.write(report) print(f"报告已保存到 {save_to}") return report def _ensure_session(self): if not self._session: raise RuntimeError("请先调用 open() 或使用 with 语句") FILE:data/archetypes.json { "_meta": { "version": "1.0", "description": "可迭代博主类型注册表 — 随分析案例自动演化", "last_updated": "2026-04-23", "total_analyzed": 0 }, "archetypes": { "A": { "name": "荒诞美学型", "desc": "用荒诞幽默包裹哲学内核，高质感视觉，品牌符号统一", "formula": "荒诞场景 × 品牌符号 × 哲思轻量化 → 审美共鸣", "title_signals": ["（劲爆）", "劲爆", "*", "·"], "content_signals": ["哲", "美学", "荒诞", "飞", "高山", "宇宙", "奖励", "永恒"], "commercial": "高端生活方式、旅行、摄影、轻奢品牌", "difficulty": "高", "examples": [], "confirmed_by": 0 }, "B": { "name": "共鸣命名型", "desc": "把私人经历转化为普世命题，给模糊情绪命名", "formula": "私人经历 → 命题化 → 给模糊状态命名 → 普世共鸣", "title_signals": ["*", "为什么", "是什么", "原来", "才知道", "命题", "灯塔", "解药"], "content_signals": ["成长", "情绪", "留学", "记录", "年末", "不确定", "命题", "感受", "理解", "接纳"], "commercial": "成长/教育/创意平台、中端生活方式", "difficulty": "中", "examples": ["xixiCharon"], "confirmed_by": 1 }, "C": { "name": "现实策略型", "desc": "打破职场/人际/金钱潜规则，提供可执行的向上策略", "formula": "困境 → 说破规则 → 提供策略 → 爽感执行", "title_signals": ["骗子", "不要脸", "装", "穷", "规则", "策略", "大法"], "content_signals": ["策略", "规则", "执行", "调整", "快速", "现实", "向上", "方法"], "commercial": "大众消费、职场、电商、实用工具", "difficulty": "中", "examples": [], "confirmed_by": 0 } }, "custom_archetypes": {} } FILE:data/bloggers.json { "_meta": { "description": "已分析博主数据库 — 每次 /xhsfx 分析后自动写入", "total": 1 }, "bloggers": [ { "creator_name": "xixiCharon", "user_id": "5f94fa23000000000100bced", "analyzed_at": "2026-04-23", "archetype": "B+A", "archetype_name": "共鸣命名型 × 荒诞美学型", "stats": { "avg_likes": 3148, "max_likes": 60000, "sample_size": 50, "viral_count": 3, "video_ratio": "46/50" }, "best_topic": "自然/户外", "best_topic_avg": 9069, "tags": [ "留学", "巴黎", "自然", "情绪成长", "INFJ" ], "commercial": "成长/教育/创意平台、中端生活方式", "formula": "私人经历(留学/山野) → 命题化 → 给模糊情绪命名 → 共鸣" } ] } FILE:examples/note_detail.py """ 示例：抓取单篇笔记详情（正文 + 评论） """ import sys, os sys.path.insert(0, os.path.join(os.path.dirname(__file__), "../..")) from xhscosmoskill import XhsClient NOTE_URL = "https://www.xiaohongshu.com/explore/6912a3c8000000000f03e0a5" with XhsClient() as xhs: note = xhs.get_note_detail(NOTE_URL) if note: print(f"标题: {note.title}") print(f"作者: {note.author}") print(f"赞/藏/评: {note.likes}/{note.collects}/{note.comments_count}") print(f"正文:\n{note.content}") print(f"\n评论 ({len(note.comments)}条):") for c in note.comments[:5]: print(f" @{c.user}: {c.text}") else: print("未能抓取到内容") FILE:examples/search_notes.py """ 示例：搜索关键词，抓取笔记列表 + 详情 """ import sys, os sys.path.insert(0, os.path.join(os.path.dirname(__file__), "../..")) from xhscosmoskill import XhsClient KEYWORD = "咖啡" LIMIT = 20 with XhsClient() as xhs: print(f"搜索: {KEYWORD}") notes = xhs.search(KEYWORD, limit=LIMIT) print(f"找到 {len(notes)} 篇笔记") for i, n in enumerate(notes[:5], 1): print(f"[{i}] {n.title} | 赞:{n.likes} | {n.type}") # 补充详情 print("\n抓取详情...") notes = xhs.batch_get_details(notes, delay=1.5) xhs.save(notes, "search_results.json") FILE:examples/user_profile.py """ 示例：抓取指定用户所有笔记（含详情） """ import sys, os sys.path.insert(0, os.path.join(os.path.dirname(__file__), "../..")) from xhscosmoskill import XhsClient # 井越的用户 ID USER_ID = "5b07eb49e8ac2b5fe1e7de09" with XhsClient(scroll_times=8) as xhs: print(f"抓取用户主页: {USER_ID}") notes = xhs.get_user_notes(USER_ID, limit=100) print(f"列表: {len(notes)} 篇") print("\n抓取每篇详情...") notes = xhs.batch_get_details(notes, delay=2.0) xhs.save(notes, f"user_{USER_ID}.json") print(f"\n完成，正文有内容: {len([n for n in notes if n.content])} 篇") FILE:formula.py """ formula.py — 爆款选题公式生成器基于 hsword 实战案例库提炼的6大模型支持 B型（共鸣命名）/ A型（荒诞美学）/ C型（现实策略）/ 混合型用法： from xhscosmoskill.formula import generate_formula_report md = generate_formula_report(notes, creator_name="xixiCharon", archetype=archetype) """ from __future__ import annotations from typing import List, Any, Dict import re # ── B型（共鸣命名型）6大模型 ────────────────────────────────── B_MODELS = [ { "id": "B1", "name": "场景 × 哲思型", "tag": "均赞最高，优先做", "formula": "[具体地点/场景] + [在这里感受到的哲学状态] + [品牌符号]", "mechanism": "地点宏大 × 感受日常 × 语言诗意 = 三重张力，用户既被画面吸引，又被感受击中", "templates": [ "在[地点]的那几天，[发现了什么]*", "[地点]让我第一次理解，什么叫[哲学状态]*", "人在[场景]里是[动态状态]的，[哲学命题]*", "去完[地点]之后，我开始相信[新的观念]*", "[地点]很[表面感受]，但其实很[真实感受]*", ], "themes": ["山野/户外/自然", "独旅领悟", "城市角落的哲思", "跨文化感受"], }, { "id": "B2", "name": "状态命名型", "tag": "高收藏率，深度粉丝", "formula": "[擅长X的人]，对[Y]的感知是[Z]的 + [品牌符号]", "mechanism": "把混沌状态翻译成可被理解的概念，用户获得「被命名」的满足感", "templates": [ "原来[做X的人]，对[Y]的感知都是[Z的]*", "我终于知道为什么[某类人]总会[某种感受]*", "擅长[X]的人，是怎么感受[时间/当下/变化]的*", "有一类人，[描述]，她们把[X]当成[Y]*", "为什么你总是[状态]？因为你其实是[新定义的人]*", ], "themes": ["记录/创作/表达", "INFJ/i人/高敏感特质", "时间感/当下感", "女性自我认知"], }, { "id": "B3", "name": "阶段宣言型", "tag": "节点内容，年终必出", "formula": "[时间节点/年龄阶段] + [这个阶段的成长判断] + [品牌符号]", "mechanism": "时间节点天然触发情绪浓度，个人叙事 × 普世经历 = 同龄共鸣", "templates": [ "[年份]年终[身份]汇报：[这一年的判断]*", "第[N]个[季节/夏末/生日]，[感受]*", "[岁数]岁的[不确定/清醒/成长]，是[如何]的存在*", "在[X]结束的前一天，我终于理解了[Y]*", "[时间段]里，我做了[事]，发现[领悟]*", ], "themes": ["年末/年初/生日", "留学生活阶段总结", "特定年龄节点", "gap年/毕业/转折"], }, { "id": "B4", "name": "独行宣言型", "tag": "独立女性共鸣", "formula": "[我独自/一个人] + [行动] + [反预期结果]", "mechanism": "独立女性认同 × 行动力展示 × 反预期结果 = 三重共鸣", "templates": [ "我独自[行动]，[结果超出预期的感受]*", "一个人[场景]，其实[反预期的发现]*", "每一个[做X的女孩/人]，都是我*", "[场景下的]一个人，是[新的理解]*", "有些事，只有一个人做了，才能[领悟]*", ], "themes": ["solo旅行/独处", "一个人的山野/城市", "独居生活", "户外/徒步"], }, { "id": "B5", "name": "情绪逆转型", "tag": "治愈系高收藏", "formula": "[消极状态] + 其实是 + [正向领悟] + [品牌符号]", "mechanism": "情绪低谷共鸣 + 逆转出口 = 「被治愈的可能性」，用户收藏的是出口本身", "templates": [ "[消极状态]，其实是[正向新解]*", "[看似坏事]的那段时间，反而让我[领悟]*", "[不确定/失去/放弃]，是像[比喻]一样的存在*", "当[困境]来的时候，我发现[反预期的事]*", "长出[新特质]，是从[某个经历]开始的*", ], "themes": ["情绪低谷/调整", "不确定阶段", "成长的代价", "留学/独居压力"], }, { "id": "B6", "name": "世界观输出型", "tag": "最强粘性，深度内容", "formula": "[持续做X的人/坚持Y的意义] + [是如何理解Z的] + [品牌符号]", "mechanism": "用户收藏的是「世界观」，会持续追更，是最强粘性内容", "templates": [ "[持续做X]的人，是在[更大的事情]上做投资*", "出发/记录/独处的意义，不是[表面]，而是[深层]*", "[这件事]，其实是[更大命题]的另一种表达*", "生活的本质，是[新定义]*", "我们[做X]，是为了[更深的东西]*", ], "themes": ["记录的意义", "出发/旅行的意义", "独立/自我的定义", "女性成长哲学"], }, ] # ── A型（荒诞美学型）公式 ───────────────────────────────────── A_MODELS = [ { "id": "A1", "name": "地点 + 反差动作/感受", "formula": "[地点的宏大] × [动作的日常] = 反差萌", "templates": ["在[高原/远方]做[日常小事]！（品牌符号）", "去[地方]的[超日常行为]！"], }, { "id": "A2", "name": "场景 + 情感类比", "formula": "[地点] × [文学化情感命名] = 高级感", "templates": ["[地点]有[文学类比]般的[时间/氛围]（品牌符号）", "这跟走进[情感类比]有什么区别！"], }, { "id": "A3", "name": "荒诞日常", "formula": "[荒唐问题/行为] × [认真叙述] = 喜剧感", "templates": ["大家都是怎么忍住不[荒诞行为]的（品牌符号）", "我在[地方]必须[荒诞行为]"], }, { "id": "A4", "name": "轻哲思命题", "formula": "[严肃命题] × [轻盈表达] = 有深度但不沉重", "templates": ["关于[严肃话题]的奇怪故事（品牌符号）", "什么是[抽象命题]（品牌符号）"], }, ] # ── C型（现实策略型）公式 ───────────────────────────────────── C_MODELS = [ { "id": "C1", "name": "困境说破型", "formula": "困境 → 说破 → 规则 → 策略 → 爽感", "templates": ["为什么[Y]都喜欢[X]？", "普通女孩怎么在[X]里拿回主动权？"], }, { "id": "C2", "name": "信息差揭秘型", "formula": "[表面现象] + [背后规则] + [可执行策略]", "templates": ["对普通人来说，[X]比努力更重要", "没有[资源]的人，怎么[目标]？"], }, ] def _parse_likes(v) -> int: if not v: return 0 s = str(v).replace(",", "").strip() if "万" in s: return int(float(s.replace("万", "")) * 10000) try: return int(s) except ValueError: return 0 def _get_top_notes(notes: List[Any], n: int = 5): def get_title(n): return n.title if hasattr(n, "title") else n.get("title", "") def get_likes(n): return _parse_likes(n.likes if hasattr(n, "likes") else n.get("likes")) return sorted(notes, key=lambda n: -get_likes(n))[:n] def _build_30_topics(creator_name: str, archetype_type: str) -> str: """生成30个可用选题方向""" typ = archetype_type[0] if archetype_type else "B" if typ == "B": return f"""### A 组：自然/户外类（均赞最高，优先做） 1. 在[山/海/草原]的那几天，[感受到了什么]* 2. [某种自然景象]，让我第一次理解什么叫[情绪状态]* 3. 去[目的地]之前我以为[X]，回来之后我只记得[Y]* 4. 一个人徒步，[具体发现]* 5. 冬天的[地点]，有[文学化情感命名]的质地* 6. 到[自然场景]里走走，才发现[领悟]* ### B 组：留学/海外类（基础盘，稳定输出） 7. 在巴黎的第[N]年，我终于理解了[某事]* 8. 留学以后，[某种改变]* 9. 一个在海外的中国女生，怎么理解[某个普世命题]* 10. 巴黎教会我的，不是[表面]，而是[真正的东西]* 11. [年份]年的留学，[年度总结词]* 12. 在异乡过[节日/特殊时刻]，[感受]* ### C 组：记录/创作类（高收藏，深度粉丝） 13. 为什么要频繁记录？[自己的答案]* 14. 擅长[记录/写作/拍照]的人，[是什么样的]* 15. [某件事]让我第一次真的开始记录* 16. 记录的意义，不是留住过去，而是[新定义]* 17. vlog 教会我的最重要一件事：[命题]* 18. 如果你也是创作者，关于[某个困惑]的答案* ### D 组：成长/情绪类（共鸣最广，拓展受众） 19. 二十[几]岁的[某种状态]，是像[比喻]一样的存在* 20. 那段[困难]时期，让我长出了[新特质]* 21. [某类人]是如何对待自己的消极情绪的* 22. 当你开始接受[某件事]，会发现[领悟]* 23. 我不再[某种行为]以后，[变化]* 24. 给所有正在[某个阶段]的人：[一句话]* ### E 组：哲学/世界观类（最高传播潜力） 25. 生活的本质，是[你的定义]* 26. 持续[做某事]，是在[更大意义上]做什么* 27. 为什么[出发/记录/独处]？因为[真正的答案]* 28. [某种看起来无用的事]，其实是[意外的价值]* 29. 我们出发，不是为了[表面]，而是为了[深层]* 30. 一个人能给自己最好的礼物，是[你的答案]*""" elif typ == "A": return """（A型：荒诞美学型30个选题方向）基础模式：地点反差 × 荒诞日常 × 轻哲思 × 品牌符号统一 1-10：地点 + 反差动作系列 11-20：场景 + 情感类比系列 21-30：荒诞命题 + 认真叙述系列具体选题请根据博主真实经历填入[地点]和[动作]。""" else: # C型 return """（C型：现实策略型30个选题方向）基础模式：困境说破 × 信息差揭秘 × 可执行策略 1-10：职场潜规则系列 11-20：消费/财务博弈系列 21-30：普通人向上移动系列""" def generate_formula_report( notes: List[Any], creator_name: str, archetype: Dict, ) -> str: """ 生成爆款选题公式 Markdown 报告参数： notes — get_user_notes() 返回值 creator_name — 博主昵称 archetype — classify_archetype() 返回值返回：Markdown 字符串 """ typ = archetype.get("type", "B") name = archetype.get("name", "共鸣命名型") desc = archetype.get("desc", "") top5 = _get_top_notes(notes, 5) def get_title(n): return n.title if hasattr(n, "title") else n.get("title", "") def get_likes(n): return _parse_likes(n.likes if hasattr(n, "likes") else n.get("likes")) top5_md = "\n".join( f"- 👍{get_likes(n):,} {get_title(n)[:50]}" for n in top5 ) # 选择对应的模型 models = B_MODELS if typ[0] == "B" else (A_MODELS if typ[0] == "A" else C_MODELS) models_md = "" for m in models: templates = "\n".join(f" - {t}" for t in m.get("templates", [])) themes_str = "、".join(m.get("themes", [])) models_md += f""" ### 模型 {m['id']}：{m['name']} {m.get('tag','')} **公式：** {m['formula']} **为什么有效：** {m.get('mechanism', '')} **可套用句式：** {templates} **适合主题：** {themes_str} --- """ topics_30 = _build_30_topics(creator_name, typ) return f"""# {creator_name} — 爆款选题公式 > 说明：本文件不是复刻 {creator_name} 的表面语气，而是提炼她真正有效的**选题逻辑、标题机制、传播结构和可迁移方法**。 --- ## 一、先说结论：为什么能做出爆款？ **账号类型：** {name}（{typ}） **核心内核：** {desc} 爆款公式： > **私人场景 × 命题化 × 诗意命名 × 品牌符号 × 普世共鸣** **高表现代表作：** {top5_md} --- ## 二、总公式 ``` 一个真实经历 / 一个具体场景 / 一个模糊的情绪状态 ↓ 提炼成普遍的人生命题 ↓ 用诗意语言给这个状态一个更精准的说法 ↓ 品牌符号收尾（如 *） ↓ 用户：就是这个感觉！ ``` --- ## 三、最常见的{len(models)}类爆款选题模型 {models_md} --- ## 四、隐形公式（最值钱的底层打法） **隐形公式1：先给感受，再给命题** 不要用命题开头，先描述具体画面，再让命题从画面里生长出来。 - 错误：「论记录的意义」 - 正确：「那天我在山里，突然理解了为什么要记录」* **隐形公式2：动词比名词强** 「流动」比「自由」有力；「后置」比「延迟」有力。动词给人画面，名词只给人概念。 **隐形公式3：品牌符号的系统性** 品牌符号（如 `*`）不是随便加的，它是人格锚点。长期累积会形成"一看就知道是谁"的识别度。 **隐形公式4：自然场景是放大器** 自然场景天然携带"宏大感"，能让任何情绪命题被放大10倍。 --- ## 五、如何迁移到自己的账号 **最优学习路径：** ``` 你的真实场景（不必模仿博主的具体场景） ↓ 找到那个你"说不清楚"的感受 ↓ 用一句话把它命名（用动词，不用形容词） ↓ 写成：[场景] + [命题] + [你的标识符] ↓ 发布，看哪句话在评论区被用户引用 ``` **标题模板直接套用：** 1. 在[地点]的时候，我第一次理解了什么叫________[你的标识符] 2. [做X]的人，对[Y]的感知是________的[标识符] 3. 我独自[行动]，[反预期的发现][标识符] 4. 原来[做某事]，是一种[新的定义][标识符] 5. 第[N]个[时间节点]，[这个阶段的感受][标识符] 6. 终于知道了，为什么[某类人]总会[某种感受][标识符] --- ## 六、可直接复用的30个选题方向 {topics_30} --- ## 七、哪些能学，哪些不能硬抄 **能学的：** - 命题化思维：私人经历 → 公共命题 - 动词命名法：用动词描述情绪 - 场景 × 哲思的意象对冲结构 - 品牌符号系统（找到你自己的"印章"） **不能硬抄的：** - 语气温度（来自真实人设，非技巧） - 具体生活场景（需要真实经历支撑） - 品牌符号本身（需先建立人格认同） --- ## 八、最终总结 **一句话：** {creator_name} 的爆款公式，核心只有一句话： > **把说不清楚的感受，用诗意的语言说清楚。** --- *基于 xhscosmoskill + hsword框架提炼 · {creator_name} · 2026-04-23* """ FILE:models.py from dataclasses import dataclass, field from typing import List, Optional @dataclass class Comment: user: str = "" text: str = "" likes: str = "" time: str = "" @dataclass class Note: id: str = "" title: str = "" content: str = "" url: str = "" author: str = "" author_id: str = "" likes: str = "" collects: str = "" comments_count: str = "" type: str = "" # "video" | "normal" images: List[str] = field(default_factory=list) tags: List[str] = field(default_factory=list) comments: List[Comment] = field(default_factory=list) raw: dict = field(default_factory=dict) def to_dict(self): return { "id": self.id, "title": self.title, "content": self.content, "url": self.url, "author": self.author, "author_id": self.author_id, "likes": self.likes, "collects": self.collects, "comments_count": self.comments_count, "type": self.type, "images": self.images, "tags": self.tags, "comments": [c.__dict__ for c in self.comments], } @dataclass class UserProfile: user_id: str = "" nickname: str = "" description: str = "" followers: str = "" following: str = "" likes_collected: str = "" notes_count: str = "" avatar: str = "" FILE:package.json { "name": "xhsfenxi-pro", "version": "2.0.0", "description": "小红书博主全链路深度分析：数据采集 + 三型分类 + 内核三段论 + 爆款选题公式", "scripts": {}, "keywords": [ "xiaohongshu", "xhsfenxi", "博主分析", "爆款选题" ], "author": "Cosmofang", "license": "MIT" } FILE:parser.py """ 解析器：从拦截到的 API JSON 或 DOM 中提取结构化数据兼容两种来源：API JSON（优先）和 DOM fallback """ from __future__ import annotations from typing import List, Optional from .models import Note, Comment, UserProfile def parse_note_from_api(raw: dict) -> Note: """从 API JSON 解析笔记""" note_card = raw.get("note_card") or raw basic = note_card.get("basic_info") or {} interact = note_card.get("interact_info") or {} user = note_card.get("user") or {} note = Note() note.id = ( raw.get("id") or note_card.get("note_id") or basic.get("note_id", "") ) note.title = ( note_card.get("display_title") or note_card.get("title") or basic.get("title", "") ) note.content = note_card.get("desc") or basic.get("desc", "") note.author = user.get("nickname", "") note.author_id = user.get("user_id", "") note.likes = str(interact.get("liked_count", "")) note.collects = str(interact.get("collected_count", "")) note.comments_count = str(interact.get("comment_count", "")) note.type = "video" if note_card.get("type") == "video" else "normal" # 图片 image_list = note_card.get("image_list") or [] note.images = [img.get("url", "") for img in image_list if img.get("url")] # 标签 tag_list = note_card.get("tag_list") or [] note.tags = [t.get("name", "") for t in tag_list if t.get("name")] note.raw = raw return note def parse_search_response(api_cache: dict) -> List[Note]: """从拦截的搜索 API 响应中提取笔记列表""" notes = [] for url, data in api_cache.items(): if "search" not in url: continue items = ( data.get("data", {}).get("items") or data.get("items") or [] ) for item in items: note_raw = item.get("note_card") or item if not note_raw: continue n = parse_note_from_api({"id": item.get("id", ""), "note_card": note_raw}) xsec = item.get("xsec_token", "") note_id = item.get("id") or note_raw.get("note_id") or n.id if note_id: n.url = f"https://www.xiaohongshu.com/explore/{note_id}?xsec_token={xsec}&xsec_source=pc_search" notes.append(n) return notes def parse_user_notes_response(api_cache: dict) -> List[Note]: """从拦截的用户主页 API 响应中提取笔记列表""" notes = [] for url, data in api_cache.items(): if "user_posted" not in url and "user/notes" not in url: continue items = ( data.get("data", {}).get("notes") or data.get("notes") or [] ) for item in items: n = parse_note_from_api(item) note_id = n.id or item.get("note_id", "") if note_id: n.url = f"https://www.xiaohongshu.com/explore/{note_id}" notes.append(n) return notes def parse_note_detail_response(api_cache: dict) -> Optional[Note]: """从拦截的笔记详情 API 响应中提取单篇笔记""" for url, data in api_cache.items(): if "feed" not in url and "note/detail" not in url: continue items = data.get("data", {}).get("items") or [] for item in items: note_card = item.get("note_card") or {} if note_card: n = parse_note_from_api({"note_card": note_card}) # 评论 comments_data = data.get("data", {}).get("comments") or [] for c in comments_data: comment = Comment( user=c.get("user_info", {}).get("nickname", ""), text=c.get("content", ""), likes=str(c.get("like_count", "")), time=c.get("create_time", ""), ) n.comments.append(comment) return n return None def parse_comment_page_response(api_cache: dict) -> List[Comment]: """从 v2/comment/page API 解析评论列表""" comments = [] for url, data in api_cache.items(): if "comment" not in url: continue items = ( data.get("data", {}).get("comments") or data.get("comments") or [] ) for c in items: user_info = c.get("user_info") or {} comment = Comment( user=user_info.get("nickname", ""), text=c.get("content", ""), likes=str(c.get("like_count", "")), time=str(c.get("create_time", "")), ) comments.append(comment) return comments def parse_dom_notes(browser) -> List[Note]: """DOM fallback：从页面元素中提取笔记（API 拦截失败时使用）""" notes = [] selectors = ["section.note-item", ".note-item", "[class*='note-item']"] items = [] for sel in selectors: items = browser.find_elements(sel) if items: break for item in items: note = Note() for sel in ["a.title span", ".title span", "span.title"]: try: note.title = item.find_element("css selector", sel).text.strip() break except Exception: pass try: note.url = item.find_element("css selector", "a.cover").get_attribute("href") except Exception: pass try: note.likes = item.find_element("css selector", "span.count").text.strip() except Exception: pass try: item.find_element("css selector", ".play-icon") note.type = "video" except Exception: note.type = "normal" if note.title or note.url: notes.append(note) return notes FILE:references/hsword-frameworks.md # hsword 框架参考手册 > 来源：openclaw_cosmo/afa/hsword/ 实战案例库 > 版本：2026-04-23 --- ## 一、内核三段论（每次分析必做） ### 框架结构 ``` 外壳是什么？ → 表面看起来像什么博主（用户最先看到的标签）真正的内核是什么？ → 一句话精炼，带""引号 → 格式：这是一个通过"X、Y、Z"来建立影响力的[定语]博主三层人设结构：表层标签 → 地点/身份/平台标签（可模仿）中层特质 → 性格/能力/气质（可包装）深层价值观 → 鼓励什么/认可什么/传递什么（必须靠长期一致才能被相信） ``` ### 已验证案例 | 博主 | 外壳 | 真正内核 | 深层价值观 | |------|------|---------|-----------| | 井越（A型）| 旅行vlogger / 文艺青年 | "荒诞滤镜下的生活哲学家" | 用奇怪但有质感的视角，记录在世界上存在过的痕迹 | | 橘一橙（B型）| 旅行/成长博主 | "把私人经历转为普世命题的命名者" | 让年轻人获得"原来不止我这样"的理解 | | 陈浪浪（C型）| 大厂博主/成长博主 | "普通女孩现实破局教练型博主" | 普通女性如何在不体面的现实里最大化利用手上的牌 | | xixiCharon（B+A）| 留学vlogger | "把说不清楚的感受用诗意语言说清楚的命名者" | 鼓励女性关注自己内心，认可流动与不确定性 | --- ## 二、三型博主爆款逻辑对比 | 维度 | A型（荒诞美学）| B型（共鸣命名）| C型（现实策略）| |------|------------|------------|------------| | 爆款驱动 | 审美震撼 + 品牌符号 | 被命名的满足感 | 说破现实的爽感 | | 标题风格 | 反差张力（宏大×日常）| 命题感+隐喻+判断 | 冲突词+反常识 | | 用户获得感 | 愉悦+审美+新视角 | 被理解+被命名 | 被激活+被赋能 | | 品牌符号 | "（劲爆）"统一后缀 | `*` 情感印章 | 自我命名先夺话语权 | | 商业适配 | 高端/旅行/轻奢 | 成长/教育/中端生活 | 大众消费/职场/电商 | | 复制难度 | 高（视觉审美积累）| 中（方法可学）| 中（框架可学）| --- ## 三、爆款选题公式体系（6大模型） ### B型（共鸣命名型）——xixiCharon 实证 #### 模型1：场景 × 哲思型 ``` 公式：[地点] + [哲学感悟] + [品牌符号] 机制：地点宏大 × 感受日常 × 语言诗意 = 三重张力实例：人在山里是流动的，不存在于过去也不在未来*（6万赞） ``` **可套用句式：** - 在[地点]的那几天，[发现了什么]* - [地点]让我第一次理解，什么叫[哲学状态]* - 人在[场景]里是[动态状态]的，[哲学命题]* #### 模型2：状态命名型 ``` 公式：[擅长X的人]，对[Y]的感知是[Z]的 + [品牌符号] 机制：把混沌状态翻译成可被理解的概念实例：擅长记录的人对当下的感知是后置的*（10万+） ``` **可套用句式：** - 原来[做X的人]，对[Y]的感知都是[Z的]* - 擅长[X]的人，是怎么感受[时间/当下/变化]的* - 有一类人，[描述]，她们把[X]当成[Y]* #### 模型3：阶段宣言型 ``` 公式：[时间节点] + [成长判断] + [品牌符号] 机制：时间节点触发情绪 × 个人叙事 × 同龄共鸣实例：2025年终留学汇报：今年似乎迈了一大步*（1.9万） ``` #### 模型4：独行宣言型 ``` 公式：[我独自/一个人] + [行动] + [反预期结果] 机制：独立女性认同 × 行动力 × 反预期实例：我独自出发，先斩后奏，斩了也不奏*（8813赞） ``` #### 模型5：情绪逆转型 ``` 公式：[消极状态] + 其实是 + [正向领悟] + [品牌符号] 机制：情绪低谷共鸣 + 逆转出口实例：二十岁的不确定是像灯塔一样的存在*（2999赞） ``` #### 模型6：世界观输出型 ``` 公式：[持续做X] + 是在[更大意义上]做什么 + [品牌符号] 机制：用户收藏"世界观"，最强粘性内容实例：记录和表达如此重要，于是有了写作和vlog*（3.2万） ``` ### A型（荒诞美学型）——井越实证 #### 公式1：地点 + 反差动作/感受 ``` 机制：地点的宏大 × 动作的日常 = 反差萌实例：在帕米尔高原摘豌豆！（劲爆）（5151赞） ``` #### 公式2：场景 + 情感类比 ``` 机制：地点 × 文学化情感命名 = 高级感实例：北海道有爵士乐般的冬日（劲爆）（912赞） ``` #### 公式3：荒诞日常 ``` 机制：荒唐问题/行为 × 认真叙述 = 喜剧感实例：大家都是怎么忍住不亲马鼻孔的（劲爆） ``` ### C型（现实策略型）——陈浪浪实证 #### 核心公式 ``` 困境 → 说破 → 规则 → 策略 → 爽感机制：先夺走羞耻感 → 说破现实 → 给出可执行策略 ``` --- ## 四、综合版"双系统选题公式" ``` 系统A（共鸣命名型）：经历 → 命题 → 命名 → 判断 → 共鸣系统B（现实策略型）：困境 → 说破 → 规则 → 策略 → 爽感混合公式（最强）：真实困境 + 可命名的阶段/处境 + 说破现实的判断 + 可执行动作建议 = 既被理解，又能行动 ``` --- ## 五、标题公式快查表（hsword综合版） | 公式 | 模板 | 适用类型 | |------|------|---------| | 为什么总像 | 为什么[X]总像[Y]？ | B型共鸣 | | 帮大家试过了 | 帮大家试过了，[X]到底值不值得 | B型验证 | | 为什么都喜欢 | 为什么[Y]都喜欢[X]？ | C型说破规则 | | 普通人/普女怎么 | 普通女孩怎么在[X]里拿回主动权？ | C型策略 | | 当一个人…会发生什么 | 当一个人不再[X]，会发生什么？ | B+C混合 | | 原来不是而是 | 原来[X]不是[Y]，而是[Z] | B型重新定义 | --- ## 六、可借鉴 vs 不能盲抄 ### 所有类型通用 | 可借鉴 | 不能盲抄 | |-------|---------| | 选题的命题感 | 表面的生活样本/场景 | | 标题的气质感 | 语气温度和个人语言习惯 | | 内容的人格统一性 | 品牌符号本身（需先建立人格） | | 把个人经历做成公共内容 | 没有真实经历的虚构叙述 | ### 可迁移内容模板（通用） ``` [你的真实场景] ↓ [找到说不清楚的感受] ↓ [用动词而不是形容词命名] ↓ [写成：场景 + 命题 + 你的标识符] ↓ [发布，看哪句话在评论区被引用] ``` --- ## 七、参考数据来源 - **新榜：** `https://www.newrank.cn/profile/xiaohongshu/{user_id}` - **第三方公开站：** TooBigData、热浪数据 - **小红书搜索页：** 可见搜索结果中的账号卡片数据（A1级） - **公开访谈/播客：** 知乎、澎湃、凤凰等（B1级） --- ## 八、分析边界声明模板 ``` 本报告基于公开网页可见信息，数据级别：A1（主页可见）/ B1（第三方公开）/ C1（综合推断）。未覆盖：全量历史笔记 / 评论区文本 / 收藏率完播率后台数据 / 商单实际转化数据。适合用途：账号定位 / 内容策略学习 / 竞品研究 / 对标选题学习。不适合：财务审计 / 精确数据核算。 ``` FILE:references/templates.md # Templates Reference — xhsfenxi v2.0 --- ## Output Files Always produce as `.docx` directly. Three standard files: | File | When | |------|------| | `账号名-结构化总结报告.docx` | Single creator analysis | | `账号名-爆款选题公式.docx` | Single creator topic formula | | `选题公式学习-综合版.docx` | Multi-creator comparison | --- ## Structured Report Skeleton (v2 — Archetype-Aware) ```markdown # 账号名结构化总结报告 > 说明：本报告基于[数据来源说明]。适合用于[账号定位/内容策略/竞品研究/爆款机制提炼]。 ## 一、执行摘要 [2–4段: 博主是谁，核心特征，最值得记住的一句话] ## 二、账号基础信息 [表格: 昵称/小红书号/粉丝/获赞收藏/笔记数/内容类型/账号阶段] [补充: 全平台历史数据（如有）] ## 三、博主背景（公开资料） [人物背景、职业起点、创作理念（来源于公开访谈）] ## 四、账号类型判断 **类型:** Type A/B/C 或混合型 **判断依据:** [2–3条可见信号] **核心内核:** [一句话总结] ## 五、博主核心内核拆解 [外壳是什么 / 真正的内核是什么 / 人设三层结构 / 最强能力] ## 六、内容结构拆解 [内容支柱分布（表格）/ 内容类型比例 / 统一主题] ## 七、标题机制与爆点分析 [最重要特征 / 品牌符号分析 / 高热内容分析（表格）/ 标题公式归纳] ## 八、与其他博主的横向对比（可选） [对比表格: 多维度 × 多博主] ## 九、商业化判断 [当前阶段 / 为什么品牌会找他 / 适合的合作方向 / 不适合的方向] ## 十、最值得学习的核心能力 [3–5个能力，每个含: 说明 + 可学习点] ## 十一、后续建议 [路径A: 深入单篇分析 / 路径B: 制作爆款选题公式 / 路径C: 跨博主综合版] ## 十二、最终总结 [1–2段核心结论] ## 附录：笔记完整列表（可选） [表格: 排名 / 类型 / 标题 / 点赞] ``` --- ## Viral Topic Formula Skeleton (v2) ```markdown # 账号名爆款选题公式 > 说明：本文件不是复刻[账号名]的表面语气，而是提炼真正有效的选题逻辑、标题机制、传播结构、人设打法与可迁移方法。 ## 一、先说结论：为什么能做出爆款？ [账号类型 + 5个叠在一起的核心能力 + 总公式一句话] ## 二、总爆款公式 [5步骤公式 + 压缩版关键词] ## 三、最常见的 3–6 类爆款选题模型 [每个模型: 公式 / 典型表达方向 / 为什么有效 / 可套用句式 / 适合主题] ## 四、标题公式：可以直接拆用 [5–6个标题公式，每个含: 结构说明 + 套用模板] ## 五、内容结构公式 [2–3个结构公式，每个含: 步骤 / 适合做的主题] ## 六、隐形公式 / 最值钱的底层打法 [3–4个] ## 七、如何迁移到自己的内容里 [4个步骤: 找现实母题 / 改写生活故事 / 用冲突词 / 4问检验] ## 八、可直接复用的 20 个选题方向 [编号列表] ## 九、哪些能学，哪些不能硬抄 [可以学的 / 不能硬抄的] ## 十、最终总结 [一句话核心结论] ``` --- ## Comparison Skeleton (v2 — Archetype Map) ```markdown # 选题公式学习（综合版） ## ——结合[博主A]与[博主B]（以及[博主C]）的内容方法论 > 说明：这份文档不是简单拼接报告，而是把各博主最有效的内容能力抽取出来，重新整理成可学习、可迁移、可直接应用的选题方法。 ## 一、为什么把这几位博主放在一起学？ [各博主关键词 / 各博主强项一句话 / 为什么组合更完整] ## 二、账号类型对照表 [表格: 博主 / 类型 / 核心方向 / 用户获得感 / 内容气质 / 标题风格 / 爆点来源] ## 三、共同底层：真正的选题不是题材，而是命题 [5个共同点 / 核心结论] ## 四、双系统（或多系统）选题公式 [系统A（共鸣命名型）: 适用场景 / 核心公式 / 压缩版 / 标题类型] [系统B（现实策略型）: 同上] [（如有）系统C（荒诞美学型）: 同上] ## 五、混合公式 [混合公式定义 + 例子 + 为什么更强] ## 六、标题公式综合版 [6–8个跨类型标题公式] ## 七、内容结构综合版 [结构A / B / C，各含步骤和用户获得感] ## 八、可直接复用的 30 个综合选题方向 [A组（偏共鸣命名）/ B组（偏现实策略）/ C组（偏荒诞美学）/ D组（混合）] ## 九、结论与行动建议 [最重要的学习提醒 + 最终公式] ``` --- ## Archetype Comparison Table Template Use this table when comparing 2–3+ accounts: ```markdown | 维度 | [博主A] | [博主B] | [博主C] | |---|---|---|---| | 账号类型 | Type A 荒诞美学 | Type B 共鸣命名 | Type C 现实策略 | | 核心方向 | | | | | 用户获得感 | | | | | 内容气质 | | | | | 标题风格 | | | | | 品牌符号 | | | | | 爆点来源 | | | | | 商业化方向 | | | | | 可复制难度 | | | | ``` --- ## Evidence Label Reference Use these labels in your internal working notes (do not need to show all in final report): | Label | Meaning | |-------|---------| | `A1` | Xiaohongshu homepage visible fact | | `A2` | Xiaohongshu screenshot fact | | `B1` | Third-party public clue (interview, news, etc.) | | `C1` | Synthesis / interpretation | --- ## Business Word Checklist When building a business-style Word file: - [ ] Cover page - [ ] TOC with `_TocH1_XXX` bookmarks (internal jump, not external link) - [ ] Page numbers - [ ] Conclusion page - [ ] Restrained typography (no loud decorative color blocks unless requested) - [ ] Tested in Word: TOC entries jump correctly - [ ] No duplicate headings accidentally inserted in body --- ## Brand Symbol Analysis Template For Type A accounts or any account with a unified recurring symbol: ```markdown ### 品牌符号分析 **符号:** [（劲爆）/ 其他] **出现频率:** [几乎所有 / 大部分 / 部分] **品牌效果:** - 识别度: [描述] - 反差机制: [描述] - 先发制人效果: [描述] **可学习点:** [其他账号如何建立自己的品牌符号] ``` FILE:references/workflow-xhsfenxi-v2.md # Workflow Reference — xhsfenxi v2.0 --- ## Source Ladder Use sources in this order whenever possible: ### Level A — Primary public evidence Highest trust for visible claims: - Xiaohongshu search result pages - creator homepage/profile pages - user-provided screenshots from Xiaohongshu - user-provided post links that are actually accessible What you can safely claim from Level A: - visible follower / note counts shown on page - visible account ID / self-introduction - visible recent titles / visible public engagement numbers - posting cadence clues visible on page - obvious content categories - brand symbols and signature phrases (like "（劲爆）") ### Level B — Secondary public clues Use as support, never as backend truth: - podcast interviews - encyclopedia entries - public creator bios elsewhere - analytics websites / public creator databases - public brand collaboration listings What Level B is good for: - filling identity / background context - confirming repeated themes - spotting public commercialization clues - triangulating a bigger narrative ### Level C — Synthesis / inference Allowed only when clearly framed as interpretation. Examples: - "This account behaves more like a Type A 荒诞美学 creator than a Type B 共鸣命名 creator." - "The recurring topic logic appears to be: contrast × literary naming × unified symbol." - "The transferable lesson is not the tone, but the framing system." Never present Level C synthesis as raw observed fact. --- ## Standard Execution Runbook ### Step 1 — Confirm the deliverable Choose one or more: - structured report - viral topic formula - archetype classification - comparison report - customized learning report - Word / business version ### Step 2 — Capture the minimum task brief Minimum useful brief: - creator name - available links or screenshots - target output format - whether a business Word version is required ### Step 3 — Archetype pre-classification Before deep analysis, make a preliminary archetype call using visible signals: | Signal | Likely archetype | |--------|-----------------| | Unified brand symbol/tag on every post | Type A 荒诞美学型 | | High-production video, absurdist or philosophical content | Type A 荒诞美学型 | | Posts name vague emotions or life stages | Type B 共鸣命名型 | | Titles feel like thoughtful questions or redefinitions | Type B 共鸣命名型 | | Titles contain conflict words ("骗子", "不要脸", "装") | Type C 现实策略型 | | Content breaks unspoken workplace/relationship/money rules | Type C 现实策略型 | | Mix of resonance + strategy | Mixed B+C | ### Step 4 — Research in layers 1. Find the account and public homepage/search data 2. Save visible account facts (follower count, note count, visible titles, engagement) 3. Pull representative visible titles — note brand symbols, title patterns 4. Add external public clues only where helpful 5. Record limitations before interpretation ### Step 5 — Extract the account system (Five Layers) Always map these five layers: 1. **Identity** — who the creator is framed as; any self-labeling strategy 2. **Audience contract** — why people follow them; what emotional/strategic need is met 3. **Topic system** — what recurring problems/desires they cover; 3–6 topic models 4. **Expression system** — title formulas, brand symbols, opening patterns 5. **Transferability** — what another account can actually learn; what must not be copied ### Step 6 — Produce the right file #### Structured report Recommended top-level sections: 1. Account snapshot 2. Archetype classification + rationale 3. Positioning judgment 4. Audience and demand 5. Content pillars 6. Title / hook system (including brand symbol analysis) 7. Narrative structure 8. Commercialization clues 9. What to learn 10. What not to copy 11. Conclusion 12. (Optional) Full note list sorted by engagement #### Viral topic formula Recommended top-level sections: 1. Why this creator produces strong topics (archetype-based) 2. Total formula 3. 3–6 recurring topic models 4. Title formulas 5. Body structure formulas 6. Distribution / comment trigger clues 7. How to migrate it to another account 8. 10–30 topic directions #### Comparison report (2–3+ accounts) Recommended top-level sections: 1. Why compare these creators together 2. Archetype map (each creator's type) 3. Shared foundations 4. Key differences 5. Hybrid formula 6. Which path fits which use case 7. Suggested topic directions 8. Final recommendation --- ## Archetype-Specific Analysis Lenses ### Type A — 荒诞美学型 Focus questions: - What is the brand symbol? (unified tag, phrase, or visual marker) - How does the title create contrast? (grand × mundane, serious × absurd) - Where does the humor come from? (earnest treatment of absurd things) - What is the underlying philosophical/emotional core? - Why is replication hard? (visual sensibility, accumulated aesthetic) Key learning for other accounts: - Find your own "brand symbol" — a unifying tag that creates recognition - Train "contrast naming" — don't describe the place, name the feeling - Practice "absurdist framing" — pick topics that are ridiculous when taken seriously --- ### Type B — 共鸣命名型 Focus questions: - What life stages or emotional states does the account name? - How does private experience become universal proposition? - What are the signature conceptual phrases? - What is the "new way to understand yourself" users get? Key learning for other accounts: - Practice "命题化" — reframe personal experience as a universal human situation - Build a personal vocabulary of named states ("Odyssey时期", etc.) - The formula: experience → proposition → naming → judgment → resonance --- ### Type C — 现实策略型 Focus questions: - What real-world rules does the account break open? - How does the self-labeling strategy work? (antifragile identity) - What is the conflict word in each title? - What is the "can-do action" users walk away with? Key learning for other accounts: - Find your "reality母题" — the real困境 you speak to - Practice rewriting weak framings into strong rule-breaking framings - The formula: 困境 → 说破 → 规则 → 策略 → 爽感 --- ## Limitation Language Use direct wording like: - "This round is based primarily on public homepage/search-page evidence." - "Single-post detail pages were not stably accessible due to platform risk control." - "Third-party figures are used as directional clues, not audited backend data." - "Archetype classification is based on visible patterns; the creator may operate differently at the full content level." --- ## Deep-Dive Upgrade Path If the user wants finer analysis, ask for 3–10 representative posts, ideally with: - screenshots of cover + title - screenshots of body pages / subtitles - comments screenshots - transcript or summary if video-based Then upgrade from account-level to post-level analysis. --- ## Word Output Workflow ### Overview Always generate Markdown first, then convert to Word. Iterating Markdown is cheaper than re-generating DOCX. ### Script Selection Guide Use the workflow archive scripts at: ``` openclaw_cosmo/afa/小红书分析与工作流归档/02-Word生成与目录修复脚本/ ``` | Script | Use when | |--------|----------| | `build_docx6.py` | Default — most stable, use this first | | `build_docx.py` → `build_docx5.py` | Earlier iterations, kept for reference/debugging | | `inspect_docx.py` | Inspect internal XML, bookmarks, link structure | | `fix_final2.py` | Final TOC / attribute fix after build | | `fix_attrs.py` | Fix XML namespace attribute writing errors | | `fix_wrong.py` | Targeted corrections for known wrong output | | `check_it.py` | Verify output correctness | ### Word TOC Fix Checklist When a generated .docx has broken TOC links: 1. Run `inspect_docx.py` — check `word/document.xml` for bookmark names and link types 2. Verify heading bookmarks use format `_TocH1_XXX` (not `#_TocH1_XXX`) 3. Verify internal links use Word-native anchor style, not external URL rels 4. If `elem.set()` errors appear, use full namespace: `elem.set('{' + W_NS + '}id', value)` 5. Run `fix_final2.py` after any XML edits 6. Test by opening in Word — click TOC entries to verify jump ### Common Errors | Error | Cause | Fix | |-------|-------|-----| | Bookmark name wrong format | Missing `_TocH1_` prefix | Rename via `fix_attrs.py` | | `anchor` contains `#` | Wrong anchor format | Strip `#` prefix | | TOC uses external link rels | Used `http://` link type for internal jump | Switch to `w:instr` field or anchor | | Duplicate headings in body | TOC headings accidentally inserted into body | Use `fix_wrong.py` to clean | | Namespace attribute error | Missing full namespace URI in `elem.set()` | Use `'{' + W_NS + '}id'` form | ### Best Practice - Keep only `build_docx6.py` as the production script - Keep all earlier versions as debugging reference, not for production - One Markdown → one `build_docx6.py` call → inspect → fix if needed FILE:references/workflow.md # 完整分析工作流手册 v2.0 > 记录：xixiCharon 深度分析全过程（三份文档合并 + hsword框架应用） > 原始 xhsfenxi 工作流见：workflow-xhsfenxi-v2.md --- ## 标准执行管道（Full Pipeline） ``` Step 0 Cookie 健康检查 Step 1 解析输入 → 检查数据库 Step 2 数据采集 Step 3 基础统计 Step 4 三型分类 Step 5 内核三段论（hsword框架） Step 6 爆款选题公式（6模型） Step 7 结构化报告 Markdown Step 8 外部文档合并（如有） Step 9 写入博主数据库 Step 10 生成黑体 Word ``` --- ## Step 0 — Cookie 健康检查 ```python from xhscosmoskill import print_cookie_status, get_best_cookies cookies = get_best_cookies() print_cookie_status(cookies) ``` **Cookie 优先顺序：** 1. `shopify-marketing/xhs_cookies.json`（最新，优先） 2. `xiaohongshu_new/xhs_cookies.json`（备用） **过期判断：** notes 返回 ≤ 1 条 → Cookie 失效 → 运行 `python3 xhs_login.py` --- ## Step 1 — 解析输入 ```python import re user_id = re.search(r'/user/profile/([a-f0-9]+)', url).group(1) from xhscosmoskill import get_blogger existing = get_blogger(creator_name) # 检查是否已分析过 ``` --- ## Step 2 — 数据采集 ```python from xhscosmoskill import XhsClient with XhsClient(cookies_file=cookies, headless=True, scroll_times=10) as xhs: notes = xhs.get_user_notes(user_id, limit=50) xhs.save(notes, f"/tmp/{creator_name}_notes.json") ``` --- ## Step 3-4 — 统计 + 分类 ```python from xhscosmoskill import compute_stats, classify_archetype, build_five_layers stats = compute_stats(notes) archetype = classify_archetype(notes) five = build_five_layers(notes, archetype) ``` --- ## Step 5 — 内核三段论（hsword框架）每次分析必须明确三层结构（参见 hsword-frameworks.md）： ``` 外壳是什么？→ 表面标签（可模仿）真正的内核？→ 一句话，带""引号三层人设：表层标签 → 自动提取中层特质 → 手动填写（性格/能力/气质）深层价值观 → 手动填写（最重要，不可复制） ``` --- ## Step 6 — 爆款选题公式 ```python from xhscosmoskill import generate_formula_report from xhscosmoskill.utils import save_md formula_md = generate_formula_report(notes, creator_name, archetype) save_md(formula_md, f"/tmp/{creator_name}-爆款选题公式.md") ``` 生成内容：总公式 + 6模型（每个含可套用句式）+ 30个选题方向 + 可迁移模板 --- ## Step 7 — 结构化报告 ```python from xhscosmoskill import analyze_account report_md = analyze_account(notes, creator_name=creator_name, mode="full") save_md(report_md, f"/tmp/{creator_name}-结构化总结报告.md") ``` --- ## Step 8 — 外部文档合并（如有）用户提供额外分析文档时，按优先级 patch： | 优先级 | 内容 | |--------|------| | 1 | 更高的点赞数（修正我们的数据）| | 2 | 新笔记数据（我们50篇未收录的）| | 3 | 新分析维度（我们没有的章节）| | 4 | 更精炼的提炼（更好的一句话总结）| **xixiCharon 合并经验：** - 文档1（txt）：补入 3.2万/7490/6231 等数据 + 视觉风格 + 风险分析 - 文档2（md）：补入 8813赞 + 头部集中度38.15% + 可摘抄感 + 新榜链接 - 验证：patch后检查7个关键词确认写入 --- ## Step 9 — 写入数据库 ```python from xhscosmoskill import save_blogger save_blogger( creator_name=creator_name, user_id=user_id, archetype=archetype, stats={"avg_likes": stats["avg_likes"], "max_likes": stats["max_likes"], "sample_size": stats["total"], "viral_count": stats["brackets"]["1万+"], "video_ratio": f"{stats['video_count']}/{stats['total']}"}, best_topic=five["best_topic"], best_topic_avg=five["best_topic_avg"], formula="核心公式一句话", ) ``` --- ## Step 10 — 生成黑体 Word ```python from xhscosmoskill.scripts.build_docx import build_word build_word(f"/tmp/{creator_name}-结构化总结报告.md", f"/tmp/{creator_name}-结构化总结报告.docx", title=creator_name, subtitle="小红书博主深度结构化分析报告") build_word(f"/tmp/{creator_name}-爆款选题公式.md", f"/tmp/{creator_name}-爆款选题公式.docx", title=creator_name, subtitle="爆款选题公式 · 6大模型 · 30个选题方向") ``` --- ## 证据分级 | 级别 | 来源 | 用法 | |------|------|------| | A1 | 小红书公开主页可见数据 | 直接陈述 | | A2 | 用户提供截图 | 直接陈述 | | B1 | 第三方公开（新榜/访谈）| 背景补充 | | C1 | 综合推断 | 明确标注"推断" | --- ## 类型迭代协议 ```python from xhscosmoskill import add_archetype, update_archetype_signals # 添加新类型（当最高分 < 10 时） add_archetype(key="D", name="知识科普型", desc="...", formula="...", title_signals=[], content_signals=[], commercial="...", difficulty="中") # 迭代已有类型信号词 update_archetype_signals("B", new_content_signals=["新词"]) ``` --- ## 已分析博主档案 | 博主 | user_id | 类型 | 最高赞 | 精神母题 | |------|---------|------|--------|---------| | xixiCharon | 5f94fa23000000000100bced | B+A | 10万+ | 一个女生如何在流动生活里确认自己 | --- ## 参考资源 - hsword 实战案例：`openclaw_cosmo/afa/hsword/` - hsword 框架手册：`references/hsword-frameworks.md` - 原始 xhsfenxi 工作流：`references/workflow-xhsfenxi-v2.md` - Word 修复脚本：`openclaw_cosmo/afa/小红书分析与工作流归档/02-Word生成与目录修复脚本/` - 新榜数据：`https://www.newrank.cn/profile/xiaohongshu/{user_id}` FILE:scripts/archetype.js #!/usr/bin/env node /** * xhsfenxi — archetype.js * Purpose: Print the archetype classification prompt for a Xiaohongshu creator. * Helps identify whether an account is Type A (荒诞美学), Type B (共鸣命名), * Type C (现实策略), or a hybrid — based on the three proven archetypes * distilled from real analyses of multiple Xiaohongshu creators. * Usage: * node scripts/archetype.js <creator-name> * node scripts/archetype.js <creator-name> --quick */ const args = process.argv.slice(2); const quick = args.includes('--quick'); const name = args.filter(a => !a.startsWith('--'))[0]; if (!name) { console.error('Usage: node scripts/archetype.js <creator-name> [--quick]'); process.exit(1); } if (quick) { console.log(` === xhsfenxi archetype cheatsheet === TYPE A — 荒诞美学型 (Absurdist Aesthetics) Example: 荒诞美学博主（vlogger型） Signals: - Unified brand symbol/tag on every post (e.g. "（劲爆）") - Absurdist or philosophical topics treated earnestly - High-production video; literary title naming - Serious × absurd contrast in every piece Formula: 荒诞滤镜下的生活哲学 TYPE B — 共鸣命名型 (Resonance & Naming) Example: 成长世界观表达博主 Signals: - Posts name vague emotional states or life stages - Private experience → universal proposition - Titles feel like thoughtful questions or redefinitions Formula: 经历 → 命题 → 命名 → 判断 → 共鸣 TYPE C — 现实策略型 (Reality & Strategy) Example: 普通女孩上行策略博主 Signals: - Titles contain conflict words ("骗子", "不要脸", "装") - Content breaks unspoken workplace/relationship/money rules - Self-labels with perceived weakness to pre-empt criticism Formula: 困境 → 说破 → 规则 → 策略 → 爽感 MIXED B+C — Most powerful hybrid Combines: resonance/naming × reality/strategy = "You understand me" + "Now I know what to do" `); process.exit(0); } console.log(` === xhsfenxi archetype classification: ${name} === TASK Classify the Xiaohongshu account "${name}" into one of these three archetypes (or a hybrid), based on visible public evidence. ───────────────────────────────────────────── THREE ARCHETYPES ───────────────────────────────────────────── TYPE A — 荒诞美学型 (Absurdist Aesthetics) Core: Wraps philosophical/serious content in absurdist humor + high visual quality Proven example: 荒诞美学vlogger型 (unified brand symbol; grand × mundane contrast; literary naming) Key signals to look for: - Does every post share a unified recurring symbol or phrase? - Do titles contrast a grand/serious setting with a mundane/absurd action? - Is the visual/production quality noticeably higher than peers? - Are serious topics treated with humor and humor treated earnestly? TYPE B — 共鸣命名型 (Resonance & Naming) Core: Names vague emotions and life stages so users feel "finally, someone said it" Proven example: 成长世界观表达型 (youth growth, worldview expression, concept naming) Key signals to look for: - Do posts give names or definitions to emotional states most people can't articulate? - Does the account turn personal stories into universal life propositions? - Do titles feel like thoughtful philosophical questions or redefinitions? - Is the tone warm, perceptive, aesthetically refined? TYPE C — 现实策略型 (Reality & Strategy) Core: Breaks unspoken real-world rules; provides executable strategies for ordinary people Proven example: 普通女孩上行策略型 (antifragile identity; rule-breaking; 普通人上行) Key signals to look for: - Do titles contain conflict words or counter-intuitive framing? - Does the account self-label with a perceived weakness to pre-empt criticism? - Does content expose hidden rules in workplace/relationships/money/consumption? - Does every piece end with a clear, actionable conclusion or stance? MIXED HYBRID Some accounts combine two archetypes. Most powerful is B+C: "I understand your situation (Type B) AND here's what you can actually do (Type C)" ───────────────────────────────────────────── CLASSIFICATION PROMPT ───────────────────────────────────────────── Based on publicly visible data for "${name}", answer: 1. Which archetype does this account most closely match? (A / B / C / Mixed) 2. What are 2–3 concrete visible signals that support this classification? 3. If mixed: what percentage is each type, and does one dominate? 4. What is the single sentence that describes this account's core "filter" on the world? 5. Is there a brand symbol or signature phrase? If yes, what is it and how does it function? Then proceed with the full analysis using this archetype as the lens. `); FILE:scripts/build_docx.py """ build_docx.py — 通用黑体 Word 生成器将 Markdown 文件转换为全黑体样式的 Word 文档支持：标题层级 / 表格 / 列表 / 引用 / 代码块 / 分割线用法（命令行）： python3 build_docx.py <md_path> <out_path> [title] [subtitle] [meta] 用法（Python）： from xhscosmoskill.scripts.build_docx import build_word build_word("/tmp/report.md", "/tmp/report.docx", title="xixiCharon", subtitle="爆款选题公式") """ import re import sys from pathlib import Path try: from docx import Document from docx.shared import Pt, RGBColor, Cm from docx.enum.text import WD_ALIGN_PARAGRAPH from docx.enum.table import WD_TABLE_ALIGNMENT from docx.oxml.ns import qn from docx.oxml import OxmlElement except ImportError: print("请先安装：pip install python-docx") sys.exit(1) # ── 配色 ────────────────────────────────────────────────────── C_BRAND = RGBColor(0x1A, 0x1A, 0x1A) C_H1 = RGBColor(0x0D, 0x0D, 0x0D) C_H2 = RGBColor(0x1A, 0x1A, 0x1A) C_ACCENT = RGBColor(0x00, 0x80, 0x60) # Shopify 绿 C_LIGHT = RGBColor(0x6B, 0x72, 0x80) C_TH_BG = RGBColor(0xF3, 0xF4, 0xF6) C_LINE = RGBColor(0x22, 0xC5, 0x5E) HEITI = "黑体" # ── 工具函数 ────────────────────────────────────────────────── def set_font(run, size, bold=False, color=None, italic=False): """全黑体设置（中英文统一）""" run.font.size = Pt(size) run.font.bold = bold run.font.italic = italic if color: run.font.color.rgb = color rPr = run._r.get_or_add_rPr() rf = rPr.find(qn('w:rFonts')) if rf is None: rf = OxmlElement('w:rFonts') rPr.insert(0, rf) for attr in ('w:ascii', 'w:hAnsi', 'w:eastAsia', 'w:cs'): rf.set(qn(attr), HEITI) def clean(s): for p, r in [ (r'\*\*(.+?)\*\*', r'\1'), (r'\*(.+?)\*', r'\1'), (r'\[(.+?)\]$.+?$', r'\1'), (r'`(.+?)`', r'\1'), (r'^#{1,6}\s+', ''), ]: s = re.sub(p, r, s) return s.strip() def set_cell_bg(cell, rgb): tcPr = cell._tc.get_or_add_tcPr() shd = OxmlElement('w:shd') shd.set(qn('w:val'), 'clear') shd.set(qn('w:color'), 'auto') shd.set(qn('w:fill'), f'{rgb[0]:02X}{rgb[1]:02X}{rgb[2]:02X}') tcPr.append(shd) def add_rule(doc, color=C_LINE): p = doc.add_paragraph() p.paragraph_format.space_before = Pt(1) p.paragraph_format.space_after = Pt(6) pPr = p._p.get_or_add_pPr() pBdr = OxmlElement('w:pBdr') bot = OxmlElement('w:bottom') bot.set(qn('w:val'), 'single') bot.set(qn('w:sz'), '4') bot.set(qn('w:space'), '1') bot.set(qn('w:color'), f'{color[0]:02X}{color[1]:02X}{color[2]:02X}') pBdr.append(bot) pPr.append(pBdr) # ── 核心构建函数 ────────────────────────────────────────────── def build_word(md_path: str, out_path: str, title: str = "", subtitle: str = "", meta: str = ""): """ 将 Markdown 转换为黑体 Word 文档参数： md_path — 输入 Markdown 文件路径 out_path — 输出 .docx 文件路径 title — 封面大标题（可选） subtitle — 封面副标题（可选） meta — 封面元数据行（可选） """ md_text = Path(md_path).read_text(encoding="utf-8") doc = Document() for sec in doc.sections: sec.top_margin = Cm(2.5) sec.bottom_margin = Cm(2.5) sec.left_margin = Cm(3.0) sec.right_margin = Cm(3.0) # 默认样式 nml = doc.styles['Normal'] nml.font.name = HEITI nml.font.size = Pt(10.5) nml._element.rPr.rFonts.set(qn('w:eastAsia'), HEITI) # 封面 if title: doc.add_paragraph() p = doc.add_paragraph() p.alignment = WD_ALIGN_PARAGRAPH.CENTER set_font(p.add_run(title), 28, bold=True, color=C_ACCENT) if subtitle: p = doc.add_paragraph() p.alignment = WD_ALIGN_PARAGRAPH.CENTER set_font(p.add_run(subtitle), 14, color=C_LIGHT) if meta: p = doc.add_paragraph() p.alignment = WD_ALIGN_PARAGRAPH.CENTER set_font(p.add_run(meta), 9, color=C_LIGHT) if title: doc.add_page_break() # 解析 Markdown lines = md_text.splitlines() i = 0 while i < len(lines): line = lines[i] # H1 if re.match(r'^# [^#]', line): p = doc.add_paragraph() p.paragraph_format.space_before = Pt(16) p.paragraph_format.space_after = Pt(4) set_font(p.add_run(clean(line)), 20, bold=True, color=C_H1) i += 1; continue # H2 if re.match(r'^## [^#]', line): p = doc.add_paragraph() p.paragraph_format.space_before = Pt(18) p.paragraph_format.space_after = Pt(2) set_font(p.add_run(clean(line)), 13, bold=True, color=C_H2) add_rule(doc) i += 1; continue # H3 if re.match(r'^### ', line): p = doc.add_paragraph() p.paragraph_format.space_before = Pt(10) p.paragraph_format.space_after = Pt(3) set_font(p.add_run(clean(line)), 11, bold=True, color=C_ACCENT) i += 1; continue # 引用 if line.startswith('> '): p = doc.add_paragraph() p.paragraph_format.left_indent = Cm(1) p.paragraph_format.space_before = Pt(3) p.paragraph_format.space_after = Pt(3) set_font(p.add_run(line[2:].strip()), 10.5, italic=True, color=C_LIGHT) i += 1; continue # 代码块 if line.startswith('```'): i += 1 code = [] while i < len(lines) and not lines[i].startswith('```'): code.append(lines[i]) i += 1 p = doc.add_paragraph('\n'.join(code)) p.paragraph_format.left_indent = Cm(0.8) p.paragraph_format.space_before = Pt(3) p.paragraph_format.space_after = Pt(3) for run in p.runs: set_font(run, 9, color=RGBColor(0x37, 0x37, 0x37)) i += 1; continue # 水平线 if line.strip() == '---': add_rule(doc, C_TH_BG) i += 1; continue # 表格 if line.strip().startswith('|'): tbl_lines = [] while i < len(lines) and lines[i].strip().startswith('|'): if not re.match(r'^\|[\s|:-]+$', lines[i].strip()): tbl_lines.append(lines[i]) i += 1 if not tbl_lines: continue rows = [ [c.strip() for c in tl.split('|')[1:-1]] for tl in tbl_lines ] rows = [r for r in rows if any(c for c in r)] if not rows: continue ncols = max(len(r) for r in rows) tbl = doc.add_table(rows=len(rows), cols=ncols) tbl.style = 'Table Grid' tbl.alignment = WD_TABLE_ALIGNMENT.LEFT for ri, row in enumerate(rows): for ci in range(ncols): cell = tbl.cell(ri, ci) txt = clean(row[ci]) if ci < len(row) else '' p = cell.paragraphs[0] p.clear() run = p.add_run(txt) set_font(run, 9.5, bold=(ri == 0), color=C_H2 if ri == 0 else C_BRAND) p.paragraph_format.space_before = Pt(2) p.paragraph_format.space_after = Pt(2) if ri == 0: set_cell_bg(cell, C_TH_BG) doc.add_paragraph() continue # 列表 if re.match(r'^[-*] |^\d+\. ', line): txt = re.sub(r'^[-*] |^\d+\. ', '', line) p = doc.add_paragraph(style='List Bullet') p.paragraph_format.left_indent = Cm(0.8) p.paragraph_format.space_before = Pt(1) p.paragraph_format.space_after = Pt(1) set_font(p.add_run(clean(txt)), 10.5, color=C_BRAND) i += 1; continue # 空行 if not line.strip(): i += 1; continue # 普通段落 p = doc.add_paragraph() p.paragraph_format.space_before = Pt(2) p.paragraph_format.space_after = Pt(4) set_font(p.add_run(clean(line)), 10.5, color=C_BRAND) i += 1 doc.save(out_path) print(f"✅ Word 已生成：{out_path}") return out_path # ── 命令行入口 ──────────────────────────────────────────────── if __name__ == "__main__": if len(sys.argv) < 3: print("用法：python3 build_docx.py <md_path> <out_path> [title] [subtitle] [meta]") sys.exit(1) md = sys.argv[1] out = sys.argv[2] ttl = sys.argv[3] if len(sys.argv) > 3 else "" sub = sys.argv[4] if len(sys.argv) > 4 else "" meta = sys.argv[5] if len(sys.argv) > 5 else "" build_word(md, out, ttl, sub, meta) FILE:scripts/docx-plan.js #!/usr/bin/env node /** * xhsfenxi — docx-plan.js * Purpose: Print the Word generation plan for a completed Xiaohongshu analysis, * including which build_docx*.py script to use, the full workflow steps, * and TOC fix guidance from the workflow archive. * Usage: * node scripts/docx-plan.js <creator-name> * node scripts/docx-plan.js <creator-name> --mode comparison * node scripts/docx-plan.js --toc-fix */ const ARCHIVE_BASE = '~/Desktop/cosmocloud/Deeplumen/cosmowork/openclaw_cosmo/afa/小红书分析与工作流归档/02-Word生成与目录修复脚本'; const args = process.argv.slice(2); const tocFix = args.includes('--toc-fix'); const modeIdx = args.indexOf('--mode'); const mode = modeIdx !== -1 ? args[modeIdx + 1] : 'single'; const name = args.filter((a, i) => { if (a.startsWith('--')) return false; if (modeIdx !== -1 && i === modeIdx + 1) return false; return true; })[0]; if (tocFix) { console.log(` === xhsfenxi Word TOC Fix Guide === Archive scripts location: ${ARCHIVE_BASE}/ STEP-BY-STEP FIX PROCESS 1. Run inspect_docx.py to diagnose the issue: python3 ${ARCHIVE_BASE}/inspect_docx.py <your-file.docx> → Check: bookmark names, link types, heading structure 2. Verify the following are correct: - Bookmark names follow format: _TocH1_XXX (NOT #_TocH1_XXX) - Internal TOC links use Word-native anchor, NOT external http:// rels - No duplicate headings accidentally inserted in document body - Namespace attributes written as: elem.set('{' + W_NS + '}id', value) 3. Run the appropriate fix script: - For attribute/namespace errors: python3 fix_attrs.py <file.docx> - For final TOC + heading cleanup: python3 fix_final2.py <file.docx> - For known wrong output patterns: python3 fix_wrong.py <file.docx> 4. Verify with check_it.py: python3 ${ARCHIVE_BASE}/check_it.py <your-fixed-file.docx> 5. Open in Word and test: click each TOC entry to confirm internal jump works. COMMON ERROR TABLE Bookmark format wrong → rename using fix_attrs.py anchor contains # → strip # prefix in fix_final2.py TOC uses external rels → switch to w:instr field or Word anchor Duplicate headings → fix_wrong.py XML namespace error → use full namespace URI in elem.set() `); process.exit(0); } if (!name && !tocFix) { console.error('Usage: node scripts/docx-plan.js <creator-name> [--mode comparison]'); console.error(' node scripts/docx-plan.js --toc-fix'); process.exit(1); } const isComparison = mode === 'comparison'; const mdFile = isComparison ? `选题公式学习-综合版.md` : `${name}-结构化总结报告.md`; const docxFile = isComparison ? `选题公式学习-综合版.docx` : `${name}-结构化总结报告-商业版.docx`; console.log(` === xhsfenxi Word generation plan: ${name || 'comparison'} === Mode: ${isComparison ? 'comparison / 综合版' : 'single creator'} ───────────────────────────────────────────── STEP 1 — Confirm Markdown is complete ───────────────────────────────────────────── Make sure this file is finalized: ${mdFile} ───────────────────────────────────────────── STEP 2 — Run build_docx6.py (recommended) ───────────────────────────────────────────── Primary script (most stable): python3 ${ARCHIVE_BASE}/build_docx6.py ${mdFile} ${docxFile} If build_docx6.py fails, fall back to earlier versions: build_docx5.py → build_docx4.py → build_docx3.py ───────────────────────────────────────────── STEP 3 — Inspect the output ───────────────────────────────────────────── python3 ${ARCHIVE_BASE}/inspect_docx.py ${docxFile} Check for: ✓ Headings have _TocH1_XXX bookmarks ✓ TOC links are internal (not external URL rels) ✓ No duplicate heading text in body ✓ Page count and section structure look correct ───────────────────────────────────────────── STEP 4 — Fix if needed ───────────────────────────────────────────── Run: node scripts/docx-plan.js --toc-fix for the full TOC fix checklist. Most common fix: python3 ${ARCHIVE_BASE}/fix_final2.py ${docxFile} ───────────────────────────────────────────── STEP 5 — Final check ───────────────────────────────────────────── python3 ${ARCHIVE_BASE}/check_it.py ${docxFile} → Open in Word, click TOC entries, verify internal jumps work ───────────────────────────────────────────── EXPECTED OUTPUT FILES ───────────────────────────────────────────── ${docxFile} ${isComparison ? '' : ` ${name}-结构化总结报告.md (keep as source of truth)\n ${name}-爆款选题公式.md (if also produced)`} ───────────────────────────────────────────── ARCHIVE LOCATION ───────────────────────────────────────────── Scripts: ${ARCHIVE_BASE}/ Stable production script: build_docx6.py Inspection tool: inspect_docx.py `); FILE:scripts/intake.js #!/usr/bin/env node /** * xhsfenxi — intake.js * Purpose: Print a structured intake template for a Xiaohongshu creator-analysis task. * Usage: * node scripts/intake.js * node scripts/intake.js --json */ const args = process.argv.slice(2); const asJson = args.includes('--json'); const template = { task: 'xiaohongshu-creator-analysis', creatorNames: ['<creator-name>'], links: ['<homepage-or-search-url>'], screenshots: ['<optional-local-path-or-url>'], deliverables: ['structured-report'], outputDir: '<optional-output-dir>', needWord: false, needBusinessVersion: false, needComparison: false, notes: 'What exactly should be learned from this creator?' }; if (asJson) { console.log(JSON.stringify(template, null, 2)); process.exit(0); } console.log(` === xhsfenxi intake template === Provide or confirm the following before deep research: - creator name(s) - homepage/search URL(s) - screenshots or links for representative posts (optional but helpful) - requested deliverable(s): structured-report / topic-formula / comparison / business-word - output directory - any specific question to answer JSON template: ${JSON.stringify(template, null, 2)} `); FILE:scripts/report-plan.js #!/usr/bin/env node /** * xhsfenxi — report-plan.js * Purpose: Print recommended deliverables and filenames for a Xiaohongshu analysis request. * Usage: * node scripts/report-plan.js <creator-name> * node scripts/report-plan.js <creator-a> <creator-b> --mode compare */ const args = process.argv.slice(2); const modeIdx = args.indexOf('--mode'); const mode = modeIdx !== -1 ? args[modeIdx + 1] : 'single'; const names = args.filter((a, i) => { if (a === '--mode') return false; if (modeIdx !== -1 && i === modeIdx + 1) return false; return true; }); if (!names.length) { console.error('Usage: node scripts/report-plan.js <creator-name>'); console.error(' node scripts/report-plan.js <creator-a> <creator-b> --mode compare'); process.exit(1); } if (mode === 'compare') { if (names.length < 2) { console.error('Compare mode requires two creator names.'); process.exit(1); } const [a, b] = names; const files = [ `选题公式学习-综合版.docx` ]; console.log(` === xhsfenxi report plan === Mode: comparison Creators: ${a} / ${b} Recommended deliverables: - comparison report - hybrid topic-formula study - optional business Word version Recommended filenames: ${files.map(f => `- ${f}`).join('\n')} `); process.exit(0); } const name = names[0]; const files = [ `${name}-结构化总结报告.docx`, `${name}-爆款选题公式.docx` ]; console.log(` === xhsfenxi report plan === Mode: single creator Creator: ${name} Recommended deliverables: - structured report - viral topic formula - optional business Word version Recommended filenames: ${files.map(f => `- ${f}`).join('\n')} `); FILE:utils.py """ utils.py — 工具函数集包含：Cookie 健康检查 / 路径常量 / 输出工具 """ import json import os import time from typing import Dict, Any _BASE = os.path.dirname(__file__) # ── 路径常量 ────────────────────────────────────────────────── COOKIES_PRIMARY = os.path.join( _BASE, "..", "..", "..", "shopify-marketing", "xhs_cookies.json" ) COOKIES_ALT = os.path.join(_BASE, "..", "xhs_cookies.json") DATA_DIR = os.path.join(_BASE, "data") ARCHETYPES_DB = os.path.join(DATA_DIR, "archetypes.json") BLOGGERS_DB = os.path.join(DATA_DIR, "bloggers.json") HSWORD_DIR = os.path.join( _BASE, "..", "..", "..", "..", "openclaw_cosmo", "afa", "hsword" ) # ── Cookie 健康检查 ─────────────────────────────────────────── def check_cookies(cookies_file: str = None) -> Dict[str, Any]: """ 检查 Cookie 文件的有效性返回： { "valid": bool, "file": str, "total": int, "expired": list[str], "expiring_soon": list[str], # 24小时内过期 "message": str } """ path = cookies_file or _find_cookies() if not path or not os.path.exists(path): return {"valid": False, "file": path, "message": "Cookie 文件不存在"} try: with open(path, encoding="utf-8") as f: cookies = json.load(f) except Exception as e: return {"valid": False, "file": path, "message": f"读取失败: {e}"} now = int(time.time()) expired = [] soon = [] for c in cookies: exp = c.get("expiry") or c.get("expires", 0) if not exp: continue diff = exp - now if diff <= 0: expired.append(c["name"]) elif diff < 86400: soon.append(c["name"]) valid = len(expired) == 0 parts = [] if expired: parts.append(f"❌ 已过期：{', '.join(expired)}") if soon: parts.append(f"⚠️ 即将过期（24h内）：{', '.join(soon)}") if valid and not soon: parts.append("✅ 全部有效") return { "valid": valid, "file": path, "total": len(cookies), "expired": expired, "expiring_soon": soon, "message": " | ".join(parts), } def _find_cookies() -> str: """自动查找最新 Cookie 文件""" candidates = [ os.path.abspath(COOKIES_PRIMARY), os.path.abspath(COOKIES_ALT), ] # 返回最近修改的有效文件 valid = [p for p in candidates if os.path.exists(p)] if not valid: return "" return max(valid, key=os.path.getmtime) def get_best_cookies() -> str: """获取当前最佳 Cookie 路径（优先使用最新且有效的）""" for path in [os.path.abspath(COOKIES_PRIMARY), os.path.abspath(COOKIES_ALT)]: if os.path.exists(path): status = check_cookies(path) if status["valid"]: return path # 如果都过期，返回最新修改的（并给出警告） return _find_cookies() # ── 输出工具 ────────────────────────────────────────────────── def print_cookie_status(cookies_file: str = None): """打印 Cookie 状态（带颜色提示）""" status = check_cookies(cookies_file) print(f"Cookie 文件: {status['file']}") print(f"状态: {status['message']}") if not status["valid"]: print("⚡ 请重新运行：python3 xhs_login.py") return status["valid"] def save_md(content: str, path: str): """保存 Markdown 文件""" with open(path, "w", encoding="utf-8") as f: f.write(content) print(f"📄 Markdown 已保存：{path}") def parse_likes(v) -> int: """解析点赞数（支持'万'格式）""" if not v: return 0 s = str(v).replace(",", "").strip() if "万" in s: return int(float(s.replace("万", "")) * 10000) try: return int(s) except ValueError: return 0 FILE:xhsfenxi/SKILL.md --- name: xhsfenxi description: | 小红书博主分析全流程 skill — 拆解账号定位、提炼爆款选题公式、输出结构化报告和商业版 Word。内置三型博主分类体系（荒诞美学型 / 共鸣命名型 / 现实策略型），基于多位真实博主的深度分析沉淀。支持单账号深度拆解、多账号对比、自定义学习报告，证据分级（公开主页→截图→第三方公开资料→推断），所有交付物可一键转为 Markdown + 商业 Word。Use it when the user asks to analyze a Xiaohongshu blogger/account, extract viral topic formulas, compare creators, or produce business-grade deliverables. keywords: - xiaohongshu - RED - 小红书 - 小红书分析 - 博主分析 - 账号拆解 - 对标分析 - 爆款选题 - 爆款选题公式 - 内容策略 - 个人IP分析 - 创作者分析 - 三型博主 - 荒诞美学型 - 共鸣命名型 - 现实策略型 - 品牌符号 - 结构化总结报告 - Word报告 - 商业版报告 - 选题公式学习 - 博主拆解 - 小红书运营 - 内容账号分析 - 竞品分析 - 账号定位 - benchmark account - topic formula - creator research - content audit - account positioning - competitor analysis - archetype classification - viral topic - markdown report - word report - 小红书博主对比 - 爆款内容拆解 - 内容IP打法 metadata: openclaw: runtime: node: ">=18" --- # xhsfenxi — 小红书博主分析 Skill > 公开页分析 · 三型博主分类 · 爆款选题公式提炼 · 多账号对比 · Markdown/商业Word交付 --- ## When to Use Use this skill when the user asks to: - analyze a Xiaohongshu / RED creator account - identify which archetype a creator belongs to (荒诞美学型 / 共鸣命名型 / 现实策略型) - summarize an account's positioning, audience, style, or content pillars - extract a **viral topic formula** or reusable content method - compare two or more benchmark creators - learn what can be copied vs what must stay differentiated - turn the analysis into a **Markdown report**, **Word report**, or **business-style deliverable** Typical trigger phrases: - "分析这个小红书博主" - "拆一下这个账号" - "做爆款选题公式" - "帮我做对标分析" - "做成商业版 Word" - "这个博主是哪一型？" - "compare these two RED creators" - "extract the topic formula from this account" --- ## Three-Archetype Classification System Distilled from real analyses of multiple Xiaohongshu creators. ### Type A — 荒诞美学型 (Absurdist Aesthetics) **Representative archetype:** Core pattern: wraps serious or philosophical content in absurdist humor and high-quality visuals. | Dimension | Characteristics | |-----------|----------------| | Content kernel | Absurdist filter on everyday life; philosophical undertone | | What users get | Entertained + moved + aesthetically immersed | | Title mechanism | Contrast (grand setting × mundane action) + unified brand symbol | | Brand symbol | A single recurring tag, e.g. "（劲爆）" | | Expression style | Humorous, high-production, literary naming | | Commercial fit | High-end lifestyle, travel, photography, luxury brands | | Replication difficulty | High — depends on accumulated visual aesthetic sensibility | **Identifying markers:** - All content tied together by a signature phrase or visual symbol - Serious topics expressed lightly; absurd topics expressed earnestly - Titles feel like literary naming, not just descriptions --- ### Type B — 共鸣命名型 (Resonance & Naming) **Representative archetype:** Core pattern: translates personal experience into universally relatable life-stage propositions, named in memorable ways. | Dimension | Characteristics | |-----------|----------------| | Content kernel | Youth growth, worldview expression, life-stage naming | | What users get | Understood + named + given a new perspective | | Title mechanism | Proposition feeling + metaphor + judgment | | Brand symbol | Signature conceptual phrases and naming patterns | | Expression style | Warm, perceptive, aesthetically refined | | Commercial fit | Growth/education/creative platforms, mid-range lifestyle | | Replication difficulty | Medium — method learnable, requires genuine observation | **Identifying markers:** - Content "names" vague emotional states people couldn't articulate - Private experience → universal proposition - Titles read like thoughtful questions or redefinitions --- ### Type C — 现实策略型 (Reality & Strategy) **Representative archetype:** Core pattern: breaks down unspoken real-world rules and provides executable strategies for ordinary people to move upward. | Dimension | Characteristics | |-----------|----------------| | Content kernel | Ordinary woman's upward mobility, reality rule-breaking | | What users get | Validated + activated + empowered with actionable moves | | Title mechanism | Conflict words + counter-intuitive framing + strong stance | | Brand symbol | Self-labeling with perceived weakness (先夺走羞耻感) | | Expression style | Sharp, realistic, strong-attitude conclusions | | Commercial fit | Mass consumer, career, e-commerce, practical tools | | Replication difficulty | Medium — framework learnable; don't copy the surface aggression | **Identifying markers:** - Titles feel slightly improper but undeniably accurate - Each piece breaks an unspoken rule or collapses an information gap - Users feel "刺但有用" — uncomfortable but useful --- ### Mixed Formula (Advanced) The most powerful content often combines two archetypes: > **Type B resonance/naming × Type C reality/strategy** > = Content that both "understands you" and "tells you what to do next" Use `scripts/archetype.js` to quickly identify which archetype(s) an account fits. --- ## Core Operating Rules 1. **Use public evidence first.** Prefer search pages, homepage/profile pages, and other publicly visible surfaces. 2. **Do not fake detail-page access.** If post detail pages are blocked by risk control, say so clearly. 3. **Grade evidence quality.** Treat sources in this order: - Level A: visible Xiaohongshu public pages / user screenshots - Level B: other public sources (podcasts, interviews, encyclopedias, analytics sites) - Level C: synthesis / inference based on repeated visible patterns 4. **Do not present public estimates as backend truth.** Third-party follower or pricing data are supporting clues, not audited facts. 5. **Ask for 3–10 representative post links or screenshots** when the user wants deep post-level analysis. 6. **Deliver something useful even when blocked.** If detail pages fail, still produce a homepage-level strategic report. 7. **Always identify the archetype.** Every analysis should name which of the three types (or hybrid) the account belongs to — this frames the entire report. --- ## Default Workflow ### 1) Clarify the task shape Lock the requested output mode before heavy research: - Structured account report - Viral topic formula - Archetype identification - Two-or-more-account comparison - "How to learn from this creator without copying them" - Business-style Word deliverable Use `scripts/intake.js` for a fast intake template. ### 2) Gather source inputs Collect whatever the user already has: - account name - Xiaohongshu homepage/search URL - screenshots - external links (podcasts, news, interviews) - target output path - whether they need Word / clickable TOC / business styling ### 3) Research with the source ladder Read `references/workflow.md` for the full ladder. Preferred tool pattern: - `web_search` for public discovery - `browser` or `browser-use` for homepage/search-page inspection - `web_fetch` for readable public pages - `image` for screenshot analysis - `read` for any local Markdown reports already created ### 4) Classify the archetype Before deep analysis, make a preliminary archetype call: - Run `node scripts/archetype.js <creator-name>` to print the classification prompt - Answer: Type A (荒诞美学) / Type B (共鸣命名) / Type C (现实策略) / Mixed - The archetype frames the entire lens of the analysis ### 5) Build the account model Extract and write down the five layers: 1. **Identity** — who the creator is framed as 2. **Audience contract** — why people follow them 3. **Topic system** — what recurring problems/desires they cover 4. **Expression system** — how titles and openings work, including any brand symbol 5. **Transferability** — what another account can actually learn ### 6) Produce the deliverables Always output the three standard DOCX files directly — no intermediate Markdown step needed: | Output file | When to produce | |-------------|----------------| | `账号名-结构化总结报告.docx` | Single creator analysis | | `账号名-爆款选题公式.docx` | Single creator topic formula | | `选题公式学习-综合版.docx` | Multi-creator comparison | Use `scripts/docx-plan.js <creator-name>` to get the exact build command. See `references/workflow.md` for the Word TOC fix checklist if needed. --- ## Deliverables Three standard output files. Always produce as `.docx` directly. ### 账号名-结构化总结报告.docx Single creator full breakdown. Sections: 1. Account snapshot 2. Archetype classification + rationale 3. Strategic positioning 4. Audience and emotional contract 5. Content pillars 6. Title / hook system + brand symbol analysis 7. Narrative structure 8. Commercialization clues 9. What to learn 10. What not to blindly copy 11. Conclusion ### 账号名-爆款选题公式.docx Single creator topic formula. Sections: 1. Why this creator produces strong topics (archetype-based) 2. Total formula 3. 3–6 recurring topic models 4. Title formulas 5. Body structure formulas 6. Distribution / comment triggers 7. How to migrate to another account 8. 10–30 ready-to-use topic directions ### 选题公式学习-综合版.docx Multi-creator comparison. Sections: 1. Why compare these creators together 2. Archetype map 3. Shared foundations + key differences 4. Hybrid formula 5. Which path fits which use case 6. 30 combined topic directions 7. Final recommendation --- ## Word Output Run `node scripts/docx-plan.js <creator-name>` to get the exact build command. Standard build: `python3 build_docx6.py <creator-name>` If TOC / bookmark issues appear: run `inspect_docx.py` then `fix_final2.py`. See `references/workflow.md` for the full fix checklist. --- ## Output Files Three files, always `.docx`: - `账号名-结构化总结报告.docx` - `账号名-爆款选题公式.docx` - `选题公式学习-综合版.docx` (comparison / multi-creator) Use `scripts/report-plan.js <creator-name>` to confirm filenames. --- ## Scripts | Command | Purpose | |--------|---------| | `node scripts/intake.js` | Print a structured intake template for a new analysis task | | `node scripts/intake.js --json` | Same as above, JSON format | | `node scripts/archetype.js <creator-name>` | Print the archetype classification prompt | | `node scripts/archetype.js <creator-name> --quick` | Print a quick-read archetype cheatsheet | | `node scripts/report-plan.js <creator-name>` | Print recommended deliverables and filenames | | `node scripts/report-plan.js <A> <B> --mode compare` | Print a comparison deliverable plan | | `node scripts/docx-plan.js <creator-name>` | Print Word generation plan using build_docx*.py | --- ## Limitation Policy Always state which situation applies: - **Homepage-level only**: enough for strategic analysis, not enough for exact post-body claims - **Mixed-source analysis**: Xiaohongshu public pages + third-party public clues - **Deep-dive mode**: based on user-provided screenshots / links / transcripts If blocked by risk control, say so and continue with the strongest available public evidence. --- ## What Good Output Looks Like A strong output is: - structured - archetype-aware (names and reasons the type) - evidence-aware (grades sources) - strategically useful - honest about access limits - directly reusable in future reports Avoid vague praise, empty creator admiration, or unsupported claims about hidden metrics. --- ## References Read only when needed: - `references/workflow.md` — detailed evidence ladder, execution runbook, Word TOC fix guide - `references/templates.md` — report templates, archetype-specific section structures, naming patterns --- ## Proven Analyses Archive Three full analyses have been completed and archived in: ``` openclaw_cosmo/afa/小红书分析与工作流归档/01-分析报告与选题公式/ ``` | Archetype | Type | Key Insight | |-----------|------|-------------| | 荒诞美学型 | Type A | 品牌符号 + 反差标题 + 哲思轻量化输出 | | 共鸣命名型 | Type B | 私人经历→公共命题，给模糊状态命名 | | 现实策略型 | Type C | 困境→说破→规则→策略→爽感 | These archetypes are the ground truth for the classification system. --- *Version: 2.1.0 · Created: 2026-04-08 · Updated: 2026-04-09* FILE:xhsfenxi/_meta.json { "ownerId": "kn79bebfnwg15sb0g7cj5z5nyd83gxh0", "slug": "xhsfenxi", "version": "2.1.0", "publishedAt": null } FILE:xhsfenxi/package.json { "name": "xhsfenxi", "version": "2.1.0", "description": "xhsfenxi — Xiaohongshu creator-account analysis workflow with three-archetype classification (荒诞美学型/共鸣命名型/现实策略型), viral topic formula extraction, comparison studies, and Markdown/Word deliverables.", "keywords": [ "xiaohongshu", "RED", "小红书", "博主分析", "账号拆解", "爆款选题", "三型博主", "荒诞美学", "共鸣命名", "现实策略", "topic-formula", "creator-research", "content-strategy", "benchmark-analysis", "archetype-classification", "word-report", "business-report" ], "author": "cosmofang", "license": "MIT", "scripts": { "intake": "node scripts/intake.js", "intake:json": "node scripts/intake.js --json", "plan": "node scripts/report-plan.js", "archetype": "node scripts/archetype.js", "archetype:quick": "node scripts/archetype.js --quick", "docx-plan": "node scripts/docx-plan.js", "toc-fix": "node scripts/docx-plan.js --toc-fix" } } FILE:xhsfenxi/references/templates.md # Templates Reference — xhsfenxi v2.0 --- ## Output Files Always produce as `.docx` directly. Three standard files: | File | When | |------|------| | `账号名-结构化总结报告.docx` | Single creator analysis | | `账号名-爆款选题公式.docx` | Single creator topic formula | | `选题公式学习-综合版.docx` | Multi-creator comparison | --- ## Structured Report Skeleton (v2 — Archetype-Aware) ```markdown # 账号名结构化总结报告 > 说明：本报告基于[数据来源说明]。适合用于[账号定位/内容策略/竞品研究/爆款机制提炼]。 ## 一、执行摘要 [2–4段: 博主是谁，核心特征，最值得记住的一句话] ## 二、账号基础信息 [表格: 昵称/小红书号/粉丝/获赞收藏/笔记数/内容类型/账号阶段] [补充: 全平台历史数据（如有）] ## 三、博主背景（公开资料） [人物背景、职业起点、创作理念（来源于公开访谈）] ## 四、账号类型判断 **类型:** Type A/B/C 或混合型 **判断依据:** [2–3条可见信号] **核心内核:** [一句话总结] ## 五、博主核心内核拆解 [外壳是什么 / 真正的内核是什么 / 人设三层结构 / 最强能力] ## 六、内容结构拆解 [内容支柱分布（表格）/ 内容类型比例 / 统一主题] ## 七、标题机制与爆点分析 [最重要特征 / 品牌符号分析 / 高热内容分析（表格）/ 标题公式归纳] ## 八、与其他博主的横向对比（可选） [对比表格: 多维度 × 多博主] ## 九、商业化判断 [当前阶段 / 为什么品牌会找他 / 适合的合作方向 / 不适合的方向] ## 十、最值得学习的核心能力 [3–5个能力，每个含: 说明 + 可学习点] ## 十一、后续建议 [路径A: 深入单篇分析 / 路径B: 制作爆款选题公式 / 路径C: 跨博主综合版] ## 十二、最终总结 [1–2段核心结论] ## 附录：笔记完整列表（可选） [表格: 排名 / 类型 / 标题 / 点赞] ``` --- ## Viral Topic Formula Skeleton (v2) ```markdown # 账号名爆款选题公式 > 说明：本文件不是复刻[账号名]的表面语气，而是提炼真正有效的选题逻辑、标题机制、传播结构、人设打法与可迁移方法。 ## 一、先说结论：为什么能做出爆款？ [账号类型 + 5个叠在一起的核心能力 + 总公式一句话] ## 二、总爆款公式 [5步骤公式 + 压缩版关键词] ## 三、最常见的 3–6 类爆款选题模型 [每个模型: 公式 / 典型表达方向 / 为什么有效 / 可套用句式 / 适合主题] ## 四、标题公式：可以直接拆用 [5–6个标题公式，每个含: 结构说明 + 套用模板] ## 五、内容结构公式 [2–3个结构公式，每个含: 步骤 / 适合做的主题] ## 六、隐形公式 / 最值钱的底层打法 [3–4个] ## 七、如何迁移到自己的内容里 [4个步骤: 找现实母题 / 改写生活故事 / 用冲突词 / 4问检验] ## 八、可直接复用的 20 个选题方向 [编号列表] ## 九、哪些能学，哪些不能硬抄 [可以学的 / 不能硬抄的] ## 十、最终总结 [一句话核心结论] ``` --- ## Comparison Skeleton (v2 — Archetype Map) ```markdown # 选题公式学习（综合版） ## ——结合[博主A]与[博主B]（以及[博主C]）的内容方法论 > 说明：这份文档不是简单拼接报告，而是把各博主最有效的内容能力抽取出来，重新整理成可学习、可迁移、可直接应用的选题方法。 ## 一、为什么把这几位博主放在一起学？ [各博主关键词 / 各博主强项一句话 / 为什么组合更完整] ## 二、账号类型对照表 [表格: 博主 / 类型 / 核心方向 / 用户获得感 / 内容气质 / 标题风格 / 爆点来源] ## 三、共同底层：真正的选题不是题材，而是命题 [5个共同点 / 核心结论] ## 四、双系统（或多系统）选题公式 [系统A（共鸣命名型）: 适用场景 / 核心公式 / 压缩版 / 标题类型] [系统B（现实策略型）: 同上] [（如有）系统C（荒诞美学型）: 同上] ## 五、混合公式 [混合公式定义 + 例子 + 为什么更强] ## 六、标题公式综合版 [6–8个跨类型标题公式] ## 七、内容结构综合版 [结构A / B / C，各含步骤和用户获得感] ## 八、可直接复用的 30 个综合选题方向 [A组（偏共鸣命名）/ B组（偏现实策略）/ C组（偏荒诞美学）/ D组（混合）] ## 九、结论与行动建议 [最重要的学习提醒 + 最终公式] ``` --- ## Archetype Comparison Table Template Use this table when comparing 2–3+ accounts: ```markdown | 维度 | [博主A] | [博主B] | [博主C] | |---|---|---|---| | 账号类型 | Type A 荒诞美学 | Type B 共鸣命名 | Type C 现实策略 | | 核心方向 | | | | | 用户获得感 | | | | | 内容气质 | | | | | 标题风格 | | | | | 品牌符号 | | | | | 爆点来源 | | | | | 商业化方向 | | | | | 可复制难度 | | | | ``` --- ## Evidence Label Reference Use these labels in your internal working notes (do not need to show all in final report): | Label | Meaning | |-------|---------| | `A1` | Xiaohongshu homepage visible fact | | `A2` | Xiaohongshu screenshot fact | | `B1` | Third-party public clue (interview, news, etc.) | | `C1` | Synthesis / interpretation | --- ## Business Word Checklist When building a business-style Word file: - [ ] Cover page - [ ] TOC with `_TocH1_XXX` bookmarks (internal jump, not external link) - [ ] Page numbers - [ ] Conclusion page - [ ] Restrained typography (no loud decorative color blocks unless requested) - [ ] Tested in Word: TOC entries jump correctly - [ ] No duplicate headings accidentally inserted in body --- ## Brand Symbol Analysis Template For Type A accounts or any account with a unified recurring symbol: ```markdown ### 品牌符号分析 **符号:** [（劲爆）/ 其他] **出现频率:** [几乎所有 / 大部分 / 部分] **品牌效果:** - 识别度: [描述] - 反差机制: [描述] - 先发制人效果: [描述] **可学习点:** [其他账号如何建立自己的品牌符号] ``` FILE:xhsfenxi/references/workflow.md # Workflow Reference — xhsfenxi v2.0 --- ## Source Ladder Use sources in this order whenever possible: ### Level A — Primary public evidence Highest trust for visible claims: - Xiaohongshu search result pages - creator homepage/profile pages - user-provided screenshots from Xiaohongshu - user-provided post links that are actually accessible What you can safely claim from Level A: - visible follower / note counts shown on page - visible account ID / self-introduction - visible recent titles / visible public engagement numbers - posting cadence clues visible on page - obvious content categories - brand symbols and signature phrases (like "（劲爆）") ### Level B — Secondary public clues Use as support, never as backend truth: - podcast interviews - encyclopedia entries - public creator bios elsewhere - analytics websites / public creator databases - public brand collaboration listings What Level B is good for: - filling identity / background context - confirming repeated themes - spotting public commercialization clues - triangulating a bigger narrative ### Level C — Synthesis / inference Allowed only when clearly framed as interpretation. Examples: - "This account behaves more like a Type A 荒诞美学 creator than a Type B 共鸣命名 creator." - "The recurring topic logic appears to be: contrast × literary naming × unified symbol." - "The transferable lesson is not the tone, but the framing system." Never present Level C synthesis as raw observed fact. --- ## Standard Execution Runbook ### Step 1 — Confirm the deliverable Choose one or more: - structured report - viral topic formula - archetype classification - comparison report - customized learning report - Word / business version ### Step 2 — Capture the minimum task brief Minimum useful brief: - creator name - available links or screenshots - target output format - whether a business Word version is required ### Step 3 — Archetype pre-classification Before deep analysis, make a preliminary archetype call using visible signals: | Signal | Likely archetype | |--------|-----------------| | Unified brand symbol/tag on every post | Type A 荒诞美学型 | | High-production video, absurdist or philosophical content | Type A 荒诞美学型 | | Posts name vague emotions or life stages | Type B 共鸣命名型 | | Titles feel like thoughtful questions or redefinitions | Type B 共鸣命名型 | | Titles contain conflict words ("骗子", "不要脸", "装") | Type C 现实策略型 | | Content breaks unspoken workplace/relationship/money rules | Type C 现实策略型 | | Mix of resonance + strategy | Mixed B+C | ### Step 4 — Research in layers 1. Find the account and public homepage/search data 2. Save visible account facts (follower count, note count, visible titles, engagement) 3. Pull representative visible titles — note brand symbols, title patterns 4. Add external public clues only where helpful 5. Record limitations before interpretation ### Step 5 — Extract the account system (Five Layers) Always map these five layers: 1. **Identity** — who the creator is framed as; any self-labeling strategy 2. **Audience contract** — why people follow them; what emotional/strategic need is met 3. **Topic system** — what recurring problems/desires they cover; 3–6 topic models 4. **Expression system** — title formulas, brand symbols, opening patterns 5. **Transferability** — what another account can actually learn; what must not be copied ### Step 6 — Produce the right file #### Structured report Recommended top-level sections: 1. Account snapshot 2. Archetype classification + rationale 3. Positioning judgment 4. Audience and demand 5. Content pillars 6. Title / hook system (including brand symbol analysis) 7. Narrative structure 8. Commercialization clues 9. What to learn 10. What not to copy 11. Conclusion 12. (Optional) Full note list sorted by engagement #### Viral topic formula Recommended top-level sections: 1. Why this creator produces strong topics (archetype-based) 2. Total formula 3. 3–6 recurring topic models 4. Title formulas 5. Body structure formulas 6. Distribution / comment trigger clues 7. How to migrate it to another account 8. 10–30 topic directions #### Comparison report (2–3+ accounts) Recommended top-level sections: 1. Why compare these creators together 2. Archetype map (each creator's type) 3. Shared foundations 4. Key differences 5. Hybrid formula 6. Which path fits which use case 7. Suggested topic directions 8. Final recommendation --- ## Archetype-Specific Analysis Lenses ### Type A — 荒诞美学型 Focus questions: - What is the brand symbol? (unified tag, phrase, or visual marker) - How does the title create contrast? (grand × mundane, serious × absurd) - Where does the humor come from? (earnest treatment of absurd things) - What is the underlying philosophical/emotional core? - Why is replication hard? (visual sensibility, accumulated aesthetic) Key learning for other accounts: - Find your own "brand symbol" — a unifying tag that creates recognition - Train "contrast naming" — don't describe the place, name the feeling - Practice "absurdist framing" — pick topics that are ridiculous when taken seriously --- ### Type B — 共鸣命名型 Focus questions: - What life stages or emotional states does the account name? - How does private experience become universal proposition? - What are the signature conceptual phrases? - What is the "new way to understand yourself" users get? Key learning for other accounts: - Practice "命题化" — reframe personal experience as a universal human situation - Build a personal vocabulary of named states ("Odyssey时期", etc.) - The formula: experience → proposition → naming → judgment → resonance --- ### Type C — 现实策略型 Focus questions: - What real-world rules does the account break open? - How does the self-labeling strategy work? (antifragile identity) - What is the conflict word in each title? - What is the "can-do action" users walk away with? Key learning for other accounts: - Find your "reality母题" — the real困境 you speak to - Practice rewriting weak framings into strong rule-breaking framings - The formula: 困境 → 说破 → 规则 → 策略 → 爽感 --- ## Limitation Language Use direct wording like: - "This round is based primarily on public homepage/search-page evidence." - "Single-post detail pages were not stably accessible due to platform risk control." - "Third-party figures are used as directional clues, not audited backend data." - "Archetype classification is based on visible patterns; the creator may operate differently at the full content level." --- ## Deep-Dive Upgrade Path If the user wants finer analysis, ask for 3–10 representative posts, ideally with: - screenshots of cover + title - screenshots of body pages / subtitles - comments screenshots - transcript or summary if video-based Then upgrade from account-level to post-level analysis. --- ## Word Output Workflow ### Overview Always generate Markdown first, then convert to Word. Iterating Markdown is cheaper than re-generating DOCX. ### Script Selection Guide Use the workflow archive scripts at: ``` openclaw_cosmo/afa/小红书分析与工作流归档/02-Word生成与目录修复脚本/ ``` | Script | Use when | |--------|----------| | `build_docx6.py` | Default — most stable, use this first | | `build_docx.py` → `build_docx5.py` | Earlier iterations, kept for reference/debugging | | `inspect_docx.py` | Inspect internal XML, bookmarks, link structure | | `fix_final2.py` | Final TOC / attribute fix after build | | `fix_attrs.py` | Fix XML namespace attribute writing errors | | `fix_wrong.py` | Targeted corrections for known wrong output | | `check_it.py` | Verify output correctness | ### Word TOC Fix Checklist When a generated .docx has broken TOC links: 1. Run `inspect_docx.py` — check `word/document.xml` for bookmark names and link types 2. Verify heading bookmarks use format `_TocH1_XXX` (not `#_TocH1_XXX`) 3. Verify internal links use Word-native anchor style, not external URL rels 4. If `elem.set()` errors appear, use full namespace: `elem.set('{' + W_NS + '}id', value)` 5. Run `fix_final2.py` after any XML edits 6. Test by opening in Word — click TOC entries to verify jump ### Common Errors | Error | Cause | Fix | |-------|-------|-----| | Bookmark name wrong format | Missing `_TocH1_` prefix | Rename via `fix_attrs.py` | | `anchor` contains `#` | Wrong anchor format | Strip `#` prefix | | TOC uses external link rels | Used `http://` link type for internal jump | Switch to `w:instr` field or anchor | | Duplicate headings in body | TOC headings accidentally inserted into body | Use `fix_wrong.py` to clean | | Namespace attribute error | Missing full namespace URI in `elem.set()` | Use `'{' + W_NS + '}id'` form | ### Best Practice - Keep only `build_docx6.py` as the production script - Keep all earlier versions as debugging reference, not for production - One Markdown → one `build_docx6.py` call → inspect → fix if needed FILE:xhsfenxi/scripts/archetype.js #!/usr/bin/env node /** * xhsfenxi — archetype.js * Purpose: Print the archetype classification prompt for a Xiaohongshu creator. * Helps identify whether an account is Type A (荒诞美学), Type B (共鸣命名), * Type C (现实策略), or a hybrid — based on the three proven archetypes * distilled from real analyses of multiple Xiaohongshu creators. * Usage: * node scripts/archetype.js <creator-name> * node scripts/archetype.js <creator-name> --quick */ const args = process.argv.slice(2); const quick = args.includes('--quick'); const name = args.filter(a => !a.startsWith('--'))[0]; if (!name) { console.error('Usage: node scripts/archetype.js <creator-name> [--quick]'); process.exit(1); } if (quick) { console.log(` === xhsfenxi archetype cheatsheet === TYPE A — 荒诞美学型 (Absurdist Aesthetics) Example: 荒诞美学博主（vlogger型） Signals: - Unified brand symbol/tag on every post (e.g. "（劲爆）") - Absurdist or philosophical topics treated earnestly - High-production video; literary title naming - Serious × absurd contrast in every piece Formula: 荒诞滤镜下的生活哲学 TYPE B — 共鸣命名型 (Resonance & Naming) Example: 成长世界观表达博主 Signals: - Posts name vague emotional states or life stages - Private experience → universal proposition - Titles feel like thoughtful questions or redefinitions Formula: 经历 → 命题 → 命名 → 判断 → 共鸣 TYPE C — 现实策略型 (Reality & Strategy) Example: 普通女孩上行策略博主 Signals: - Titles contain conflict words ("骗子", "不要脸", "装") - Content breaks unspoken workplace/relationship/money rules - Self-labels with perceived weakness to pre-empt criticism Formula: 困境 → 说破 → 规则 → 策略 → 爽感 MIXED B+C — Most powerful hybrid Combines: resonance/naming × reality/strategy = "You understand me" + "Now I know what to do" `); process.exit(0); } console.log(` === xhsfenxi archetype classification: ${name} === TASK Classify the Xiaohongshu account "${name}" into one of these three archetypes (or a hybrid), based on visible public evidence. ───────────────────────────────────────────── THREE ARCHETYPES ───────────────────────────────────────────── TYPE A — 荒诞美学型 (Absurdist Aesthetics) Core: Wraps philosophical/serious content in absurdist humor + high visual quality Proven example: 荒诞美学vlogger型 (unified brand symbol; grand × mundane contrast; literary naming) Key signals to look for: - Does every post share a unified recurring symbol or phrase? - Do titles contrast a grand/serious setting with a mundane/absurd action? - Is the visual/production quality noticeably higher than peers? - Are serious topics treated with humor and humor treated earnestly? TYPE B — 共鸣命名型 (Resonance & Naming) Core: Names vague emotions and life stages so users feel "finally, someone said it" Proven example: 成长世界观表达型 (youth growth, worldview expression, concept naming) Key signals to look for: - Do posts give names or definitions to emotional states most people can't articulate? - Does the account turn personal stories into universal life propositions? - Do titles feel like thoughtful philosophical questions or redefinitions? - Is the tone warm, perceptive, aesthetically refined? TYPE C — 现实策略型 (Reality & Strategy) Core: Breaks unspoken real-world rules; provides executable strategies for ordinary people Proven example: 普通女孩上行策略型 (antifragile identity; rule-breaking; 普通人上行) Key signals to look for: - Do titles contain conflict words or counter-intuitive framing? - Does the account self-label with a perceived weakness to pre-empt criticism? - Does content expose hidden rules in workplace/relationships/money/consumption? - Does every piece end with a clear, actionable conclusion or stance? MIXED HYBRID Some accounts combine two archetypes. Most powerful is B+C: "I understand your situation (Type B) AND here's what you can actually do (Type C)" ───────────────────────────────────────────── CLASSIFICATION PROMPT ───────────────────────────────────────────── Based on publicly visible data for "${name}", answer: 1. Which archetype does this account most closely match? (A / B / C / Mixed) 2. What are 2–3 concrete visible signals that support this classification? 3. If mixed: what percentage is each type, and does one dominate? 4. What is the single sentence that describes this account's core "filter" on the world? 5. Is there a brand symbol or signature phrase? If yes, what is it and how does it function? Then proceed with the full analysis using this archetype as the lens. `); FILE:xhsfenxi/scripts/docx-plan.js #!/usr/bin/env node /** * xhsfenxi — docx-plan.js * Purpose: Print the Word generation plan for a completed Xiaohongshu analysis, * including which build_docx*.py script to use, the full workflow steps, * and TOC fix guidance from the workflow archive. * Usage: * node scripts/docx-plan.js <creator-name> * node scripts/docx-plan.js <creator-name> --mode comparison * node scripts/docx-plan.js --toc-fix */ const ARCHIVE_BASE = '~/Desktop/cosmocloud/Deeplumen/cosmowork/openclaw_cosmo/afa/小红书分析与工作流归档/02-Word生成与目录修复脚本'; const args = process.argv.slice(2); const tocFix = args.includes('--toc-fix'); const modeIdx = args.indexOf('--mode'); const mode = modeIdx !== -1 ? args[modeIdx + 1] : 'single'; const name = args.filter((a, i) => { if (a.startsWith('--')) return false; if (modeIdx !== -1 && i === modeIdx + 1) return false; return true; })[0]; if (tocFix) { console.log(` === xhsfenxi Word TOC Fix Guide === Archive scripts location: ${ARCHIVE_BASE}/ STEP-BY-STEP FIX PROCESS 1. Run inspect_docx.py to diagnose the issue: python3 ${ARCHIVE_BASE}/inspect_docx.py <your-file.docx> → Check: bookmark names, link types, heading structure 2. Verify the following are correct: - Bookmark names follow format: _TocH1_XXX (NOT #_TocH1_XXX) - Internal TOC links use Word-native anchor, NOT external http:// rels - No duplicate headings accidentally inserted in document body - Namespace attributes written as: elem.set('{' + W_NS + '}id', value) 3. Run the appropriate fix script: - For attribute/namespace errors: python3 fix_attrs.py <file.docx> - For final TOC + heading cleanup: python3 fix_final2.py <file.docx> - For known wrong output patterns: python3 fix_wrong.py <file.docx> 4. Verify with check_it.py: python3 ${ARCHIVE_BASE}/check_it.py <your-fixed-file.docx> 5. Open in Word and test: click each TOC entry to confirm internal jump works. COMMON ERROR TABLE Bookmark format wrong → rename using fix_attrs.py anchor contains # → strip # prefix in fix_final2.py TOC uses external rels → switch to w:instr field or Word anchor Duplicate headings → fix_wrong.py XML namespace error → use full namespace URI in elem.set() `); process.exit(0); } if (!name && !tocFix) { console.error('Usage: node scripts/docx-plan.js <creator-name> [--mode comparison]'); console.error(' node scripts/docx-plan.js --toc-fix'); process.exit(1); } const isComparison = mode === 'comparison'; const mdFile = isComparison ? `选题公式学习-综合版.md` : `${name}-结构化总结报告.md`; const docxFile = isComparison ? `选题公式学习-综合版.docx` : `${name}-结构化总结报告-商业版.docx`; console.log(` === xhsfenxi Word generation plan: ${name || 'comparison'} === Mode: ${isComparison ? 'comparison / 综合版' : 'single creator'} ───────────────────────────────────────────── STEP 1 — Confirm Markdown is complete ───────────────────────────────────────────── Make sure this file is finalized: ${mdFile} ───────────────────────────────────────────── STEP 2 — Run build_docx6.py (recommended) ───────────────────────────────────────────── Primary script (most stable): python3 ${ARCHIVE_BASE}/build_docx6.py ${mdFile} ${docxFile} If build_docx6.py fails, fall back to earlier versions: build_docx5.py → build_docx4.py → build_docx3.py ───────────────────────────────────────────── STEP 3 — Inspect the output ───────────────────────────────────────────── python3 ${ARCHIVE_BASE}/inspect_docx.py ${docxFile} Check for: ✓ Headings have _TocH1_XXX bookmarks ✓ TOC links are internal (not external URL rels) ✓ No duplicate heading text in body ✓ Page count and section structure look correct ───────────────────────────────────────────── STEP 4 — Fix if needed ───────────────────────────────────────────── Run: node scripts/docx-plan.js --toc-fix for the full TOC fix checklist. Most common fix: python3 ${ARCHIVE_BASE}/fix_final2.py ${docxFile} ───────────────────────────────────────────── STEP 5 — Final check ───────────────────────────────────────────── python3 ${ARCHIVE_BASE}/check_it.py ${docxFile} → Open in Word, click TOC entries, verify internal jumps work ───────────────────────────────────────────── EXPECTED OUTPUT FILES ───────────────────────────────────────────── ${docxFile} ${isComparison ? '' : ` ${name}-结构化总结报告.md (keep as source of truth)\n ${name}-爆款选题公式.md (if also produced)`} ───────────────────────────────────────────── ARCHIVE LOCATION ───────────────────────────────────────────── Scripts: ${ARCHIVE_BASE}/ Stable production script: build_docx6.py Inspection tool: inspect_docx.py `); FILE:xhsfenxi/scripts/intake.js #!/usr/bin/env node /** * xhsfenxi — intake.js * Purpose: Print a structured intake template for a Xiaohongshu creator-analysis task. * Usage: * node scripts/intake.js * node scripts/intake.js --json */ const args = process.argv.slice(2); const asJson = args.includes('--json'); const template = { task: 'xiaohongshu-creator-analysis', creatorNames: ['<creator-name>'], links: ['<homepage-or-search-url>'], screenshots: ['<optional-local-path-or-url>'], deliverables: ['structured-report'], outputDir: '<optional-output-dir>', needWord: false, needBusinessVersion: false, needComparison: false, notes: 'What exactly should be learned from this creator?' }; if (asJson) { console.log(JSON.stringify(template, null, 2)); process.exit(0); } console.log(` === xhsfenxi intake template === Provide or confirm the following before deep research: - creator name(s) - homepage/search URL(s) - screenshots or links for representative posts (optional but helpful) - requested deliverable(s): structured-report / topic-formula / comparison / business-word - output directory - any specific question to answer JSON template: ${JSON.stringify(template, null, 2)} `); FILE:xhsfenxi/scripts/report-plan.js #!/usr/bin/env node /** * xhsfenxi — report-plan.js * Purpose: Print recommended deliverables and filenames for a Xiaohongshu analysis request. * Usage: * node scripts/report-plan.js <creator-name> * node scripts/report-plan.js <creator-a> <creator-b> --mode compare */ const args = process.argv.slice(2); const modeIdx = args.indexOf('--mode'); const mode = modeIdx !== -1 ? args[modeIdx + 1] : 'single'; const names = args.filter((a, i) => { if (a === '--mode') return false; if (modeIdx !== -1 && i === modeIdx + 1) return false; return true; }); if (!names.length) { console.error('Usage: node scripts/report-plan.js <creator-name>'); console.error(' node scripts/report-plan.js <creator-a> <creator-b> --mode compare'); process.exit(1); } if (mode === 'compare') { if (names.length < 2) { console.error('Compare mode requires two creator names.'); process.exit(1); } const [a, b] = names; const files = [ `选题公式学习-综合版.docx` ]; console.log(` === xhsfenxi report plan === Mode: comparison Creators: ${a} / ${b} Recommended deliverables: - comparison report - hybrid topic-formula study - optional business Word version Recommended filenames: ${files.map(f => `- ${f}`).join('\n')} `); process.exit(0); } const name = names[0]; const files = [ `${name}-结构化总结报告.docx`, `${name}-爆款选题公式.docx` ]; console.log(` === xhsfenxi report plan === Mode: single creator Creator: ${name} Recommended deliverables: - structured report - viral topic formula - optional business Word version Recommended filenames: ${files.map(f => `- ${f}`).join('\n')} `); </article></body></html>

--- name: xhsfenxi-pro version: 2.0.0 description: | 小红书全链路分析 skill。数据采集（SeleniumBase XHR拦截）+ 博主深度分析（三型分类 + 五层模型 + hsword内核三段论）+ 爆款选题公式（6模型 + 30选题方向）+ 结构化报告 + 黑体Word交付。整合 xhscosmoskill（采集引擎）+ xhsfenxi（分析框架）+ hsword（实战案例库）。 keywords: - xiaohongshu - 小红书分析 - 博主分析 - 爆款选题公式 - 三型博主 - 选题模型 - Word报告 - 账号拆解 --- # xhs-cosmo — 小红书全链路分析 Skill > 数据采集 × 三型分类 × 内核三段论 × 6模型爆款选题公式 × 黑体Word交付 --- ## 工具路径 ``` LIB_ROOT: /Users/zezedabaobei/Desktop/cosmocloud/Deeplumen/cosmowork/xiaohongshu_new COOKIES: /Users/zezedabaobei/Desktop/cosmocloud/Deeplumen/cosmowork/shopify-marketing/xhs_cookies.json COOKIES_ALT: /Users/zezedabaobei/Desktop/cosmocloud/Deeplumen/cosmowork/xiaohongshu_new/xhs_cookies.json DATA_DIR: xhscosmoskill/data/ (archetypes.json, bloggers.json) HSWORD_REF: openclaw_cosmo/afa/hsword/ (实战案例库) BUILD_DOCX: xhscosmoskill/scripts/build_docx.py ``` --- ## 能力总览 | 层 | 来源 | 功能 | |----|------|------| | **数据采集** | xhscosmoskill | 用户主页笔记、关键词搜索、笔记详情、评论 | | **分析框架** | xhsfenxi | 三型分类、五层账号模型、证据分级 | | **内核三段论** | hsword | 外壳/真正内核/三层人设结构 | | **爆款选题公式** | hsword | 6模型 × 可套用句式 × 30个选题方向 | | **可迁移框架** | hsword | 如何把博主方法论迁移到自己账号 | | **报告生成** | 综合 | 结构化报告 + 爆款选题公式 + 多账号对比 | | **Word交付** | scripts/build_docx.py | 全黑体样式 + 绿色装饰线 | --- ## 三型博主分类系统 ### Type A — 荒诞美学型 - **内核：** 荒诞幽默包裹哲学内核，品牌符号统一（如"（劲爆）"） - **公式：** 荒诞场景 × 品牌符号 × 哲思轻量化 → 审美共鸣 - **代表：** 井越 ### Type B — 共鸣命名型 - **内核：** 私人经历 → 普世命题，给模糊情绪命名，`*` 号品牌符号 - **公式：** 私人场景 × 命题化 × 诗意命名 × `*` 印章 → 普世共鸣 - **代表：** xixiCharon、橘一橙NiceFriend ### Type C — 现实策略型 - **内核：** 打破潜规则，提供可执行向上策略，反体面表达 - **公式：** 困境 → 说破规则 → 提供策略 → 爽感执行 - **代表：** 丑穷女孩陈浪浪 ### 混合型（最强组合） - **B+A：** 既"给情绪命名"又"有审美质感" — xixiCharon - **B+C：** 既"懂你"又"告诉你下一步怎么做" --- ## hsword 内核三段论（必做步骤）每次分析必须明确三层： ``` 外壳是什么？（表面看起来像什么博主） ↓ 真正的内核是什么？（一句话，带""引号的精炼） ↓ 三层人设结构：表层标签 → 身份/场景/标签中层特质 → 性格/能力/气质深层价值观 → 鼓励什么/认可什么/传递什么 ``` **关键洞察：** 真正让账号成立的是第三层。第一层可模仿，第二层可包装，但第三层必须靠长期内容一致才能被用户相信。 --- ## 爆款选题公式体系（6大模型） ### 模型1：场景 × 哲思型（B型均赞最高） ``` [具体地点/场景] + [在这里感受到的哲学状态] + [品牌符号] ``` 机制：地点宏大 × 感受日常 × 语言诗意 = 三重张力 ### 模型2：状态命名型（高收藏率） ``` [擅长/习惯做X的人] + 对[某事物]的感知是[新的理解]的 + [品牌符号] ``` 机制：把混沌状态翻译成可被理解的概念 → 用户"被命名"的满足感 ### 模型3：阶段宣言型（节点必出） ``` [时间节点/年龄阶段] + [这个阶段的成长判断] + [品牌符号] ``` 机制：时间节点触发情绪浓度 × 个人叙事 × 同龄共鸣 ### 模型4：独行宣言型（独立女性共鸣） ``` [我独自/一个人] + [行动] + [反预期结果] ``` 机制：独立女性认同 × 行动力展示 × 反预期 = 三重共鸣 ### 模型5：情绪逆转型（治愈系高收藏） ``` [消极状态] + 其实是/让我明白了 + [正向领悟] + [品牌符号] ``` 机制：情绪低谷共鸣 + 逆转出口 = "被治愈的可能性" ### 模型6：世界观输出型（最强粘性） ``` [持续做X的人/坚持Y的意义] + [是如何理解Z的] + [品牌符号] ``` 机制：用户收藏的是"世界观"，会持续追更 --- ## 完整分析管道（Full Pipeline） ``` Step 0 Cookie 健康检查 ↓ Step 1 解析输入（URL/ID/名称）→ 检查数据库是否已有记录 ↓ Step 2 数据采集（get_user_notes, limit=50, scroll_times=10） ↓ Step 3 基础统计（compute_stats） ↓ Step 4 三型分类（classify_archetype） ↓ Step 5 内核三段论（build_five_layers + 手动补充外壳/内核/深层价值观） ↓ Step 6 爆款选题公式生成（generate_formula_report） ↓ Step 7 生成结构化报告 Markdown（mode='full'） ↓ Step 8 外部文档合并（如有用户提供额外分析文档） ↓ Step 9 写入博主数据库（save_blogger） ↓ Step 10 生成 Word（scripts/build_docx.py） ``` --- ## 数据采集 API ```python import sys sys.path.insert(0, "/Users/zezedabaobei/Desktop/cosmocloud/Deeplumen/cosmowork/xiaohongshu_new") from xhscosmoskill import XhsClient from xhscosmoskill.analyzer import analyze_account, classify_archetype, compute_stats, build_five_layers from xhscosmoskill.formula import generate_formula_report from xhscosmoskill.archetype_registry import save_blogger, list_archetypes, get_blogger COOKIES = "/Users/zezedabaobei/Desktop/cosmocloud/Deeplumen/cosmowork/shopify-marketing/xhs_cookies.json" with XhsClient(cookies_file=COOKIES, headless=True, scroll_times=10) as xhs: notes = xhs.get_user_notes(user_id, limit=50) report = xhs.analyze_account(notes, creator_name=name, mode="full") ``` --- ## 报告生成 API | 函数 | 说明 | |------|------| | `analyze_account(notes, creator_name, mode)` | 主入口，mode: full/formula/snapshot | | `classify_archetype(notes)` | 三型分类，返回类型+置信度 | | `build_five_layers(notes, archetype)` | 五层账号模型 | | `compute_stats(notes)` | 基础统计 | | `generate_formula_report(notes, creator_name, archetype)` | 生成爆款选题公式报告 | --- ## Word 生成 ```python # 单命令生成黑体Word python3 scripts/build_docx.py <subtitle> ``` 或在代码中： ```python from xhscosmoskill.scripts.build_docx import build_word build_word(md_path="/tmp/report.md", out_path="/tmp/report.docx", title="xixiCharon", subtitle="爆款选题公式") ``` --- ## Cookie 管理 ```python # 优先使用（最新） COOKIES = ".../shopify-marketing/xhs_cookies.json" # 备用 COOKIES_ALT = ".../xiaohongshu_new/xhs_cookies.json" # Cookie 健康检查 from xhscosmoskill.utils import check_cookies status = check_cookies(COOKIES) # 返回 {valid: bool, expired_keys: list} ``` **Cookie 过期标志：** notes 返回 ≤ 1 条 → 提示重新运行 `xhs_login.py` --- ## 数据库操作 ```python from xhscosmoskill.archetype_registry import ( save_blogger, # 写入/更新博主记录 get_blogger, # 按名称查询 list_bloggers, # 列出所有博主 list_archetypes, # 查看当前类型库 add_archetype, # 新增自定义类型 update_archetype_signals # 迭代更新类型信号词 ) ``` --- ## 交付物规范 | 文件 | 格式 | 说明 | |------|------|------| | `{博主名}-结构化总结报告.md/.docx` | Markdown + Word | 单账号深度分析（15节）| | `{博主名}-爆款选题公式.md/.docx` | Markdown + Word | 6模型 + 30选题方向 | | `选题公式学习-综合版.md/.docx` | Markdown + Word | 多账号对比 | --- ## 证据分级 | 级别 | 来源 | 使用方式 | |------|------|---------| | A1 | 小红书公开主页可见数据 | 直接陈述 | | A2 | 用户提供截图 | 直接陈述 | | B1 | 第三方公开资料（采访/新榜/百科）| 作为背景补充 | | C1 | 综合推断 | 明确标注为"推断" | --- ## 参考资源 - **实战案例库（hsword）：** `openclaw_cosmo/afa/hsword/` - 井越（Type A）：荒诞美学型完整报告 + 爆款公式 - 橘一橙NiceFriend（Type B）：共鸣命名型完整报告 + 爆款公式 - 丑穷女孩陈浪浪（Type C）：现实策略型完整报告 + 爆款公式 - 选题公式学习-综合版：双系统 + 混合公式 + 6种标题公式 - **分析框架参考：** `references/workflow.md` - **报告模板：** `references/templates.md` - **hsword框架：** `references/hsword-frameworks.md` - **新榜数据：** `https://www.newrank.cn/profile/xiaohongshu/{user_id}` --- ## 已分析博主档案 | 博主 | 类型 | 最高赞 | 均赞 | 分析日期 | |------|------|--------|------|---------| | xixiCharon | B+A | 10万+ | 3,148 | 2026-04-23 | *更多博主将在 `/xhsfx` 调用后自动写入 data/bloggers.json* --- *整合自 xhscosmoskill v1.0 + xhsfenxi v2.1 + hsword实战案例 · 版本 2.0.0 · 2026-04-23* --- ## Purpose & Capability **Xhsfenxi Pro** is a full-stack Xiaohongshu (小红书) blogger analysis skill. It combines automated data collection with a structured deep-analysis framework derived from real-world case studies (hsword archive). | Capability | Description | |-----------|-------------| | Data Collection | Scrape user notes via SeleniumBase XHR interception — no API key required | | Three-Archetype Classification | Classify bloggers as Type A (Absurdist Aesthetics) / B (Resonance Naming) / C (Reality Strategy) or hybrid | | Core Identity Analysis (hsword) | Three-layer persona deconstruction: surface labels / mid-layer traits / deep values | | Viral Topic Formula | 6-model formula system with reusable sentence templates and 30 ready-to-use topic directions | | Structured Reports | Full 15-section analysis report in Markdown | | Word Output | Black Heiti-font Word documents via `scripts/build_docx.py` | | Iterative Archetype DB | `data/archetypes.json` evolves as more bloggers are analyzed | **Does NOT:** - Require any Xiaohongshu API token (uses browser-based XHR interception) - Access private/protected notes or accounts - Store or transmit user credentials - Generate fake engagement data or fabricate analysis results --- ## Instruction Scope **In scope — will handle:** - "Analyze this Xiaohongshu blogger URL" - "Generate viral topic formula for this account" - "What archetype is this blogger?" - "Produce a Word report for this creator" - "Compare two bloggers" - Any `/xhsfx` command invocation **Out of scope — will not handle:** - Accessing private accounts or bypassing platform security - Publishing or posting content to Xiaohongshu on behalf of users - Real-time follower/engagement data (uses public page data only) - Non-Xiaohongshu platforms **When cookies expire:** The browser session cookie file expires approximately every 30 days. If `notes` returns ≤ 1 result, prompt the user to re-run `python3 xhs_login.py` to refresh cookies. The skill will not silently fail — it will detect and report the expired state. --- ## Credentials This skill uses **no API tokens or platform credentials for analysis**. | Action | Credential | Scope | |--------|-----------|-------| | Data collection | Xiaohongshu session cookie (browser-based) | Local file only — `xhs_cookies.json` | | Report generation | None | Local file writes only | | Word output | None | Local file writes only | Cookie file locations (in priority order): 1. `shopify-marketing/xhs_cookies.json` (preferred — most recent) 2. `xiaohongshu_new/xhs_cookies.json` (fallback) **Does NOT hardcode tokens, API keys, or account credentials.** Cookie files are local only and never transmitted. --- ## Persistence & Privilege | Path | Content | When written | |------|---------|-------------| | `data/archetypes.json` | Archetype registry — evolves as bloggers are analyzed | On each `save_blogger()` call | | `data/bloggers.json` | Analyzed blogger database | On each `save_blogger()` call | | `/tmp/{creator}-*.md` | Markdown analysis reports | On report generation | | `/tmp/{creator}-*.docx` | Word documents | On `build_word()` call | **Does NOT write to:** - Any system directories outside the skill directory and `/tmp/` - Shell configuration files (`~/.zshrc` etc.) - Xiaohongshu platform (read-only) **Uninstall:** Delete the skill directory. Cookie files at `xhs_cookies.json` paths can be deleted separately. --- ## Install Mechanism ```bash clawhub install xhsfenxi-pro ``` **Prerequisites:** ```bash pip install seleniumbase python-docx ``` **First-time cookie setup:** ```bash python3 xhs_login.py # Opens browser → log in to Xiaohongshu → cookies saved automatically ``` **Verify installation:** ```python import sys sys.path.insert(0, "/path/to/xhscosmoskill/..") import xhscosmoskill print(xhscosmoskill.__version__) # should print: 2.0.0 from xhscosmoskill import print_cookie_status print_cookie_status() # ✅ 全部有效 / ❌ 已过期 ``` **Quick analysis:** ```python from xhscosmoskill import XhsClient, analyze_account with XhsClient() as xhs: notes = xhs.get_user_notes("USER_ID_HERE", limit=50) report = xhs.analyze_account(notes, creator_name="BloggerName") print(report) ``` FILE:README.md # xhscosmoskill — 小红书数据抓取工具 > Agent 可直接调用的小红书内容抓取客户端。 > 基于 SeleniumBase UC Mode + XHR 拦截，绕过反爬检测，无需逆向签名算法。 --- ## 核心原理 ``` 用户登录一次 → 保存 Cookie ↓ SeleniumBase UC Mode 打开浏览器（绕过 headless 检测） ↓ 注入 JS 拦截器（劫持 fetch/XHR） ↓ 浏览器自动带签名请求小红书 API ↓ 拦截器捕获 API 响应 JSON → 返回结构化数据 ``` 不需要逆向 X-s/X-t 签名，因为我们让浏览器自己完成签名，然后从内存中读取响应。 --- ## 前置步骤（只需一次） ```bash # 1. 安装依赖 pip install seleniumbase # 2. 登录小红书，保存 cookie cd /Users/zezedabaobei/Desktop/cosmocloud/Deeplumen/cosmowork/xiaohongshu_new python3 xhs_login.py # → 浏览器弹出，手动登录，按 Enter 保存 # → 生成 xhs_cookies.json（有效期约 30 天） ``` --- ## 快速使用 ```python import sys sys.path.insert(0, "/Users/zezedabaobei/Desktop/cosmocloud/Deeplumen/cosmowork/xiaohongshu_new") from xhscosmoskill import XhsClient with XhsClient() as xhs: # 搜索关键词 notes = xhs.search("咖啡", limit=20) # 获取用户所有笔记 notes = xhs.get_user_notes("5b07eb49e8ac2b5fe1e7de09", limit=50) # 获取单篇详情（正文 + 评论） note = xhs.get_note_detail("https://www.xiaohongshu.com/explore/xxx") # 批量补充详情 notes = xhs.batch_get_details(notes, delay=2.0) # 保存 xhs.save(notes, "output.json") ``` --- ## API 文档 ### `XhsClient(cookies_file, headless, scroll_times)` | 参数 | 类型 | 默认 | 说明 | |------|------|------|------| | cookies_file | str | `../xhs_cookies.json` | Cookie 文件路径 | | headless | bool | False | 无头模式（稳定后可开启） | | scroll_times | int | 5 | 默认滚动次数（影响加载量） | --- ### `xhs.search(keyword, limit, scroll_times) → List[Note]` 搜索笔记。 ```python notes = xhs.search("咖啡", limit=30) # notes[0].title, notes[0].likes, notes[0].url, notes[0].type ``` --- ### `xhs.get_user_notes(user_id, limit, scroll_times) → List[Note]` 获取用户主页所有笔记。`user_id` 是 URL 中的 hex 字符串。 ```python # URL: xiaohongshu.com/user/profile/5b07eb49e8ac2b5fe1e7de09 notes = xhs.get_user_notes("5b07eb49e8ac2b5fe1e7de09", limit=100) ``` --- ### `xhs.get_note_detail(note_url) → Note | None` 获取单篇笔记详情，包含正文、评论、互动数据。 ```python note = xhs.get_note_detail("https://www.xiaohongshu.com/explore/xxx") print(note.content) # 正文全文 print(note.comments) # List[Comment] ``` --- ### `xhs.batch_get_details(notes, delay) → List[Note]` 在搜索/列表结果基础上批量补充正文和评论。 ```python notes = xhs.search("咖啡", limit=20) notes = xhs.batch_get_details(notes, delay=2.0) # delay 建议 1.5~3 秒 ``` --- ### `xhs.save(notes, filepath)` 保存为 JSON 文件。 --- ## 数据模型 ### `Note` ``` id str 笔记 ID title str 标题 content str 正文全文 url str 完整 URL author str 作者昵称 author_id str 作者 ID likes str 点赞数（字符串，如 "3万"） collects str 收藏数 comments_count str 评论数 type str "video" | "normal" images list 图片 URL 列表 tags list 话题标签列表 comments list List[Comment] ``` ### `Comment` ``` user str 评论者昵称 text str 评论内容 likes str 评论点赞数 time str 评论时间 ``` --- ## 稳定性说明 | 场景 | 稳定性 | 说明 | |------|--------|------| | 搜索列表 | ✅ 稳定 | DOM 兜底，基本必成 | | 用户主页列表 | ✅ 稳定 | 同上 | | 笔记详情正文 | ⚠️ 较好 | API 拦截优先，DOM 兜底 | | 评论数据 | ⚠️ 较好 | 依赖 API 拦截成功 | - Cookie 有效期约 30 天，过期后重跑 `xhs_login.py` - 建议单账号每次请求间隔 1.5 秒以上，避免风控 - 首次使用建议 `headless=False` 观察浏览器行为 --- ## 文件结构 ``` xhscosmoskill/ ├── __init__.py 入口，导出 XhsClient ├── client.py 主客户端类（对外 API） ├── browser.py 浏览器会话 + XHR 拦截器 ├── parser.py API JSON / DOM 解析器 ├── models.py 数据模型（Note / Comment / UserProfile） ├── README.md 本文件 └── examples/ ├── search_notes.py 搜索示例 ├── user_profile.py 用户主页示例 └── note_detail.py 单篇详情示例 ``` --- ## 常见问题 **Q: `Cookie 文件不存在` 报错** A: 运行 `python3 xhs_login.py` 先登录一次。 **Q: content 字段为空** A: 该笔记的详情页 API 未被拦截到。可尝试增加 `wait` 时间，或该笔记需要登录才可见。 **Q: 运行很慢** A: 正常，每次操作需要等待页面渲染。可在稳定后开启 `headless=True`。 **Q: 被封号怎么办** A: 降低频率（`delay=3`），换账号重新 `xhs_login.py`。 FILE:__init__.py from .client import XhsClient from .analyzer import analyze_account, classify_archetype, compute_stats, build_five_layers from .formula import generate_formula_report from .archetype_registry import ( save_blogger, get_blogger, list_bloggers, list_archetypes, add_archetype, update_archetype_signals, ) from .utils import check_cookies, get_best_cookies, print_cookie_status __all__ = [ "XhsClient", "analyze_account", "classify_archetype", "compute_stats", "build_five_layers", "generate_formula_report", "save_blogger", "get_blogger", "list_bloggers", "list_archetypes", "add_archetype", "update_archetype_signals", "check_cookies", "get_best_cookies", "print_cookie_status", ] __version__ = "2.0.0" FILE:_meta.json { "ownerId": "kn79bebfnwg15sb0g7cj5z5nyd83gxh0", "slug": "xhsfenxi-pro", "version": "2.0.0", "publishedAt": null } FILE:analyzer.py """ analyzer.py — 小红书博主分析引擎整合 xhsfenxi 三型博主分类体系 + 五层账号模型可与 XhsClient 采集结果直接对接用法: from xhscosmoskill import XhsClient from xhscosmoskill.analyzer import analyze_account with XhsClient() as xhs: notes = xhs.get_user_notes("user_id", limit=50) report = analyze_account(notes, creator_name="xixiCharon") print(report) """ from __future__ import annotations from typing import List, Dict, Any import re from collections import Counter # ── 三型博主分类系统 ────────────────────────────────────────── ARCHETYPE_SIGNALS = { "A": { "name": "荒诞美学型", "desc": "用荒诞幽默包裹哲学内核，高质感视觉，品牌符号统一", "title_signals": ["（劲爆）", "劲爆", "＊", "*", "·"], "content_signals": ["哲", "美学", "荒诞", "飞", "高山", "宇宙", "奖励", "永恒"], "commercial": "高端生活方式、旅行、摄影、轻奢品牌", "difficulty": "高 — 依赖长期审美积累", }, "B": { "name": "共鸣命名型", "desc": "把私人经历转化为普世命题，给模糊情绪命名", "title_signals": ["*", "为什么", "是什么", "原来", "才知道", "命题", "灯塔", "解药"], "content_signals": ["成长", "情绪", "留学", "记录", "年末", "不确定", "命题", "感受", "理解", "接纳"], "commercial": "成长/教育/创意平台、中端生活方式", "difficulty": "中 — 方法可学，需要真实观察力", }, "C": { "name": "现实策略型", "desc": "打破职场/人际/金钱潜规则，提供可执行的向上策略", "title_signals": ["骗子", "不要脸", "装", "穷", "规则", "策略", "大法"], "content_signals": ["策略", "规则", "执行", "调整", "快速", "现实", "向上", "方法"], "commercial": "大众消费、职场、电商、实用工具", "difficulty": "中 — 框架可学，切忌复制表面攻击性", }, } def classify_archetype(notes: List[Any]) -> Dict[str, Any]: """ 基于笔记标题关键词，初步判断博主类型。返回: { "type": "B", "name": "共鸣命名型", "scores": {...}, "rationale": [...] } """ scores = {"A": 0, "B": 0, "C": 0} evidence = {"A": [], "B": [], "C": []} for note in notes: title = note.title or "" if hasattr(note, "title") else note.get("title", "") content = note.content or "" if hasattr(note, "content") else note.get("content", "") text = title + " " + content for typ, cfg in ARCHETYPE_SIGNALS.items(): for sig in cfg["title_signals"]: if sig in title: scores[typ] += 2 evidence[typ].append(f"标题含「{sig}」: {title[:30]}") for sig in cfg["content_signals"]: if sig in text: scores[typ] += 1 # 归一化 total = sum(scores.values()) or 1 pct = {k: round(v / total * 100) for k, v in scores.items()} best = max(scores, key=scores.get) second = sorted(scores, key=scores.get, reverse=True)[1] # 混合判断 is_mixed = scores[second] >= scores[best] * 0.6 if is_mixed: archetype_type = f"{best}+{second}" archetype_name = f"{ARCHETYPE_SIGNALS[best]['name']} × {ARCHETYPE_SIGNALS[second]['name']}" else: archetype_type = best archetype_name = ARCHETYPE_SIGNALS[best]["name"] rationale = evidence[best][:3] + (evidence[second][:2] if is_mixed else []) return { "type": archetype_type, "name": archetype_name, "desc": ARCHETYPE_SIGNALS[best]["desc"], "scores": scores, "pct": pct, "rationale": rationale, "commercial": ARCHETYPE_SIGNALS[best]["commercial"], "difficulty": ARCHETYPE_SIGNALS[best]["difficulty"], "is_mixed": is_mixed, } # ── 数据工具 ────────────────────────────────────────────────── def _parse_likes(v) -> int: if not v: return 0 s = str(v).replace(",", "").strip() if "万" in s: return int(float(s.replace("万", "")) * 10000) try: return int(s) except ValueError: return 0 def _engagement_tier(likes: int) -> str: if likes >= 50000: return "爆款 🔥" if likes >= 10000: return "高量 ⭐" if likes >= 5000: return "中高量" if likes >= 1000: return "稳定量" return "普通" # ── 五层账号模型 ────────────────────────────────────────────── def build_five_layers(notes: List[Any], archetype: Dict) -> Dict[str, Any]: """ 提取五层账号模型: 1. Identity — 博主是谁（人设定位） 2. Contract — 用户为什么关注（情感契约） 3. Topic — 反复覆盖的问题/欲望（选题体系） 4. Expression — 标题/开头公式、品牌符号 5. Transfer — 可学习点 vs 不能硬抄 """ titles = [] likes_list = [] for n in notes: t = n.title if hasattr(n, "title") else n.get("title", "") lk = _parse_likes(n.likes if hasattr(n, "likes") else n.get("likes")) if t: titles.append((t, lk)) likes_list.append(lk) # 选题体系：基于标题词频聚类 topic_keywords = { "自然/户外": ["山", "自然", "户外", "攀岩", "跑", "溪", "飞", "森林", "海"], "留学/海外": ["留学", "巴黎", "欧洲", "外国", "服装生", "独居", "课后"], "情绪/成长": ["情绪", "成长", "失败", "不确定", "淤青", "消极", "烦恼", "困境"], "旅行/探索": ["旅行", "solotrip", "solo", "gap", "世界", "探索"], "生活哲学": ["生活", "命题", "本质", "人生", "记录", "计划", "感受", "存在", "宇宙"], "创作/记录": ["记录", "vlog", "日记", "分享", "创造力"], } topic_dist: Dict[str, List] = {k: [] for k in topic_keywords} for title, lk in titles: for topic, kws in topic_keywords.items(): if any(kw in title for kw in kws): topic_dist[topic].append((title, lk)) best_topic = max(topic_dist, key=lambda k: sum(x[1] for x in topic_dist[k]) / max(len(topic_dist[k]), 1)) # 标题模式分析 has_asterisk = sum(1 for t, _ in titles if "*" in t or "＊" in t) has_question = sum(1 for t, _ in titles if "为什么" in t or "如何" in t or "？" in t) has_contrast = sum(1 for t, _ in titles if "但" in t or "，" in t) has_emotion_label = sum(1 for t, _ in titles if any(w in t for w in ["模糊", "流动", "感受", "治愈", "珍贵"])) # 爆款标题 top_titles = sorted(titles, key=lambda x: -x[1])[:5] return { "identity": _infer_identity(titles, archetype), "contract": _infer_contract(archetype), "topic_system": {k: len(v) for k, v in topic_dist.items() if v}, "best_topic": best_topic, "best_topic_avg": round( sum(x[1] for x in topic_dist[best_topic]) / max(len(topic_dist[best_topic]), 1) ), "expression": { "asterisk_ratio": f"{has_asterisk}/{len(titles)}", "question_titles": has_question, "contrast_titles": has_contrast, "emotion_labels": has_emotion_label, "top_titles": top_titles, }, "transfer": _infer_transfer(archetype), } def _infer_identity(titles, archetype) -> str: typ = archetype["type"][0] if typ == "A": return "用荒诞美学构建独特视角的创作者，以品牌符号建立强识别度" if typ == "B": return "把私人经历转化为公共命题、擅长给模糊情绪命名的思考者" return "洞察现实规则、提供可执行策略的实用派创作者" def _infer_contract(archetype) -> str: typ = archetype["type"][0] if typ == "A": return "看了就愉快 + 被美学震撼 + 看到了另一种看待生活的方式" if typ == "B": return "被理解 + 获得命名 + 对自己的状态有了新的视角" return "被验证 + 被激活 + 获得可执行的向上移动路径" def _infer_transfer(archetype) -> Dict[str, List[str]]: typ = archetype["type"][0] if typ == "A": return { "可学": ["建立自己的品牌符号", "练习反差命名而非描述地点", "荒诞框架选题"], "不能硬抄": ["视觉审美需长期积累", "语气独特性不可复制"], } if typ == "B": return { "可学": ["命题化思维：私人经历→公共命题", "建立个人概念词汇库", "五步公式：经历→命题→命名→判断→共鸣"], "不能硬抄": ["真实性是基础，无法凭空虚构", "语气温度需与真实人设一致"], } return { "可学": ["找到自己的现实母题", "冲突词标题练习", "公式：困境→说破→规则→策略→爽感"], "不能硬抄": ["表面攻击性语气", "特定人设的羞耻感先夺法"], } # ── 数据统计 ────────────────────────────────────────────────── def compute_stats(notes: List[Any]) -> Dict[str, Any]: likes = [_parse_likes(n.likes if hasattr(n, "likes") else n.get("likes")) for n in notes] types = Counter( (n.type if hasattr(n, "type") else n.get("type", "unknown")) for n in notes ) brackets = {"1万+": 0, "5000-9999": 0, "1000-4999": 0, "500-999": 0, "<500": 0} for lk in likes: if lk >= 10000: brackets["1万+"] += 1 elif lk >= 5000: brackets["5000-9999"] += 1 elif lk >= 1000: brackets["1000-4999"] += 1 elif lk >= 500: brackets["500-999"] += 1 else: brackets["<500"] += 1 top = sorted( [(n.title if hasattr(n, "title") else n.get("title", ""), _parse_likes(n.likes if hasattr(n, "likes") else n.get("likes"))) for n in notes], key=lambda x: -x[1], )[:10] return { "total": len(notes), "video_count": types.get("video", 0), "normal_count": types.get("normal", 0), "total_likes": sum(likes), "avg_likes": round(sum(likes) / len(likes)) if likes else 0, "max_likes": max(likes) if likes else 0, "brackets": brackets, "top10": top, } # ── 主入口 ──────────────────────────────────────────────────── def analyze_account( notes: List[Any], creator_name: str = "未知博主", mode: str = "full", ) -> str: """ 对采集到的笔记列表做完整分析，返回 Markdown 报告。参数: notes — XhsClient.get_user_notes() 的返回值 creator_name — 博主名称 mode — 'full' 完整报告 | 'formula' 只输出选题公式 | 'snapshot' 快速摘要 """ stats = compute_stats(notes) archetype = classify_archetype(notes) five = build_five_layers(notes, archetype) if mode == "snapshot": return _render_snapshot(creator_name, stats, archetype) if mode == "formula": return _render_formula(creator_name, stats, archetype, five) return _render_full_report(creator_name, stats, archetype, five) # ── 渲染函数 ────────────────────────────────────────────────── def _render_snapshot(name, stats, archetype) -> str: return f"""# {name} — 快速摘要 | 指标 | 数值 | |------|------| | 分析笔记 | {stats['total']} 篇 | | 视频占比 | {stats['video_count']}/{stats['total']} | | 平均点赞 | {stats['avg_likes']:,} | | 最高单篇 | {stats['max_likes']:,} | | 博主类型 | **{archetype['name']}** ({archetype['type']}) | | 商业适配 | {archetype['commercial']} | """ def _render_formula(name, stats, archetype, five) -> str: top_topic = five["best_topic"] top_avg = five["best_topic_avg"] transfer = five["transfer"] top_titles_md = "\n".join( f"- 👍{lk:,} {t[:50]}" for t, lk in five["expression"]["top_titles"] ) can_learn = "\n".join(f"- ✅ {x}" for x in transfer["可学"]) cant_copy = "\n".join(f"- ❌ {x}" for x in transfer["不能硬抄"]) return f"""# {name} — 爆款选题公式 > 类型：{archetype['name']} | 平均点赞：{stats['avg_likes']:,} | 最高：{stats['max_likes']:,} ## 一、为什么能做出爆款？博主类型：**{archetype['name']}** {archetype['desc']} 核心公式：{_get_core_formula(archetype['type'][0])} ## 二、最强选题方向最高互动主题：**{top_topic}**（均赞 {top_avg:,}） ## 三、爆款标题参考 {top_titles_md} ## 四、标题模式 - 含 * 符号：{five['expression']['asterisk_ratio']}（情感标记） - 疑问句式：{five['expression']['question_titles']} 篇 - 情绪词命名：{five['expression']['emotion_labels']} 篇 ## 五、可学 / 不能硬抄 {can_learn} {cant_copy} """ def _render_full_report(name, stats, archetype, five) -> str: topic_table = "\n".join( f"| {t} | {c}篇 |" for t, c in sorted(five["topic_system"].items(), key=lambda x: -x[1]) ) brackets_md = "\n".join( f"| {k} | {'█' * v} {v}篇 |" for k, v in stats["brackets"].items() ) top10_md = "\n".join( f"| {i+1} | 👍{lk:,} | {_engagement_tier(lk)} | {t[:45]} |" for i, (t, lk) in enumerate(stats["top10"]) ) rationale_md = "\n".join(f"- {r}" for r in archetype["rationale"]) or "- 基于内容风格综合判断" can_learn = "\n".join(f"- ✅ {x}" for x in five["transfer"]["可学"]) cant_copy = "\n".join(f"- ❌ {x}" for x in five["transfer"]["不能硬抄"]) return f"""# {name} — 结构化分析报告 > 数据来源：小红书公开主页，样本量 {stats['total']} 篇 --- ## 一、账号快照 | 指标 | 数值 | |------|------| | 分析样本 | {stats['total']} 篇 | | 视频/图文 | {stats['video_count']} / {stats['normal_count']} | | 总点赞 | {stats['total_likes']:,} | | 平均点赞 | {stats['avg_likes']:,} | | 最高单篇 | {stats['max_likes']:,} | | 爆款（1万+） | {stats['brackets']['1万+']} 篇 | --- ## 二、账号类型判断 **类型：** {archetype['name']} ({archetype['type']}) **核心内核：** {archetype['desc']} **判断依据：** {rationale_md} --- ## 三、人设定位（Identity） {five['identity']} --- ## 四、用户情感契约（Audience Contract）用户关注这个博主，是为了： {five['contract']} --- ## 五、选题体系（Topic System） | 主题方向 | 篇数 | |----------|------| {topic_table} **最强方向：** {five['best_topic']}（均赞 {five['best_topic_avg']:,}） --- ## 六、表达体系（Expression System） | 维度 | 数据 | |------|------| | 含 * 情感符号 | {five['expression']['asterisk_ratio']} 篇 | | 疑问句标题 | {five['expression']['question_titles']} 篇 | | 情绪词命名 | {five['expression']['emotion_labels']} 篇 | **核心公式：** {_get_core_formula(archetype['type'][0])} --- ## 七、高互动内容 Top 10 | # | 点赞 | 量级 | 标题 | |---|------|------|------| {top10_md} --- ## 八、点赞分布 | 段位 | 分布 | |------|------| {brackets_md} --- ## 九、商业化判断 - **适合方向：** {archetype['commercial']} - **复制难度：** {archetype['difficulty']} --- ## 十、可学 / 不能硬抄 **可以学习：** {can_learn} **不能硬抄：** {cant_copy} --- ## 十一、后续建议 - 路径 A：深入拆解 3-5 篇高互动笔记的正文结构 - 路径 B：制作专项爆款选题公式（`mode='formula'`） - 路径 C：引入对标账号做多账号综合对比 """ def _get_core_formula(typ: str) -> str: formulas = { "A": "荒诞场景 × 品牌符号 × 哲思轻量化 → 审美共鸣", "B": "私人经历 → 命题化 → 给模糊状态命名 → 普世共鸣", "C": "困境 → 说破规则 → 提供策略 → 爽感执行", } return formulas.get(typ, "内容积累 → 风格稳定 → 用户粘性") FILE:archetype_registry.py """ archetype_registry.py — 可迭代博主类型注册表随分析案例积累，自动更新类型数据库，支持添加新类型 """ from __future__ import annotations import json import os from datetime import date from typing import Dict, Any, Optional _BASE = os.path.dirname(__file__) ARCHETYPES_FILE = os.path.join(_BASE, "data", "archetypes.json") BLOGGERS_FILE = os.path.join(_BASE, "data", "bloggers.json") # ── 读写工具 ────────────────────────────────────────────────── def _load(path: str) -> dict: with open(path, encoding="utf-8") as f: return json.load(f) def _save(path: str, data: dict): with open(path, "w", encoding="utf-8") as f: json.dump(data, f, ensure_ascii=False, indent=2) # ── 类型注册表操作 ──────────────────────────────────────────── def load_archetypes() -> Dict[str, Any]: """加载当前所有类型（内置 + 自定义）""" db = _load(ARCHETYPES_FILE) merged = {**db["archetypes"], **db.get("custom_archetypes", {})} return merged def add_archetype(key: str, name: str, desc: str, formula: str, title_signals: list, content_signals: list, commercial: str, difficulty: str = "中") -> str: """ 添加新博主类型（当现有三型无法覆盖时）参数: key — 类型代码，如 "D"、"E" 或自定义字符串 name — 类型名，如 "知识科普型" ... 返回: 成功提示 """ db = _load(ARCHETYPES_FILE) if key in db["archetypes"]: return f"⚠️ 类型 {key} 已存在于内置类型，请用其他 key" db["custom_archetypes"][key] = { "name": name, "desc": desc, "formula": formula, "title_signals": title_signals, "content_signals": content_signals, "commercial": commercial, "difficulty": difficulty, "examples": [], "confirmed_by": 0, } db["_meta"]["last_updated"] = str(date.today()) _save(ARCHETYPES_FILE, db) return f"✅ 新增类型 {key}「{name}」" def update_archetype_signals(key: str, new_title_signals: list = None, new_content_signals: list = None) -> str: """从新分析的博主身上迭代更新信号词""" db = _load(ARCHETYPES_FILE) target = db["archetypes"].get(key) or db["custom_archetypes"].get(key) if not target: return f"❌ 类型 {key} 不存在" if new_title_signals: existing = set(target["title_signals"]) added = [s for s in new_title_signals if s not in existing] target["title_signals"].extend(added) if new_content_signals: existing = set(target["content_signals"]) added = [s for s in new_content_signals if s not in existing] target["content_signals"].extend(added) target["confirmed_by"] = target.get("confirmed_by", 0) + 1 db["_meta"]["last_updated"] = str(date.today()) _save(ARCHETYPES_FILE, db) return f"✅ 类型 {key} 信号词已更新" def list_archetypes() -> str: """打印当前所有类型（用于 /xhsfx 开头展示）""" db = _load(ARCHETYPES_FILE) lines = [f"📊 当前博主类型库（{db['_meta']['last_updated']} 更新）\n"] for key, cfg in db["archetypes"].items(): ex = f" — 案例: {', '.join(cfg['examples'][:2])}" if cfg["examples"] else "" lines.append(f" **{key}** {cfg['name']} × {cfg['confirmed_by']}个案例{ex}") custom = db.get("custom_archetypes", {}) if custom: lines.append("\n **自定义类型：**") for key, cfg in custom.items(): lines.append(f" **{key}** {cfg['name']} × {cfg['confirmed_by']}个案例") return "\n".join(lines) # ── 博主数据库操作 ──────────────────────────────────────────── def save_blogger(creator_name: str, user_id: str, archetype: dict, stats: dict, best_topic: str, best_topic_avg: int, tags: list = None, formula: str = "") -> str: """ 分析完成后写入博主数据库，并更新对应类型的案例数 """ db = _load(BLOGGERS_FILE) # 检查是否已存在 existing = next((b for b in db["bloggers"] if b["user_id"] == user_id), None) if existing: existing.update({ "analyzed_at": str(date.today()), "archetype": archetype["type"], "archetype_name": archetype["name"], "stats": stats, "best_topic": best_topic, "best_topic_avg": best_topic_avg, }) action = "更新" else: db["bloggers"].append({ "creator_name": creator_name, "user_id": user_id, "analyzed_at": str(date.today()), "archetype": archetype["type"], "archetype_name": archetype["name"], "stats": stats, "best_topic": best_topic, "best_topic_avg": best_topic_avg, "tags": tags or [], "formula": formula, }) db["_meta"]["total"] = len(db["bloggers"]) action = "新增" _save(BLOGGERS_FILE, db) # 同步更新类型案例 _update_archetype_example(archetype["type"][0], creator_name) return f"✅ 博主「{creator_name}」已{action}到数据库（共 {db['_meta']['total']} 位）" def _update_archetype_example(type_key: str, creator_name: str): """把博主名写入对应类型的 examples 列表""" db = _load(ARCHETYPES_FILE) target = db["archetypes"].get(type_key) or db["custom_archetypes"].get(type_key) if target and creator_name not in target["examples"]: target["examples"].append(creator_name) target["confirmed_by"] = len(target["examples"]) db["_meta"]["total_analyzed"] = sum( len(v["examples"]) for v in {**db["archetypes"], **db.get("custom_archetypes", {})}.values() ) _save(ARCHETYPES_FILE, db) def list_bloggers(archetype_filter: str = None) -> str: """列出已分析博主，可按类型过滤""" db = _load(BLOGGERS_FILE) bloggers = db["bloggers"] if archetype_filter: bloggers = [b for b in bloggers if archetype_filter in b["archetype"]] lines = [f"📋 已分析博主（共 {len(bloggers)} 位）\n"] for b in bloggers: lines.append( f" · **{b['creator_name']}** [{b['archetype_name']}] " f"均赞{b['stats'].get('avg_likes',0):,} 最高{b['stats'].get('max_likes',0):,} " f"| {b['analyzed_at']}" ) return "\n".join(lines) def get_blogger(creator_name: str) -> Optional[dict]: """按名称查找已分析博主记录""" db = _load(BLOGGERS_FILE) return next((b for b in db["bloggers"] if b["creator_name"] == creator_name), None) FILE:browser.py """ 浏览器引擎：SeleniumBase UC Mode + XHR 拦截核心思路：注入 JS 拦截 fetch/XHR，捕获小红书真实 API 响应 JSON 不需要逆向签名算法，直接从浏览器内存中读取已签名的响应 """ import json import time import os from seleniumbase import SB # 注入到页面的 JS 拦截器 _INTERCEPT_JS = """ (function() { if (window.__xhs_interceptor_installed) return; window.__xhs_interceptor_installed = true; window.__xhs_api_cache = {}; // 拦截 fetch const _fetch = window.fetch; window.fetch = async function(...args) { const url = typeof args[0] === 'string' ? args[0] : args[0]?.url || ''; const res = await _fetch.apply(this, args); if (url.includes('/api/sns/') || url.includes('edith.xiaohongshu')) { try { const clone = res.clone(); clone.json().then(data => { window.__xhs_api_cache[url] = data; }).catch(() => {}); } catch(e) {} } return res; }; // 拦截 XHR const _open = XMLHttpRequest.prototype.open; const _send = XMLHttpRequest.prototype.send; XMLHttpRequest.prototype.open = function(method, url) { this.__url = url; return _open.apply(this, arguments); }; XMLHttpRequest.prototype.send = function() { this.addEventListener('load', function() { if (this.__url && (this.__url.includes('/api/sns/') || this.__url.includes('edith.xiaohongshu'))) { try { window.__xhs_api_cache[this.__url] = JSON.parse(this.responseText); } catch(e) {} } }); return _send.apply(this, arguments); }; })(); """ class BrowserSession: def __init__(self, cookies_file: str, headless: bool = False): self.cookies_file = cookies_file self.headless = headless self._sb = None self._ctx = None def __enter__(self): self._ctx = SB(uc=True, headless2=self.headless) self._sb = self._ctx.__enter__() self._init() return self def __exit__(self, *args): if self._ctx: self._ctx.__exit__(*args) def _init(self): self._sb.uc_open_with_reconnect("https://www.xiaohongshu.com", 4) time.sleep(2) self._inject_cookies() # Register CDP script BEFORE refresh so it runs on reload self._register_cdp_interceptor() self._sb.refresh() time.sleep(2) self._sb.execute_script(_INTERCEPT_JS) def _inject_cookies(self): if not os.path.exists(self.cookies_file): raise FileNotFoundError(f"Cookie 文件不存在: {self.cookies_file}\n请先运行 xhs_login.py") with open(self.cookies_file, encoding="utf-8") as f: cookies = json.load(f) for c in cookies: c.pop("sameSite", None) c.pop("httpOnly", None) try: self._sb.driver.add_cookie(c) except Exception: pass def _register_cdp_interceptor(self): """注册 CDP onNewDocument 脚本，每次页面加载前执行（捕获 page-load API）""" try: self._sb.driver.execute_cdp_cmd("Page.addScriptToEvaluateOnNewDocument", { "source": _INTERCEPT_JS }) except Exception: pass def _install_interceptor(self): self._sb.execute_script(_INTERCEPT_JS) def navigate(self, url: str, wait: float = 3.0): """导航到页面：用 driver.get 保留 CDP 会话（使拦截器在 page-load 前运行）""" # Re-register in case CDP session was reset self._register_cdp_interceptor() # Use driver.get to keep CDP alive (interceptor runs before page JS) self._sb.driver.get(url) time.sleep(wait) # Also inject immediately for any late-arriving race conditions self._install_interceptor() def scroll_and_collect(self, times: int = 3, gap: float = 2.0): """滚动页面触发懒加载 API""" for _ in range(times): self._sb.execute_script("window.scrollBy(0, 1500)") time.sleep(gap) def get_api_cache(self) -> dict: """读取所有已拦截的 API 响应""" try: return self._sb.execute_script("return window.__xhs_api_cache || {}") or {} except Exception: return {} def clear_cache(self): self._sb.execute_script("window.__xhs_api_cache = {}") def get_page_source(self) -> str: return self._sb.get_page_source() def find_elements(self, selector: str): return self._sb.find_elements(selector) def execute_script(self, script: str): return self._sb.execute_script(script) @property def driver(self): return self._sb.driver FILE:client.py """ XhsClient — 小红书数据抓取 + 分析客户端基于 SeleniumBase UC Mode + XHR 拦截，无需逆向签名算法整合 xhsfenxi 三型博主分类体系 + 五层账号模型用法: from xhscosmoskill import XhsClient with XhsClient() as xhs: notes = xhs.search("咖啡", limit=30) user_notes = xhs.get_user_notes("5b07eb49e8ac2b5fe1e7de09") report = xhs.analyze_account(user_notes, creator_name="博主名") detail = xhs.get_note_detail("https://www.xiaohongshu.com/explore/xxx") """ import json import time import os from typing import List, Optional from .browser import BrowserSession from .models import Note, UserProfile from .parser import ( parse_search_response, parse_user_notes_response, parse_note_detail_response, parse_comment_page_response, parse_dom_notes, ) from .analyzer import analyze_account as _analyze_account DEFAULT_COOKIES = os.path.join(os.path.dirname(__file__), "..", "xhs_cookies.json") class XhsClient: """ 小红书数据抓取客户端参数: cookies_file: cookie 文件路径（运行 xhs_login.py 生成） headless: 是否无头模式（默认 False，首次建议 False 观察行为） scroll_times: 每次操作滚动次数（影响加载数量）支持 context manager（with 语句）或手动 open()/close() """ def __init__( self, cookies_file: str = DEFAULT_COOKIES, headless: bool = True, scroll_times: int = 5, ): self.cookies_file = os.path.abspath(cookies_file) self.headless = headless self.scroll_times = scroll_times self._session: Optional[BrowserSession] = None def open(self): self._session = BrowserSession(self.cookies_file, self.headless) self._session.__enter__() return self def close(self): if self._session: self._session.__exit__(None, None, None) self._session = None def __enter__(self): return self.open() def __exit__(self, *args): self.close() # ─── 公开接口 ────────────────────────────────────────────── def search(self, keyword: str, limit: int = 20, scroll_times: int = None) -> List[Note]: """ 搜索笔记参数: keyword: 搜索关键词 limit: 最大返回数量 scroll_times: 滚动次数（越多加载越多，默认用初始化值）返回: List[Note] """ self._ensure_session() url = f"https://www.xiaohongshu.com/search_result?keyword={keyword}&source=web_explore_feed" self._session.clear_cache() self._session.navigate(url, wait=3) self._session.scroll_and_collect(scroll_times or self.scroll_times) cache = self._session.get_api_cache() notes = parse_search_response(cache) # API 拦截失败时 fallback 到 DOM if not notes: notes = parse_dom_notes(self._session) return notes[:limit] def get_user_notes(self, user_id: str, limit: int = 50, scroll_times: int = None) -> List[Note]: """ 获取用户主页所有笔记参数: user_id: 小红书用户 ID（URL 中的那串 hex） limit: 最大返回数量 scroll_times: 滚动次数返回: List[Note] """ self._ensure_session() url = f"https://www.xiaohongshu.com/user/profile/{user_id}" self._session.clear_cache() self._session.navigate(url, wait=4) seen_ids = set() notes = [] n_scroll = scroll_times or self.scroll_times for _ in range(n_scroll): self._session.execute_script("window.scrollBy(0, 1500)") time.sleep(2) cache = self._session.get_api_cache() batch = parse_user_notes_response(cache) # DOM fallback if API empty if not batch: batch = parse_dom_notes(self._session) for n in batch: key = n.id or n.url or n.title if key and key not in seen_ids: seen_ids.add(key) notes.append(n) if len(notes) >= limit: break return notes[:limit] def get_note_detail(self, note_url: str) -> Optional[Note]: """ 获取单篇笔记详情（正文、评论、互动数据）参数: note_url: 笔记完整 URL 返回: Note 或 None """ self._ensure_session() self._session.clear_cache() self._session.navigate(note_url, wait=4) cache = self._session.get_api_cache() note = parse_note_detail_response(cache) # API 失败时从 DOM 提取正文和标题 if not note: note = Note() note.url = note_url for sel in ["#detail-desc", "span#detail-desc", ".desc.expand", "[class*='desc']"]: try: el = self._session.find_elements(sel) if el and el[0].text.strip(): note.content = el[0].text.strip() break except Exception: pass for sel in ["#detail-title", ".note-content .title", "h1", "[class*='title']"]: try: el = self._session.find_elements(sel) if el and el[0].text.strip(): note.title = el[0].text.strip() break except Exception: pass # 无论 API 是否成功，都尝试从 comment/page 补充评论 if not note.comments: note.comments = parse_comment_page_response(cache) return note if (note and (note.content or note.title)) else None def post_comment(self, note_url: str, text: str) -> bool: """ 在笔记下发表评论参数: note_url: 笔记完整 URL text: 评论内容返回: True 成功 / False 失败 """ self._ensure_session() self._session.navigate(note_url, wait=4) try: from selenium.webdriver.common.action_chains import ActionChains from selenium.webdriver.common.keys import Keys driver = self._session.driver # 1. JS 点击评论区占位符，激活输入框 activate_selectors = [ ".comment-input-inner-container", "[class*='comment-input']", ".input-inner-container", "[class*='commentInput']", "[class*='input-box']", ] activated = False for sel in activate_selectors: els = self._session.find_elements(sel) if els: driver.execute_script("arguments[0].scrollIntoView({block:'center'})", els[0]) time.sleep(0.5) driver.execute_script("arguments[0].click()", els[0]) activated = True time.sleep(1) break if not activated: print(" ✗ 找不到评论输入框") return False # 2. 找激活后的 contenteditable 或 textarea editable_selectors = [ "[contenteditable='true']", "textarea", "[class*='ql-editor']", "[class*='editor']", ] editable = None for sel in editable_selectors: els = self._session.find_elements(sel) if els: editable = els[0] break if editable: ActionChains(driver).move_to_element(editable).click().send_keys(text).perform() else: # fallback: 直接用 active element ActionChains(driver).send_keys(text).perform() time.sleep(1) # 3. 点击发送按钮 submit_selectors = [ "[class*='submit']", "[class*='send']", "button.submit", "[class*='confirm']", ] submitted = False for sel in submit_selectors: els = self._session.find_elements(sel) # 过滤不可见元素 for el in els: try: if el.is_displayed() and el.is_enabled(): driver.execute_script("arguments[0].click()", el) submitted = True break except Exception: continue if submitted: break if not submitted: # 最后 fallback: Enter ActionChains(driver).send_keys(Keys.ENTER).perform() time.sleep(2) print(f" ✓ 评论已发送: {text[:30]}") return True except Exception as e: print(f" ✗ 评论失败: {e}") return False def get_note_comments(self, note_url: str, limit: int = 20) -> List[dict]: """ 单独获取笔记评论参数: note_url: 笔记 URL limit: 最大评论数返回: List[dict] 含 user/text/likes 字段 """ note = self.get_note_detail(note_url) if note: return [c.__dict__ for c in note.comments[:limit]] return [] def batch_get_details(self, notes: List[Note], delay: float = 1.5) -> List[Note]: """ 批量获取笔记详情（在已有列表基础上补充正文和评论）参数: notes: search() 或 get_user_notes() 返回的列表 delay: 每篇请求间隔秒数（建议 1.5~3）返回: 补充了 content/comments 的 Notes 列表 """ enriched = [] for i, note in enumerate(notes): if not note.url: enriched.append(note) continue print(f" [{i+1}/{len(notes)}] {note.title[:30] or note.url[:40]}") detail = self.get_note_detail(note.url) if detail: detail.title = detail.title or note.title detail.likes = detail.likes or note.likes detail.type = detail.type or note.type enriched.append(detail) else: enriched.append(note) time.sleep(delay) return enriched # ─── 工具方法 ─────────────────────────────────────────────── def save(self, notes: List[Note], filepath: str): """保存结果到 JSON 文件""" data = [n.to_dict() for n in notes] with open(filepath, "w", encoding="utf-8") as f: json.dump(data, f, ensure_ascii=False, indent=2) print(f"已保存 {len(data)} 条到 {filepath}") def analyze_account( self, notes: List[Note], creator_name: str = "未知博主", mode: str = "full", save_to: str = None, ) -> str: """ 对已采集的笔记列表做完整分析，返回 Markdown 报告。参数: notes — get_user_notes() 的返回值 creator_name — 博主昵称（用于报告标题） mode — 'full' 完整报告 | 'formula' 选题公式 | 'snapshot' 快速摘要 save_to — 可选，保存报告到文件路径（.md）返回: Markdown 字符串 """ report = _analyze_account(notes, creator_name=creator_name, mode=mode) if save_to: with open(save_to, "w", encoding="utf-8") as f: f.write(report) print(f"报告已保存到 {save_to}") return report def _ensure_session(self): if not self._session: raise RuntimeError("请先调用 open() 或使用 with 语句") FILE:data/archetypes.json { "_meta": { "version": "1.0", "description": "可迭代博主类型注册表 — 随分析案例自动演化", "last_updated": "2026-04-23", "total_analyzed": 0 }, "archetypes": { "A": { "name": "荒诞美学型", "desc": "用荒诞幽默包裹哲学内核，高质感视觉，品牌符号统一", "formula": "荒诞场景 × 品牌符号 × 哲思轻量化 → 审美共鸣", "title_signals": ["（劲爆）", "劲爆", "*", "·"], "content_signals": ["哲", "美学", "荒诞", "飞", "高山", "宇宙", "奖励", "永恒"], "commercial": "高端生活方式、旅行、摄影、轻奢品牌", "difficulty": "高", "examples": [], "confirmed_by": 0 }, "B": { "name": "共鸣命名型", "desc": "把私人经历转化为普世命题，给模糊情绪命名", "formula": "私人经历 → 命题化 → 给模糊状态命名 → 普世共鸣", "title_signals": ["*", "为什么", "是什么", "原来", "才知道", "命题", "灯塔", "解药"], "content_signals": ["成长", "情绪", "留学", "记录", "年末", "不确定", "命题", "感受", "理解", "接纳"], "commercial": "成长/教育/创意平台、中端生活方式", "difficulty": "中", "examples": ["xixiCharon"], "confirmed_by": 1 }, "C": { "name": "现实策略型", "desc": "打破职场/人际/金钱潜规则，提供可执行的向上策略", "formula": "困境 → 说破规则 → 提供策略 → 爽感执行", "title_signals": ["骗子", "不要脸", "装", "穷", "规则", "策略", "大法"], "content_signals": ["策略", "规则", "执行", "调整", "快速", "现实", "向上", "方法"], "commercial": "大众消费、职场、电商、实用工具", "difficulty": "中", "examples": [], "confirmed_by": 0 } }, "custom_archetypes": {} } FILE:data/bloggers.json { "_meta": { "description": "已分析博主数据库 — 每次 /xhsfx 分析后自动写入", "total": 1 }, "bloggers": [ { "creator_name": "xixiCharon", "user_id": "5f94fa23000000000100bced", "analyzed_at": "2026-04-23", "archetype": "B+A", "archetype_name": "共鸣命名型 × 荒诞美学型", "stats": { "avg_likes": 3148, "max_likes": 60000, "sample_size": 50, "viral_count": 3, "video_ratio": "46/50" }, "best_topic": "自然/户外", "best_topic_avg": 9069, "tags": [ "留学", "巴黎", "自然", "情绪成长", "INFJ" ], "commercial": "成长/教育/创意平台、中端生活方式", "formula": "私人经历(留学/山野) → 命题化 → 给模糊情绪命名 → 共鸣" } ] } FILE:examples/note_detail.py """ 示例：抓取单篇笔记详情（正文 + 评论） """ import sys, os sys.path.insert(0, os.path.join(os.path.dirname(__file__), "../..")) from xhscosmoskill import XhsClient NOTE_URL = "https://www.xiaohongshu.com/explore/6912a3c8000000000f03e0a5" with XhsClient() as xhs: note = xhs.get_note_detail(NOTE_URL) if note: print(f"标题: {note.title}") print(f"作者: {note.author}") print(f"赞/藏/评: {note.likes}/{note.collects}/{note.comments_count}") print(f"正文:\n{note.content}") print(f"\n评论 ({len(note.comments)}条):") for c in note.comments[:5]: print(f" @{c.user}: {c.text}") else: print("未能抓取到内容") FILE:examples/search_notes.py """ 示例：搜索关键词，抓取笔记列表 + 详情 """ import sys, os sys.path.insert(0, os.path.join(os.path.dirname(__file__), "../..")) from xhscosmoskill import XhsClient KEYWORD = "咖啡" LIMIT = 20 with XhsClient() as xhs: print(f"搜索: {KEYWORD}") notes = xhs.search(KEYWORD, limit=LIMIT) print(f"找到 {len(notes)} 篇笔记") for i, n in enumerate(notes[:5], 1): print(f"[{i}] {n.title} | 赞:{n.likes} | {n.type}") # 补充详情 print("\n抓取详情...") notes = xhs.batch_get_details(notes, delay=1.5) xhs.save(notes, "search_results.json") FILE:examples/user_profile.py """ 示例：抓取指定用户所有笔记（含详情） """ import sys, os sys.path.insert(0, os.path.join(os.path.dirname(__file__), "../..")) from xhscosmoskill import XhsClient # 井越的用户 ID USER_ID = "5b07eb49e8ac2b5fe1e7de09" with XhsClient(scroll_times=8) as xhs: print(f"抓取用户主页: {USER_ID}") notes = xhs.get_user_notes(USER_ID, limit=100) print(f"列表: {len(notes)} 篇") print("\n抓取每篇详情...") notes = xhs.batch_get_details(notes, delay=2.0) xhs.save(notes, f"user_{USER_ID}.json") print(f"\n完成，正文有内容: {len([n for n in notes if n.content])} 篇") FILE:formula.py """ formula.py — 爆款选题公式生成器基于 hsword 实战案例库提炼的6大模型支持 B型（共鸣命名）/ A型（荒诞美学）/ C型（现实策略）/ 混合型用法： from xhscosmoskill.formula import generate_formula_report md = generate_formula_report(notes, creator_name="xixiCharon", archetype=archetype) """ from __future__ import annotations from typing import List, Any, Dict import re # ── B型（共鸣命名型）6大模型 ────────────────────────────────── B_MODELS = [ { "id": "B1", "name": "场景 × 哲思型", "tag": "均赞最高，优先做", "formula": "[具体地点/场景] + [在这里感受到的哲学状态] + [品牌符号]", "mechanism": "地点宏大 × 感受日常 × 语言诗意 = 三重张力，用户既被画面吸引，又被感受击中", "templates": [ "在[地点]的那几天，[发现了什么]*", "[地点]让我第一次理解，什么叫[哲学状态]*", "人在[场景]里是[动态状态]的，[哲学命题]*", "去完[地点]之后，我开始相信[新的观念]*", "[地点]很[表面感受]，但其实很[真实感受]*", ], "themes": ["山野/户外/自然", "独旅领悟", "城市角落的哲思", "跨文化感受"], }, { "id": "B2", "name": "状态命名型", "tag": "高收藏率，深度粉丝", "formula": "[擅长X的人]，对[Y]的感知是[Z]的 + [品牌符号]", "mechanism": "把混沌状态翻译成可被理解的概念，用户获得「被命名」的满足感", "templates": [ "原来[做X的人]，对[Y]的感知都是[Z的]*", "我终于知道为什么[某类人]总会[某种感受]*", "擅长[X]的人，是怎么感受[时间/当下/变化]的*", "有一类人，[描述]，她们把[X]当成[Y]*", "为什么你总是[状态]？因为你其实是[新定义的人]*", ], "themes": ["记录/创作/表达", "INFJ/i人/高敏感特质", "时间感/当下感", "女性自我认知"], }, { "id": "B3", "name": "阶段宣言型", "tag": "节点内容，年终必出", "formula": "[时间节点/年龄阶段] + [这个阶段的成长判断] + [品牌符号]", "mechanism": "时间节点天然触发情绪浓度，个人叙事 × 普世经历 = 同龄共鸣", "templates": [ "[年份]年终[身份]汇报：[这一年的判断]*", "第[N]个[季节/夏末/生日]，[感受]*", "[岁数]岁的[不确定/清醒/成长]，是[如何]的存在*", "在[X]结束的前一天，我终于理解了[Y]*", "[时间段]里，我做了[事]，发现[领悟]*", ], "themes": ["年末/年初/生日", "留学生活阶段总结", "特定年龄节点", "gap年/毕业/转折"], }, { "id": "B4", "name": "独行宣言型", "tag": "独立女性共鸣", "formula": "[我独自/一个人] + [行动] + [反预期结果]", "mechanism": "独立女性认同 × 行动力展示 × 反预期结果 = 三重共鸣", "templates": [ "我独自[行动]，[结果超出预期的感受]*", "一个人[场景]，其实[反预期的发现]*", "每一个[做X的女孩/人]，都是我*", "[场景下的]一个人，是[新的理解]*", "有些事，只有一个人做了，才能[领悟]*", ], "themes": ["solo旅行/独处", "一个人的山野/城市", "独居生活", "户外/徒步"], }, { "id": "B5", "name": "情绪逆转型", "tag": "治愈系高收藏", "formula": "[消极状态] + 其实是 + [正向领悟] + [品牌符号]", "mechanism": "情绪低谷共鸣 + 逆转出口 = 「被治愈的可能性」，用户收藏的是出口本身", "templates": [ "[消极状态]，其实是[正向新解]*", "[看似坏事]的那段时间，反而让我[领悟]*", "[不确定/失去/放弃]，是像[比喻]一样的存在*", "当[困境]来的时候，我发现[反预期的事]*", "长出[新特质]，是从[某个经历]开始的*", ], "themes": ["情绪低谷/调整", "不确定阶段", "成长的代价", "留学/独居压力"], }, { "id": "B6", "name": "世界观输出型", "tag": "最强粘性，深度内容", "formula": "[持续做X的人/坚持Y的意义] + [是如何理解Z的] + [品牌符号]", "mechanism": "用户收藏的是「世界观」，会持续追更，是最强粘性内容", "templates": [ "[持续做X]的人，是在[更大的事情]上做投资*", "出发/记录/独处的意义，不是[表面]，而是[深层]*", "[这件事]，其实是[更大命题]的另一种表达*", "生活的本质，是[新定义]*", "我们[做X]，是为了[更深的东西]*", ], "themes": ["记录的意义", "出发/旅行的意义", "独立/自我的定义", "女性成长哲学"], }, ] # ── A型（荒诞美学型）公式 ───────────────────────────────────── A_MODELS = [ { "id": "A1", "name": "地点 + 反差动作/感受", "formula": "[地点的宏大] × [动作的日常] = 反差萌", "templates": ["在[高原/远方]做[日常小事]！（品牌符号）", "去[地方]的[超日常行为]！"], }, { "id": "A2", "name": "场景 + 情感类比", "formula": "[地点] × [文学化情感命名] = 高级感", "templates": ["[地点]有[文学类比]般的[时间/氛围]（品牌符号）", "这跟走进[情感类比]有什么区别！"], }, { "id": "A3", "name": "荒诞日常", "formula": "[荒唐问题/行为] × [认真叙述] = 喜剧感", "templates": ["大家都是怎么忍住不[荒诞行为]的（品牌符号）", "我在[地方]必须[荒诞行为]"], }, { "id": "A4", "name": "轻哲思命题", "formula": "[严肃命题] × [轻盈表达] = 有深度但不沉重", "templates": ["关于[严肃话题]的奇怪故事（品牌符号）", "什么是[抽象命题]（品牌符号）"], }, ] # ── C型（现实策略型）公式 ───────────────────────────────────── C_MODELS = [ { "id": "C1", "name": "困境说破型", "formula": "困境 → 说破 → 规则 → 策略 → 爽感", "templates": ["为什么[Y]都喜欢[X]？", "普通女孩怎么在[X]里拿回主动权？"], }, { "id": "C2", "name": "信息差揭秘型", "formula": "[表面现象] + [背后规则] + [可执行策略]", "templates": ["对普通人来说，[X]比努力更重要", "没有[资源]的人，怎么[目标]？"], }, ] def _parse_likes(v) -> int: if not v: return 0 s = str(v).replace(",", "").strip() if "万" in s: return int(float(s.replace("万", "")) * 10000) try: return int(s) except ValueError: return 0 def _get_top_notes(notes: List[Any], n: int = 5): def get_title(n): return n.title if hasattr(n, "title") else n.get("title", "") def get_likes(n): return _parse_likes(n.likes if hasattr(n, "likes") else n.get("likes")) return sorted(notes, key=lambda n: -get_likes(n))[:n] def _build_30_topics(creator_name: str, archetype_type: str) -> str: """生成30个可用选题方向""" typ = archetype_type[0] if archetype_type else "B" if typ == "B": return f"""### A 组：自然/户外类（均赞最高，优先做） 1. 在[山/海/草原]的那几天，[感受到了什么]* 2. [某种自然景象]，让我第一次理解什么叫[情绪状态]* 3. 去[目的地]之前我以为[X]，回来之后我只记得[Y]* 4. 一个人徒步，[具体发现]* 5. 冬天的[地点]，有[文学化情感命名]的质地* 6. 到[自然场景]里走走，才发现[领悟]* ### B 组：留学/海外类（基础盘，稳定输出） 7. 在巴黎的第[N]年，我终于理解了[某事]* 8. 留学以后，[某种改变]* 9. 一个在海外的中国女生，怎么理解[某个普世命题]* 10. 巴黎教会我的，不是[表面]，而是[真正的东西]* 11. [年份]年的留学，[年度总结词]* 12. 在异乡过[节日/特殊时刻]，[感受]* ### C 组：记录/创作类（高收藏，深度粉丝） 13. 为什么要频繁记录？[自己的答案]* 14. 擅长[记录/写作/拍照]的人，[是什么样的]* 15. [某件事]让我第一次真的开始记录* 16. 记录的意义，不是留住过去，而是[新定义]* 17. vlog 教会我的最重要一件事：[命题]* 18. 如果你也是创作者，关于[某个困惑]的答案* ### D 组：成长/情绪类（共鸣最广，拓展受众） 19. 二十[几]岁的[某种状态]，是像[比喻]一样的存在* 20. 那段[困难]时期，让我长出了[新特质]* 21. [某类人]是如何对待自己的消极情绪的* 22. 当你开始接受[某件事]，会发现[领悟]* 23. 我不再[某种行为]以后，[变化]* 24. 给所有正在[某个阶段]的人：[一句话]* ### E 组：哲学/世界观类（最高传播潜力） 25. 生活的本质，是[你的定义]* 26. 持续[做某事]，是在[更大意义上]做什么* 27. 为什么[出发/记录/独处]？因为[真正的答案]* 28. [某种看起来无用的事]，其实是[意外的价值]* 29. 我们出发，不是为了[表面]，而是为了[深层]* 30. 一个人能给自己最好的礼物，是[你的答案]*""" elif typ == "A": return """（A型：荒诞美学型30个选题方向）基础模式：地点反差 × 荒诞日常 × 轻哲思 × 品牌符号统一 1-10：地点 + 反差动作系列 11-20：场景 + 情感类比系列 21-30：荒诞命题 + 认真叙述系列具体选题请根据博主真实经历填入[地点]和[动作]。""" else: # C型 return """（C型：现实策略型30个选题方向）基础模式：困境说破 × 信息差揭秘 × 可执行策略 1-10：职场潜规则系列 11-20：消费/财务博弈系列 21-30：普通人向上移动系列""" def generate_formula_report( notes: List[Any], creator_name: str, archetype: Dict, ) -> str: """ 生成爆款选题公式 Markdown 报告参数： notes — get_user_notes() 返回值 creator_name — 博主昵称 archetype — classify_archetype() 返回值返回：Markdown 字符串 """ typ = archetype.get("type", "B") name = archetype.get("name", "共鸣命名型") desc = archetype.get("desc", "") top5 = _get_top_notes(notes, 5) def get_title(n): return n.title if hasattr(n, "title") else n.get("title", "") def get_likes(n): return _parse_likes(n.likes if hasattr(n, "likes") else n.get("likes")) top5_md = "\n".join( f"- 👍{get_likes(n):,} {get_title(n)[:50]}" for n in top5 ) # 选择对应的模型 models = B_MODELS if typ[0] == "B" else (A_MODELS if typ[0] == "A" else C_MODELS) models_md = "" for m in models: templates = "\n".join(f" - {t}" for t in m.get("templates", [])) themes_str = "、".join(m.get("themes", [])) models_md += f""" ### 模型 {m['id']}：{m['name']} {m.get('tag','')} **公式：** {m['formula']} **为什么有效：** {m.get('mechanism', '')} **可套用句式：** {templates} **适合主题：** {themes_str} --- """ topics_30 = _build_30_topics(creator_name, typ) return f"""# {creator_name} — 爆款选题公式 > 说明：本文件不是复刻 {creator_name} 的表面语气，而是提炼她真正有效的**选题逻辑、标题机制、传播结构和可迁移方法**。 --- ## 一、先说结论：为什么能做出爆款？ **账号类型：** {name}（{typ}） **核心内核：** {desc} 爆款公式： > **私人场景 × 命题化 × 诗意命名 × 品牌符号 × 普世共鸣** **高表现代表作：** {top5_md} --- ## 二、总公式 ``` 一个真实经历 / 一个具体场景 / 一个模糊的情绪状态 ↓ 提炼成普遍的人生命题 ↓ 用诗意语言给这个状态一个更精准的说法 ↓ 品牌符号收尾（如 *） ↓ 用户：就是这个感觉！ ``` --- ## 三、最常见的{len(models)}类爆款选题模型 {models_md} --- ## 四、隐形公式（最值钱的底层打法） **隐形公式1：先给感受，再给命题** 不要用命题开头，先描述具体画面，再让命题从画面里生长出来。 - 错误：「论记录的意义」 - 正确：「那天我在山里，突然理解了为什么要记录」* **隐形公式2：动词比名词强** 「流动」比「自由」有力；「后置」比「延迟」有力。动词给人画面，名词只给人概念。 **隐形公式3：品牌符号的系统性** 品牌符号（如 `*`）不是随便加的，它是人格锚点。长期累积会形成"一看就知道是谁"的识别度。 **隐形公式4：自然场景是放大器** 自然场景天然携带"宏大感"，能让任何情绪命题被放大10倍。 --- ## 五、如何迁移到自己的账号 **最优学习路径：** ``` 你的真实场景（不必模仿博主的具体场景） ↓ 找到那个你"说不清楚"的感受 ↓ 用一句话把它命名（用动词，不用形容词） ↓ 写成：[场景] + [命题] + [你的标识符] ↓ 发布，看哪句话在评论区被用户引用 ``` **标题模板直接套用：** 1. 在[地点]的时候，我第一次理解了什么叫________[你的标识符] 2. [做X]的人，对[Y]的感知是________的[标识符] 3. 我独自[行动]，[反预期的发现][标识符] 4. 原来[做某事]，是一种[新的定义][标识符] 5. 第[N]个[时间节点]，[这个阶段的感受][标识符] 6. 终于知道了，为什么[某类人]总会[某种感受][标识符] --- ## 六、可直接复用的30个选题方向 {topics_30} --- ## 七、哪些能学，哪些不能硬抄 **能学的：** - 命题化思维：私人经历 → 公共命题 - 动词命名法：用动词描述情绪 - 场景 × 哲思的意象对冲结构 - 品牌符号系统（找到你自己的"印章"） **不能硬抄的：** - 语气温度（来自真实人设，非技巧） - 具体生活场景（需要真实经历支撑） - 品牌符号本身（需先建立人格认同） --- ## 八、最终总结 **一句话：** {creator_name} 的爆款公式，核心只有一句话： > **把说不清楚的感受，用诗意的语言说清楚。** --- *基于 xhscosmoskill + hsword框架提炼 · {creator_name} · 2026-04-23* """ FILE:models.py from dataclasses import dataclass, field from typing import List, Optional @dataclass class Comment: user: str = "" text: str = "" likes: str = "" time: str = "" @dataclass class Note: id: str = "" title: str = "" content: str = "" url: str = "" author: str = "" author_id: str = "" likes: str = "" collects: str = "" comments_count: str = "" type: str = "" # "video" | "normal" images: List[str] = field(default_factory=list) tags: List[str] = field(default_factory=list) comments: List[Comment] = field(default_factory=list) raw: dict = field(default_factory=dict) def to_dict(self): return { "id": self.id, "title": self.title, "content": self.content, "url": self.url, "author": self.author, "author_id": self.author_id, "likes": self.likes, "collects": self.collects, "comments_count": self.comments_count, "type": self.type, "images": self.images, "tags": self.tags, "comments": [c.__dict__ for c in self.comments], } @dataclass class UserProfile: user_id: str = "" nickname: str = "" description: str = "" followers: str = "" following: str = "" likes_collected: str = "" notes_count: str = "" avatar: str = "" FILE:package.json { "name": "xhsfenxi-pro", "version": "2.0.0", "description": "小红书博主全链路深度分析：数据采集 + 三型分类 + 内核三段论 + 爆款选题公式", "scripts": {}, "keywords": [ "xiaohongshu", "xhsfenxi", "博主分析", "爆款选题" ], "author": "Cosmofang", "license": "MIT" } FILE:parser.py """ 解析器：从拦截到的 API JSON 或 DOM 中提取结构化数据兼容两种来源：API JSON（优先）和 DOM fallback """ from __future__ import annotations from typing import List, Optional from .models import Note, Comment, UserProfile def parse_note_from_api(raw: dict) -> Note: """从 API JSON 解析笔记""" note_card = raw.get("note_card") or raw basic = note_card.get("basic_info") or {} interact = note_card.get("interact_info") or {} user = note_card.get("user") or {} note = Note() note.id = ( raw.get("id") or note_card.get("note_id") or basic.get("note_id", "") ) note.title = ( note_card.get("display_title") or note_card.get("title") or basic.get("title", "") ) note.content = note_card.get("desc") or basic.get("desc", "") note.author = user.get("nickname", "") note.author_id = user.get("user_id", "") note.likes = str(interact.get("liked_count", "")) note.collects = str(interact.get("collected_count", "")) note.comments_count = str(interact.get("comment_count", "")) note.type = "video" if note_card.get("type") == "video" else "normal" # 图片 image_list = note_card.get("image_list") or [] note.images = [img.get("url", "") for img in image_list if img.get("url")] # 标签 tag_list = note_card.get("tag_list") or [] note.tags = [t.get("name", "") for t in tag_list if t.get("name")] note.raw = raw return note def parse_search_response(api_cache: dict) -> List[Note]: """从拦截的搜索 API 响应中提取笔记列表""" notes = [] for url, data in api_cache.items(): if "search" not in url: continue items = ( data.get("data", {}).get("items") or data.get("items") or [] ) for item in items: note_raw = item.get("note_card") or item if not note_raw: continue n = parse_note_from_api({"id": item.get("id", ""), "note_card": note_raw}) xsec = item.get("xsec_token", "") note_id = item.get("id") or note_raw.get("note_id") or n.id if note_id: n.url = f"https://www.xiaohongshu.com/explore/{note_id}?xsec_token={xsec}&xsec_source=pc_search" notes.append(n) return notes def parse_user_notes_response(api_cache: dict) -> List[Note]: """从拦截的用户主页 API 响应中提取笔记列表""" notes = [] for url, data in api_cache.items(): if "user_posted" not in url and "user/notes" not in url: continue items = ( data.get("data", {}).get("notes") or data.get("notes") or [] ) for item in items: n = parse_note_from_api(item) note_id = n.id or item.get("note_id", "") if note_id: n.url = f"https://www.xiaohongshu.com/explore/{note_id}" notes.append(n) return notes def parse_note_detail_response(api_cache: dict) -> Optional[Note]: """从拦截的笔记详情 API 响应中提取单篇笔记""" for url, data in api_cache.items(): if "feed" not in url and "note/detail" not in url: continue items = data.get("data", {}).get("items") or [] for item in items: note_card = item.get("note_card") or {} if note_card: n = parse_note_from_api({"note_card": note_card}) # 评论 comments_data = data.get("data", {}).get("comments") or [] for c in comments_data: comment = Comment( user=c.get("user_info", {}).get("nickname", ""), text=c.get("content", ""), likes=str(c.get("like_count", "")), time=c.get("create_time", ""), ) n.comments.append(comment) return n return None def parse_comment_page_response(api_cache: dict) -> List[Comment]: """从 v2/comment/page API 解析评论列表""" comments = [] for url, data in api_cache.items(): if "comment" not in url: continue items = ( data.get("data", {}).get("comments") or data.get("comments") or [] ) for c in items: user_info = c.get("user_info") or {} comment = Comment( user=user_info.get("nickname", ""), text=c.get("content", ""), likes=str(c.get("like_count", "")), time=str(c.get("create_time", "")), ) comments.append(comment) return comments def parse_dom_notes(browser) -> List[Note]: """DOM fallback：从页面元素中提取笔记（API 拦截失败时使用）""" notes = [] selectors = ["section.note-item", ".note-item", "[class*='note-item']"] items = [] for sel in selectors: items = browser.find_elements(sel) if items: break for item in items: note = Note() for sel in ["a.title span", ".title span", "span.title"]: try: note.title = item.find_element("css selector", sel).text.strip() break except Exception: pass try: note.url = item.find_element("css selector", "a.cover").get_attribute("href") except Exception: pass try: note.likes = item.find_element("css selector", "span.count").text.strip() except Exception: pass try: item.find_element("css selector", ".play-icon") note.type = "video" except Exception: note.type = "normal" if note.title or note.url: notes.append(note) return notes FILE:references/hsword-frameworks.md # hsword 框架参考手册 > 来源：openclaw_cosmo/afa/hsword/ 实战案例库 > 版本：2026-04-23 --- ## 一、内核三段论（每次分析必做） ### 框架结构 ``` 外壳是什么？ → 表面看起来像什么博主（用户最先看到的标签）真正的内核是什么？ → 一句话精炼，带""引号 → 格式：这是一个通过"X、Y、Z"来建立影响力的[定语]博主三层人设结构：表层标签 → 地点/身份/平台标签（可模仿）中层特质 → 性格/能力/气质（可包装）深层价值观 → 鼓励什么/认可什么/传递什么（必须靠长期一致才能被相信） ``` ### 已验证案例 | 博主 | 外壳 | 真正内核 | 深层价值观 | |------|------|---------|-----------| | 井越（A型）| 旅行vlogger / 文艺青年 | "荒诞滤镜下的生活哲学家" | 用奇怪但有质感的视角，记录在世界上存在过的痕迹 | | 橘一橙（B型）| 旅行/成长博主 | "把私人经历转为普世命题的命名者" | 让年轻人获得"原来不止我这样"的理解 | | 陈浪浪（C型）| 大厂博主/成长博主 | "普通女孩现实破局教练型博主" | 普通女性如何在不体面的现实里最大化利用手上的牌 | | xixiCharon（B+A）| 留学vlogger | "把说不清楚的感受用诗意语言说清楚的命名者" | 鼓励女性关注自己内心，认可流动与不确定性 | --- ## 二、三型博主爆款逻辑对比 | 维度 | A型（荒诞美学）| B型（共鸣命名）| C型（现实策略）| |------|------------|------------|------------| | 爆款驱动 | 审美震撼 + 品牌符号 | 被命名的满足感 | 说破现实的爽感 | | 标题风格 | 反差张力（宏大×日常）| 命题感+隐喻+判断 | 冲突词+反常识 | | 用户获得感 | 愉悦+审美+新视角 | 被理解+被命名 | 被激活+被赋能 | | 品牌符号 | "（劲爆）"统一后缀 | `*` 情感印章 | 自我命名先夺话语权 | | 商业适配 | 高端/旅行/轻奢 | 成长/教育/中端生活 | 大众消费/职场/电商 | | 复制难度 | 高（视觉审美积累）| 中（方法可学）| 中（框架可学）| --- ## 三、爆款选题公式体系（6大模型） ### B型（共鸣命名型）——xixiCharon 实证 #### 模型1：场景 × 哲思型 ``` 公式：[地点] + [哲学感悟] + [品牌符号] 机制：地点宏大 × 感受日常 × 语言诗意 = 三重张力实例：人在山里是流动的，不存在于过去也不在未来*（6万赞） ``` **可套用句式：** - 在[地点]的那几天，[发现了什么]* - [地点]让我第一次理解，什么叫[哲学状态]* - 人在[场景]里是[动态状态]的，[哲学命题]* #### 模型2：状态命名型 ``` 公式：[擅长X的人]，对[Y]的感知是[Z]的 + [品牌符号] 机制：把混沌状态翻译成可被理解的概念实例：擅长记录的人对当下的感知是后置的*（10万+） ``` **可套用句式：** - 原来[做X的人]，对[Y]的感知都是[Z的]* - 擅长[X]的人，是怎么感受[时间/当下/变化]的* - 有一类人，[描述]，她们把[X]当成[Y]* #### 模型3：阶段宣言型 ``` 公式：[时间节点] + [成长判断] + [品牌符号] 机制：时间节点触发情绪 × 个人叙事 × 同龄共鸣实例：2025年终留学汇报：今年似乎迈了一大步*（1.9万） ``` #### 模型4：独行宣言型 ``` 公式：[我独自/一个人] + [行动] + [反预期结果] 机制：独立女性认同 × 行动力 × 反预期实例：我独自出发，先斩后奏，斩了也不奏*（8813赞） ``` #### 模型5：情绪逆转型 ``` 公式：[消极状态] + 其实是 + [正向领悟] + [品牌符号] 机制：情绪低谷共鸣 + 逆转出口实例：二十岁的不确定是像灯塔一样的存在*（2999赞） ``` #### 模型6：世界观输出型 ``` 公式：[持续做X] + 是在[更大意义上]做什么 + [品牌符号] 机制：用户收藏"世界观"，最强粘性内容实例：记录和表达如此重要，于是有了写作和vlog*（3.2万） ``` ### A型（荒诞美学型）——井越实证 #### 公式1：地点 + 反差动作/感受 ``` 机制：地点的宏大 × 动作的日常 = 反差萌实例：在帕米尔高原摘豌豆！（劲爆）（5151赞） ``` #### 公式2：场景 + 情感类比 ``` 机制：地点 × 文学化情感命名 = 高级感实例：北海道有爵士乐般的冬日（劲爆）（912赞） ``` #### 公式3：荒诞日常 ``` 机制：荒唐问题/行为 × 认真叙述 = 喜剧感实例：大家都是怎么忍住不亲马鼻孔的（劲爆） ``` ### C型（现实策略型）——陈浪浪实证 #### 核心公式 ``` 困境 → 说破 → 规则 → 策略 → 爽感机制：先夺走羞耻感 → 说破现实 → 给出可执行策略 ``` --- ## 四、综合版"双系统选题公式" ``` 系统A（共鸣命名型）：经历 → 命题 → 命名 → 判断 → 共鸣系统B（现实策略型）：困境 → 说破 → 规则 → 策略 → 爽感混合公式（最强）：真实困境 + 可命名的阶段/处境 + 说破现实的判断 + 可执行动作建议 = 既被理解，又能行动 ``` --- ## 五、标题公式快查表（hsword综合版） | 公式 | 模板 | 适用类型 | |------|------|---------| | 为什么总像 | 为什么[X]总像[Y]？ | B型共鸣 | | 帮大家试过了 | 帮大家试过了，[X]到底值不值得 | B型验证 | | 为什么都喜欢 | 为什么[Y]都喜欢[X]？ | C型说破规则 | | 普通人/普女怎么 | 普通女孩怎么在[X]里拿回主动权？ | C型策略 | | 当一个人…会发生什么 | 当一个人不再[X]，会发生什么？ | B+C混合 | | 原来不是而是 | 原来[X]不是[Y]，而是[Z] | B型重新定义 | --- ## 六、可借鉴 vs 不能盲抄 ### 所有类型通用 | 可借鉴 | 不能盲抄 | |-------|---------| | 选题的命题感 | 表面的生活样本/场景 | | 标题的气质感 | 语气温度和个人语言习惯 | | 内容的人格统一性 | 品牌符号本身（需先建立人格） | | 把个人经历做成公共内容 | 没有真实经历的虚构叙述 | ### 可迁移内容模板（通用） ``` [你的真实场景] ↓ [找到说不清楚的感受] ↓ [用动词而不是形容词命名] ↓ [写成：场景 + 命题 + 你的标识符] ↓ [发布，看哪句话在评论区被引用] ``` --- ## 七、参考数据来源 - **新榜：** `https://www.newrank.cn/profile/xiaohongshu/{user_id}` - **第三方公开站：** TooBigData、热浪数据 - **小红书搜索页：** 可见搜索结果中的账号卡片数据（A1级） - **公开访谈/播客：** 知乎、澎湃、凤凰等（B1级） --- ## 八、分析边界声明模板 ``` 本报告基于公开网页可见信息，数据级别：A1（主页可见）/ B1（第三方公开）/ C1（综合推断）。未覆盖：全量历史笔记 / 评论区文本 / 收藏率完播率后台数据 / 商单实际转化数据。适合用途：账号定位 / 内容策略学习 / 竞品研究 / 对标选题学习。不适合：财务审计 / 精确数据核算。 ``` FILE:references/templates.md # Templates Reference — xhsfenxi v2.0 --- ## Output Files Always produce as `.docx` directly. Three standard files: | File | When | |------|------| | `账号名-结构化总结报告.docx` | Single creator analysis | | `账号名-爆款选题公式.docx` | Single creator topic formula | | `选题公式学习-综合版.docx` | Multi-creator comparison | --- ## Structured Report Skeleton (v2 — Archetype-Aware) ```markdown # 账号名结构化总结报告 > 说明：本报告基于[数据来源说明]。适合用于[账号定位/内容策略/竞品研究/爆款机制提炼]。 ## 一、执行摘要 [2–4段: 博主是谁，核心特征，最值得记住的一句话] ## 二、账号基础信息 [表格: 昵称/小红书号/粉丝/获赞收藏/笔记数/内容类型/账号阶段] [补充: 全平台历史数据（如有）] ## 三、博主背景（公开资料） [人物背景、职业起点、创作理念（来源于公开访谈）] ## 四、账号类型判断 **类型:** Type A/B/C 或混合型 **判断依据:** [2–3条可见信号] **核心内核:** [一句话总结] ## 五、博主核心内核拆解 [外壳是什么 / 真正的内核是什么 / 人设三层结构 / 最强能力] ## 六、内容结构拆解 [内容支柱分布（表格）/ 内容类型比例 / 统一主题] ## 七、标题机制与爆点分析 [最重要特征 / 品牌符号分析 / 高热内容分析（表格）/ 标题公式归纳] ## 八、与其他博主的横向对比（可选） [对比表格: 多维度 × 多博主] ## 九、商业化判断 [当前阶段 / 为什么品牌会找他 / 适合的合作方向 / 不适合的方向] ## 十、最值得学习的核心能力 [3–5个能力，每个含: 说明 + 可学习点] ## 十一、后续建议 [路径A: 深入单篇分析 / 路径B: 制作爆款选题公式 / 路径C: 跨博主综合版] ## 十二、最终总结 [1–2段核心结论] ## 附录：笔记完整列表（可选） [表格: 排名 / 类型 / 标题 / 点赞] ``` --- ## Viral Topic Formula Skeleton (v2) ```markdown # 账号名爆款选题公式 > 说明：本文件不是复刻[账号名]的表面语气，而是提炼真正有效的选题逻辑、标题机制、传播结构、人设打法与可迁移方法。 ## 一、先说结论：为什么能做出爆款？ [账号类型 + 5个叠在一起的核心能力 + 总公式一句话] ## 二、总爆款公式 [5步骤公式 + 压缩版关键词] ## 三、最常见的 3–6 类爆款选题模型 [每个模型: 公式 / 典型表达方向 / 为什么有效 / 可套用句式 / 适合主题] ## 四、标题公式：可以直接拆用 [5–6个标题公式，每个含: 结构说明 + 套用模板] ## 五、内容结构公式 [2–3个结构公式，每个含: 步骤 / 适合做的主题] ## 六、隐形公式 / 最值钱的底层打法 [3–4个] ## 七、如何迁移到自己的内容里 [4个步骤: 找现实母题 / 改写生活故事 / 用冲突词 / 4问检验] ## 八、可直接复用的 20 个选题方向 [编号列表] ## 九、哪些能学，哪些不能硬抄 [可以学的 / 不能硬抄的] ## 十、最终总结 [一句话核心结论] ``` --- ## Comparison Skeleton (v2 — Archetype Map) ```markdown # 选题公式学习（综合版） ## ——结合[博主A]与[博主B]（以及[博主C]）的内容方法论 > 说明：这份文档不是简单拼接报告，而是把各博主最有效的内容能力抽取出来，重新整理成可学习、可迁移、可直接应用的选题方法。 ## 一、为什么把这几位博主放在一起学？ [各博主关键词 / 各博主强项一句话 / 为什么组合更完整] ## 二、账号类型对照表 [表格: 博主 / 类型 / 核心方向 / 用户获得感 / 内容气质 / 标题风格 / 爆点来源] ## 三、共同底层：真正的选题不是题材，而是命题 [5个共同点 / 核心结论] ## 四、双系统（或多系统）选题公式 [系统A（共鸣命名型）: 适用场景 / 核心公式 / 压缩版 / 标题类型] [系统B（现实策略型）: 同上] [（如有）系统C（荒诞美学型）: 同上] ## 五、混合公式 [混合公式定义 + 例子 + 为什么更强] ## 六、标题公式综合版 [6–8个跨类型标题公式] ## 七、内容结构综合版 [结构A / B / C，各含步骤和用户获得感] ## 八、可直接复用的 30 个综合选题方向 [A组（偏共鸣命名）/ B组（偏现实策略）/ C组（偏荒诞美学）/ D组（混合）] ## 九、结论与行动建议 [最重要的学习提醒 + 最终公式] ``` --- ## Archetype Comparison Table Template Use this table when comparing 2–3+ accounts: ```markdown | 维度 | [博主A] | [博主B] | [博主C] | |---|---|---|---| | 账号类型 | Type A 荒诞美学 | Type B 共鸣命名 | Type C 现实策略 | | 核心方向 | | | | | 用户获得感 | | | | | 内容气质 | | | | | 标题风格 | | | | | 品牌符号 | | | | | 爆点来源 | | | | | 商业化方向 | | | | | 可复制难度 | | | | ``` --- ## Evidence Label Reference Use these labels in your internal working notes (do not need to show all in final report): | Label | Meaning | |-------|---------| | `A1` | Xiaohongshu homepage visible fact | | `A2` | Xiaohongshu screenshot fact | | `B1` | Third-party public clue (interview, news, etc.) | | `C1` | Synthesis / interpretation | --- ## Business Word Checklist When building a business-style Word file: - [ ] Cover page - [ ] TOC with `_TocH1_XXX` bookmarks (internal jump, not external link) - [ ] Page numbers - [ ] Conclusion page - [ ] Restrained typography (no loud decorative color blocks unless requested) - [ ] Tested in Word: TOC entries jump correctly - [ ] No duplicate headings accidentally inserted in body --- ## Brand Symbol Analysis Template For Type A accounts or any account with a unified recurring symbol: ```markdown ### 品牌符号分析 **符号:** [（劲爆）/ 其他] **出现频率:** [几乎所有 / 大部分 / 部分] **品牌效果:** - 识别度: [描述] - 反差机制: [描述] - 先发制人效果: [描述] **可学习点:** [其他账号如何建立自己的品牌符号] ``` FILE:references/workflow-xhsfenxi-v2.md # Workflow Reference — xhsfenxi v2.0 --- ## Source Ladder Use sources in this order whenever possible: ### Level A — Primary public evidence Highest trust for visible claims: - Xiaohongshu search result pages - creator homepage/profile pages - user-provided screenshots from Xiaohongshu - user-provided post links that are actually accessible What you can safely claim from Level A: - visible follower / note counts shown on page - visible account ID / self-introduction - visible recent titles / visible public engagement numbers - posting cadence clues visible on page - obvious content categories - brand symbols and signature phrases (like "（劲爆）") ### Level B — Secondary public clues Use as support, never as backend truth: - podcast interviews - encyclopedia entries - public creator bios elsewhere - analytics websites / public creator databases - public brand collaboration listings What Level B is good for: - filling identity / background context - confirming repeated themes - spotting public commercialization clues - triangulating a bigger narrative ### Level C — Synthesis / inference Allowed only when clearly framed as interpretation. Examples: - "This account behaves more like a Type A 荒诞美学 creator than a Type B 共鸣命名 creator." - "The recurring topic logic appears to be: contrast × literary naming × unified symbol." - "The transferable lesson is not the tone, but the framing system." Never present Level C synthesis as raw observed fact. --- ## Standard Execution Runbook ### Step 1 — Confirm the deliverable Choose one or more: - structured report - viral topic formula - archetype classification - comparison report - customized learning report - Word / business version ### Step 2 — Capture the minimum task brief Minimum useful brief: - creator name - available links or screenshots - target output format - whether a business Word version is required ### Step 3 — Archetype pre-classification Before deep analysis, make a preliminary archetype call using visible signals: | Signal | Likely archetype | |--------|-----------------| | Unified brand symbol/tag on every post | Type A 荒诞美学型 | | High-production video, absurdist or philosophical content | Type A 荒诞美学型 | | Posts name vague emotions or life stages | Type B 共鸣命名型 | | Titles feel like thoughtful questions or redefinitions | Type B 共鸣命名型 | | Titles contain conflict words ("骗子", "不要脸", "装") | Type C 现实策略型 | | Content breaks unspoken workplace/relationship/money rules | Type C 现实策略型 | | Mix of resonance + strategy | Mixed B+C | ### Step 4 — Research in layers 1. Find the account and public homepage/search data 2. Save visible account facts (follower count, note count, visible titles, engagement) 3. Pull representative visible titles — note brand symbols, title patterns 4. Add external public clues only where helpful 5. Record limitations before interpretation ### Step 5 — Extract the account system (Five Layers) Always map these five layers: 1. **Identity** — who the creator is framed as; any self-labeling strategy 2. **Audience contract** — why people follow them; what emotional/strategic need is met 3. **Topic system** — what recurring problems/desires they cover; 3–6 topic models 4. **Expression system** — title formulas, brand symbols, opening patterns 5. **Transferability** — what another account can actually learn; what must not be copied ### Step 6 — Produce the right file #### Structured report Recommended top-level sections: 1. Account snapshot 2. Archetype classification + rationale 3. Positioning judgment 4. Audience and demand 5. Content pillars 6. Title / hook system (including brand symbol analysis) 7. Narrative structure 8. Commercialization clues 9. What to learn 10. What not to copy 11. Conclusion 12. (Optional) Full note list sorted by engagement #### Viral topic formula Recommended top-level sections: 1. Why this creator produces strong topics (archetype-based) 2. Total formula 3. 3–6 recurring topic models 4. Title formulas 5. Body structure formulas 6. Distribution / comment trigger clues 7. How to migrate it to another account 8. 10–30 topic directions #### Comparison report (2–3+ accounts) Recommended top-level sections: 1. Why compare these creators together 2. Archetype map (each creator's type) 3. Shared foundations 4. Key differences 5. Hybrid formula 6. Which path fits which use case 7. Suggested topic directions 8. Final recommendation --- ## Archetype-Specific Analysis Lenses ### Type A — 荒诞美学型 Focus questions: - What is the brand symbol? (unified tag, phrase, or visual marker) - How does the title create contrast? (grand × mundane, serious × absurd) - Where does the humor come from? (earnest treatment of absurd things) - What is the underlying philosophical/emotional core? - Why is replication hard? (visual sensibility, accumulated aesthetic) Key learning for other accounts: - Find your own "brand symbol" — a unifying tag that creates recognition - Train "contrast naming" — don't describe the place, name the feeling - Practice "absurdist framing" — pick topics that are ridiculous when taken seriously --- ### Type B — 共鸣命名型 Focus questions: - What life stages or emotional states does the account name? - How does private experience become universal proposition? - What are the signature conceptual phrases? - What is the "new way to understand yourself" users get? Key learning for other accounts: - Practice "命题化" — reframe personal experience as a universal human situation - Build a personal vocabulary of named states ("Odyssey时期", etc.) - The formula: experience → proposition → naming → judgment → resonance --- ### Type C — 现实策略型 Focus questions: - What real-world rules does the account break open? - How does the self-labeling strategy work? (antifragile identity) - What is the conflict word in each title? - What is the "can-do action" users walk away with? Key learning for other accounts: - Find your "reality母题" — the real困境 you speak to - Practice rewriting weak framings into strong rule-breaking framings - The formula: 困境 → 说破 → 规则 → 策略 → 爽感 --- ## Limitation Language Use direct wording like: - "This round is based primarily on public homepage/search-page evidence." - "Single-post detail pages were not stably accessible due to platform risk control." - "Third-party figures are used as directional clues, not audited backend data." - "Archetype classification is based on visible patterns; the creator may operate differently at the full content level." --- ## Deep-Dive Upgrade Path If the user wants finer analysis, ask for 3–10 representative posts, ideally with: - screenshots of cover + title - screenshots of body pages / subtitles - comments screenshots - transcript or summary if video-based Then upgrade from account-level to post-level analysis. --- ## Word Output Workflow ### Overview Always generate Markdown first, then convert to Word. Iterating Markdown is cheaper than re-generating DOCX. ### Script Selection Guide Use the workflow archive scripts at: ``` openclaw_cosmo/afa/小红书分析与工作流归档/02-Word生成与目录修复脚本/ ``` | Script | Use when | |--------|----------| | `build_docx6.py` | Default — most stable, use this first | | `build_docx.py` → `build_docx5.py` | Earlier iterations, kept for reference/debugging | | `inspect_docx.py` | Inspect internal XML, bookmarks, link structure | | `fix_final2.py` | Final TOC / attribute fix after build | | `fix_attrs.py` | Fix XML namespace attribute writing errors | | `fix_wrong.py` | Targeted corrections for known wrong output | | `check_it.py` | Verify output correctness | ### Word TOC Fix Checklist When a generated .docx has broken TOC links: 1. Run `inspect_docx.py` — check `word/document.xml` for bookmark names and link types 2. Verify heading bookmarks use format `_TocH1_XXX` (not `#_TocH1_XXX`) 3. Verify internal links use Word-native anchor style, not external URL rels 4. If `elem.set()` errors appear, use full namespace: `elem.set('{' + W_NS + '}id', value)` 5. Run `fix_final2.py` after any XML edits 6. Test by opening in Word — click TOC entries to verify jump ### Common Errors | Error | Cause | Fix | |-------|-------|-----| | Bookmark name wrong format | Missing `_TocH1_` prefix | Rename via `fix_attrs.py` | | `anchor` contains `#` | Wrong anchor format | Strip `#` prefix | | TOC uses external link rels | Used `http://` link type for internal jump | Switch to `w:instr` field or anchor | | Duplicate headings in body | TOC headings accidentally inserted into body | Use `fix_wrong.py` to clean | | Namespace attribute error | Missing full namespace URI in `elem.set()` | Use `'{' + W_NS + '}id'` form | ### Best Practice - Keep only `build_docx6.py` as the production script - Keep all earlier versions as debugging reference, not for production - One Markdown → one `build_docx6.py` call → inspect → fix if needed FILE:references/workflow.md # 完整分析工作流手册 v2.0 > 记录：xixiCharon 深度分析全过程（三份文档合并 + hsword框架应用） > 原始 xhsfenxi 工作流见：workflow-xhsfenxi-v2.md --- ## 标准执行管道（Full Pipeline） ``` Step 0 Cookie 健康检查 Step 1 解析输入 → 检查数据库 Step 2 数据采集 Step 3 基础统计 Step 4 三型分类 Step 5 内核三段论（hsword框架） Step 6 爆款选题公式（6模型） Step 7 结构化报告 Markdown Step 8 外部文档合并（如有） Step 9 写入博主数据库 Step 10 生成黑体 Word ``` --- ## Step 0 — Cookie 健康检查 ```python from xhscosmoskill import print_cookie_status, get_best_cookies cookies = get_best_cookies() print_cookie_status(cookies) ``` **Cookie 优先顺序：** 1. `shopify-marketing/xhs_cookies.json`（最新，优先） 2. `xiaohongshu_new/xhs_cookies.json`（备用） **过期判断：** notes 返回 ≤ 1 条 → Cookie 失效 → 运行 `python3 xhs_login.py` --- ## Step 1 — 解析输入 ```python import re user_id = re.search(r'/user/profile/([a-f0-9]+)', url).group(1) from xhscosmoskill import get_blogger existing = get_blogger(creator_name) # 检查是否已分析过 ``` --- ## Step 2 — 数据采集 ```python from xhscosmoskill import XhsClient with XhsClient(cookies_file=cookies, headless=True, scroll_times=10) as xhs: notes = xhs.get_user_notes(user_id, limit=50) xhs.save(notes, f"/tmp/{creator_name}_notes.json") ``` --- ## Step 3-4 — 统计 + 分类 ```python from xhscosmoskill import compute_stats, classify_archetype, build_five_layers stats = compute_stats(notes) archetype = classify_archetype(notes) five = build_five_layers(notes, archetype) ``` --- ## Step 5 — 内核三段论（hsword框架）每次分析必须明确三层结构（参见 hsword-frameworks.md）： ``` 外壳是什么？→ 表面标签（可模仿）真正的内核？→ 一句话，带""引号三层人设：表层标签 → 自动提取中层特质 → 手动填写（性格/能力/气质）深层价值观 → 手动填写（最重要，不可复制） ``` --- ## Step 6 — 爆款选题公式 ```python from xhscosmoskill import generate_formula_report from xhscosmoskill.utils import save_md formula_md = generate_formula_report(notes, creator_name, archetype) save_md(formula_md, f"/tmp/{creator_name}-爆款选题公式.md") ``` 生成内容：总公式 + 6模型（每个含可套用句式）+ 30个选题方向 + 可迁移模板 --- ## Step 7 — 结构化报告 ```python from xhscosmoskill import analyze_account report_md = analyze_account(notes, creator_name=creator_name, mode="full") save_md(report_md, f"/tmp/{creator_name}-结构化总结报告.md") ``` --- ## Step 8 — 外部文档合并（如有）用户提供额外分析文档时，按优先级 patch： | 优先级 | 内容 | |--------|------| | 1 | 更高的点赞数（修正我们的数据）| | 2 | 新笔记数据（我们50篇未收录的）| | 3 | 新分析维度（我们没有的章节）| | 4 | 更精炼的提炼（更好的一句话总结）| **xixiCharon 合并经验：** - 文档1（txt）：补入 3.2万/7490/6231 等数据 + 视觉风格 + 风险分析 - 文档2（md）：补入 8813赞 + 头部集中度38.15% + 可摘抄感 + 新榜链接 - 验证：patch后检查7个关键词确认写入 --- ## Step 9 — 写入数据库 ```python from xhscosmoskill import save_blogger save_blogger( creator_name=creator_name, user_id=user_id, archetype=archetype, stats={"avg_likes": stats["avg_likes"], "max_likes": stats["max_likes"], "sample_size": stats["total"], "viral_count": stats["brackets"]["1万+"], "video_ratio": f"{stats['video_count']}/{stats['total']}"}, best_topic=five["best_topic"], best_topic_avg=five["best_topic_avg"], formula="核心公式一句话", ) ``` --- ## Step 10 — 生成黑体 Word ```python from xhscosmoskill.scripts.build_docx import build_word build_word(f"/tmp/{creator_name}-结构化总结报告.md", f"/tmp/{creator_name}-结构化总结报告.docx", title=creator_name, subtitle="小红书博主深度结构化分析报告") build_word(f"/tmp/{creator_name}-爆款选题公式.md", f"/tmp/{creator_name}-爆款选题公式.docx", title=creator_name, subtitle="爆款选题公式 · 6大模型 · 30个选题方向") ``` --- ## 证据分级 | 级别 | 来源 | 用法 | |------|------|------| | A1 | 小红书公开主页可见数据 | 直接陈述 | | A2 | 用户提供截图 | 直接陈述 | | B1 | 第三方公开（新榜/访谈）| 背景补充 | | C1 | 综合推断 | 明确标注"推断" | --- ## 类型迭代协议 ```python from xhscosmoskill import add_archetype, update_archetype_signals # 添加新类型（当最高分 < 10 时） add_archetype(key="D", name="知识科普型", desc="...", formula="...", title_signals=[], content_signals=[], commercial="...", difficulty="中") # 迭代已有类型信号词 update_archetype_signals("B", new_content_signals=["新词"]) ``` --- ## 已分析博主档案 | 博主 | user_id | 类型 | 最高赞 | 精神母题 | |------|---------|------|--------|---------| | xixiCharon | 5f94fa23000000000100bced | B+A | 10万+ | 一个女生如何在流动生活里确认自己 | --- ## 参考资源 - hsword 实战案例：`openclaw_cosmo/afa/hsword/` - hsword 框架手册：`references/hsword-frameworks.md` - 原始 xhsfenxi 工作流：`references/workflow-xhsfenxi-v2.md` - Word 修复脚本：`openclaw_cosmo/afa/小红书分析与工作流归档/02-Word生成与目录修复脚本/` - 新榜数据：`https://www.newrank.cn/profile/xiaohongshu/{user_id}` FILE:scripts/archetype.js #!/usr/bin/env node /** * xhsfenxi — archetype.js * Purpose: Print the archetype classification prompt for a Xiaohongshu creator. * Helps identify whether an account is Type A (荒诞美学), Type B (共鸣命名), * Type C (现实策略), or a hybrid — based on the three proven archetypes * distilled from real analyses of multiple Xiaohongshu creators. * Usage: * node scripts/archetype.js <creator-name> * node scripts/archetype.js <creator-name> --quick */ const args = process.argv.slice(2); const quick = args.includes('--quick'); const name = args.filter(a => !a.startsWith('--'))[0]; if (!name) { console.error('Usage: node scripts/archetype.js <creator-name> [--quick]'); process.exit(1); } if (quick) { console.log(` === xhsfenxi archetype cheatsheet === TYPE A — 荒诞美学型 (Absurdist Aesthetics) Example: 荒诞美学博主（vlogger型） Signals: - Unified brand symbol/tag on every post (e.g. "（劲爆）") - Absurdist or philosophical topics treated earnestly - High-production video; literary title naming - Serious × absurd contrast in every piece Formula: 荒诞滤镜下的生活哲学 TYPE B — 共鸣命名型 (Resonance & Naming) Example: 成长世界观表达博主 Signals: - Posts name vague emotional states or life stages - Private experience → universal proposition - Titles feel like thoughtful questions or redefinitions Formula: 经历 → 命题 → 命名 → 判断 → 共鸣 TYPE C — 现实策略型 (Reality & Strategy) Example: 普通女孩上行策略博主 Signals: - Titles contain conflict words ("骗子", "不要脸", "装") - Content breaks unspoken workplace/relationship/money rules - Self-labels with perceived weakness to pre-empt criticism Formula: 困境 → 说破 → 规则 → 策略 → 爽感 MIXED B+C — Most powerful hybrid Combines: resonance/naming × reality/strategy = "You understand me" + "Now I know what to do" `); process.exit(0); } console.log(` === xhsfenxi archetype classification: ${name} === TASK Classify the Xiaohongshu account "${name}" into one of these three archetypes (or a hybrid), based on visible public evidence. ───────────────────────────────────────────── THREE ARCHETYPES ───────────────────────────────────────────── TYPE A — 荒诞美学型 (Absurdist Aesthetics) Core: Wraps philosophical/serious content in absurdist humor + high visual quality Proven example: 荒诞美学vlogger型 (unified brand symbol; grand × mundane contrast; literary naming) Key signals to look for: - Does every post share a unified recurring symbol or phrase? - Do titles contrast a grand/serious setting with a mundane/absurd action? - Is the visual/production quality noticeably higher than peers? - Are serious topics treated with humor and humor treated earnestly? TYPE B — 共鸣命名型 (Resonance & Naming) Core: Names vague emotions and life stages so users feel "finally, someone said it" Proven example: 成长世界观表达型 (youth growth, worldview expression, concept naming) Key signals to look for: - Do posts give names or definitions to emotional states most people can't articulate? - Does the account turn personal stories into universal life propositions? - Do titles feel like thoughtful philosophical questions or redefinitions? - Is the tone warm, perceptive, aesthetically refined? TYPE C — 现实策略型 (Reality & Strategy) Core: Breaks unspoken real-world rules; provides executable strategies for ordinary people Proven example: 普通女孩上行策略型 (antifragile identity; rule-breaking; 普通人上行) Key signals to look for: - Do titles contain conflict words or counter-intuitive framing? - Does the account self-label with a perceived weakness to pre-empt criticism? - Does content expose hidden rules in workplace/relationships/money/consumption? - Does every piece end with a clear, actionable conclusion or stance? MIXED HYBRID Some accounts combine two archetypes. Most powerful is B+C: "I understand your situation (Type B) AND here's what you can actually do (Type C)" ───────────────────────────────────────────── CLASSIFICATION PROMPT ───────────────────────────────────────────── Based on publicly visible data for "${name}", answer: 1. Which archetype does this account most closely match? (A / B / C / Mixed) 2. What are 2–3 concrete visible signals that support this classification? 3. If mixed: what percentage is each type, and does one dominate? 4. What is the single sentence that describes this account's core "filter" on the world? 5. Is there a brand symbol or signature phrase? If yes, what is it and how does it function? Then proceed with the full analysis using this archetype as the lens. `); FILE:scripts/build_docx.py """ build_docx.py — 通用黑体 Word 生成器将 Markdown 文件转换为全黑体样式的 Word 文档支持：标题层级 / 表格 / 列表 / 引用 / 代码块 / 分割线用法（命令行）： python3 build_docx.py <md_path> <out_path> [title] [subtitle] [meta] 用法（Python）： from xhscosmoskill.scripts.build_docx import build_word build_word("/tmp/report.md", "/tmp/report.docx", title="xixiCharon", subtitle="爆款选题公式") """ import re import sys from pathlib import Path try: from docx import Document from docx.shared import Pt, RGBColor, Cm from docx.enum.text import WD_ALIGN_PARAGRAPH from docx.enum.table import WD_TABLE_ALIGNMENT from docx.oxml.ns import qn from docx.oxml import OxmlElement except ImportError: print("请先安装：pip install python-docx") sys.exit(1) # ── 配色 ────────────────────────────────────────────────────── C_BRAND = RGBColor(0x1A, 0x1A, 0x1A) C_H1 = RGBColor(0x0D, 0x0D, 0x0D) C_H2 = RGBColor(0x1A, 0x1A, 0x1A) C_ACCENT = RGBColor(0x00, 0x80, 0x60) # Shopify 绿 C_LIGHT = RGBColor(0x6B, 0x72, 0x80) C_TH_BG = RGBColor(0xF3, 0xF4, 0xF6) C_LINE = RGBColor(0x22, 0xC5, 0x5E) HEITI = "黑体" # ── 工具函数 ────────────────────────────────────────────────── def set_font(run, size, bold=False, color=None, italic=False): """全黑体设置（中英文统一）""" run.font.size = Pt(size) run.font.bold = bold run.font.italic = italic if color: run.font.color.rgb = color rPr = run._r.get_or_add_rPr() rf = rPr.find(qn('w:rFonts')) if rf is None: rf = OxmlElement('w:rFonts') rPr.insert(0, rf) for attr in ('w:ascii', 'w:hAnsi', 'w:eastAsia', 'w:cs'): rf.set(qn(attr), HEITI) def clean(s): for p, r in [ (r'\*\*(.+?)\*\*', r'\1'), (r'\*(.+?)\*', r'\1'), (r'\[(.+?)\]$.+?$', r'\1'), (r'`(.+?)`', r'\1'), (r'^#{1,6}\s+', ''), ]: s = re.sub(p, r, s) return s.strip() def set_cell_bg(cell, rgb): tcPr = cell._tc.get_or_add_tcPr() shd = OxmlElement('w:shd') shd.set(qn('w:val'), 'clear') shd.set(qn('w:color'), 'auto') shd.set(qn('w:fill'), f'{rgb[0]:02X}{rgb[1]:02X}{rgb[2]:02X}') tcPr.append(shd) def add_rule(doc, color=C_LINE): p = doc.add_paragraph() p.paragraph_format.space_before = Pt(1) p.paragraph_format.space_after = Pt(6) pPr = p._p.get_or_add_pPr() pBdr = OxmlElement('w:pBdr') bot = OxmlElement('w:bottom') bot.set(qn('w:val'), 'single') bot.set(qn('w:sz'), '4') bot.set(qn('w:space'), '1') bot.set(qn('w:color'), f'{color[0]:02X}{color[1]:02X}{color[2]:02X}') pBdr.append(bot) pPr.append(pBdr) # ── 核心构建函数 ────────────────────────────────────────────── def build_word(md_path: str, out_path: str, title: str = "", subtitle: str = "", meta: str = ""): """ 将 Markdown 转换为黑体 Word 文档参数： md_path — 输入 Markdown 文件路径 out_path — 输出 .docx 文件路径 title — 封面大标题（可选） subtitle — 封面副标题（可选） meta — 封面元数据行（可选） """ md_text = Path(md_path).read_text(encoding="utf-8") doc = Document() for sec in doc.sections: sec.top_margin = Cm(2.5) sec.bottom_margin = Cm(2.5) sec.left_margin = Cm(3.0) sec.right_margin = Cm(3.0) # 默认样式 nml = doc.styles['Normal'] nml.font.name = HEITI nml.font.size = Pt(10.5) nml._element.rPr.rFonts.set(qn('w:eastAsia'), HEITI) # 封面 if title: doc.add_paragraph() p = doc.add_paragraph() p.alignment = WD_ALIGN_PARAGRAPH.CENTER set_font(p.add_run(title), 28, bold=True, color=C_ACCENT) if subtitle: p = doc.add_paragraph() p.alignment = WD_ALIGN_PARAGRAPH.CENTER set_font(p.add_run(subtitle), 14, color=C_LIGHT) if meta: p = doc.add_paragraph() p.alignment = WD_ALIGN_PARAGRAPH.CENTER set_font(p.add_run(meta), 9, color=C_LIGHT) if title: doc.add_page_break() # 解析 Markdown lines = md_text.splitlines() i = 0 while i < len(lines): line = lines[i] # H1 if re.match(r'^# [^#]', line): p = doc.add_paragraph() p.paragraph_format.space_before = Pt(16) p.paragraph_format.space_after = Pt(4) set_font(p.add_run(clean(line)), 20, bold=True, color=C_H1) i += 1; continue # H2 if re.match(r'^## [^#]', line): p = doc.add_paragraph() p.paragraph_format.space_before = Pt(18) p.paragraph_format.space_after = Pt(2) set_font(p.add_run(clean(line)), 13, bold=True, color=C_H2) add_rule(doc) i += 1; continue # H3 if re.match(r'^### ', line): p = doc.add_paragraph() p.paragraph_format.space_before = Pt(10) p.paragraph_format.space_after = Pt(3) set_font(p.add_run(clean(line)), 11, bold=True, color=C_ACCENT) i += 1; continue # 引用 if line.startswith('> '): p = doc.add_paragraph() p.paragraph_format.left_indent = Cm(1) p.paragraph_format.space_before = Pt(3) p.paragraph_format.space_after = Pt(3) set_font(p.add_run(line[2:].strip()), 10.5, italic=True, color=C_LIGHT) i += 1; continue # 代码块 if line.startswith('```'): i += 1 code = [] while i < len(lines) and not lines[i].startswith('```'): code.append(lines[i]) i += 1 p = doc.add_paragraph('\n'.join(code)) p.paragraph_format.left_indent = Cm(0.8) p.paragraph_format.space_before = Pt(3) p.paragraph_format.space_after = Pt(3) for run in p.runs: set_font(run, 9, color=RGBColor(0x37, 0x37, 0x37)) i += 1; continue # 水平线 if line.strip() == '---': add_rule(doc, C_TH_BG) i += 1; continue # 表格 if line.strip().startswith('|'): tbl_lines = [] while i < len(lines) and lines[i].strip().startswith('|'): if not re.match(r'^\|[\s|:-]+$', lines[i].strip()): tbl_lines.append(lines[i]) i += 1 if not tbl_lines: continue rows = [ [c.strip() for c in tl.split('|')[1:-1]] for tl in tbl_lines ] rows = [r for r in rows if any(c for c in r)] if not rows: continue ncols = max(len(r) for r in rows) tbl = doc.add_table(rows=len(rows), cols=ncols) tbl.style = 'Table Grid' tbl.alignment = WD_TABLE_ALIGNMENT.LEFT for ri, row in enumerate(rows): for ci in range(ncols): cell = tbl.cell(ri, ci) txt = clean(row[ci]) if ci < len(row) else '' p = cell.paragraphs[0] p.clear() run = p.add_run(txt) set_font(run, 9.5, bold=(ri == 0), color=C_H2 if ri == 0 else C_BRAND) p.paragraph_format.space_before = Pt(2) p.paragraph_format.space_after = Pt(2) if ri == 0: set_cell_bg(cell, C_TH_BG) doc.add_paragraph() continue # 列表 if re.match(r'^[-*] |^\d+\. ', line): txt = re.sub(r'^[-*] |^\d+\. ', '', line) p = doc.add_paragraph(style='List Bullet') p.paragraph_format.left_indent = Cm(0.8) p.paragraph_format.space_before = Pt(1) p.paragraph_format.space_after = Pt(1) set_font(p.add_run(clean(txt)), 10.5, color=C_BRAND) i += 1; continue # 空行 if not line.strip(): i += 1; continue # 普通段落 p = doc.add_paragraph() p.paragraph_format.space_before = Pt(2) p.paragraph_format.space_after = Pt(4) set_font(p.add_run(clean(line)), 10.5, color=C_BRAND) i += 1 doc.save(out_path) print(f"✅ Word 已生成：{out_path}") return out_path # ── 命令行入口 ──────────────────────────────────────────────── if __name__ == "__main__": if len(sys.argv) < 3: print("用法：python3 build_docx.py <md_path> <out_path> [title] [subtitle] [meta]") sys.exit(1) md = sys.argv[1] out = sys.argv[2] ttl = sys.argv[3] if len(sys.argv) > 3 else "" sub = sys.argv[4] if len(sys.argv) > 4 else "" meta = sys.argv[5] if len(sys.argv) > 5 else "" build_word(md, out, ttl, sub, meta) FILE:scripts/docx-plan.js #!/usr/bin/env node /** * xhsfenxi — docx-plan.js * Purpose: Print the Word generation plan for a completed Xiaohongshu analysis, * including which build_docx*.py script to use, the full workflow steps, * and TOC fix guidance from the workflow archive. * Usage: * node scripts/docx-plan.js <creator-name> * node scripts/docx-plan.js <creator-name> --mode comparison * node scripts/docx-plan.js --toc-fix */ const ARCHIVE_BASE = '~/Desktop/cosmocloud/Deeplumen/cosmowork/openclaw_cosmo/afa/小红书分析与工作流归档/02-Word生成与目录修复脚本'; const args = process.argv.slice(2); const tocFix = args.includes('--toc-fix'); const modeIdx = args.indexOf('--mode'); const mode = modeIdx !== -1 ? args[modeIdx + 1] : 'single'; const name = args.filter((a, i) => { if (a.startsWith('--')) return false; if (modeIdx !== -1 && i === modeIdx + 1) return false; return true; })[0]; if (tocFix) { console.log(` === xhsfenxi Word TOC Fix Guide === Archive scripts location: ${ARCHIVE_BASE}/ STEP-BY-STEP FIX PROCESS 1. Run inspect_docx.py to diagnose the issue: python3 ${ARCHIVE_BASE}/inspect_docx.py <your-file.docx> → Check: bookmark names, link types, heading structure 2. Verify the following are correct: - Bookmark names follow format: _TocH1_XXX (NOT #_TocH1_XXX) - Internal TOC links use Word-native anchor, NOT external http:// rels - No duplicate headings accidentally inserted in document body - Namespace attributes written as: elem.set('{' + W_NS + '}id', value) 3. Run the appropriate fix script: - For attribute/namespace errors: python3 fix_attrs.py <file.docx> - For final TOC + heading cleanup: python3 fix_final2.py <file.docx> - For known wrong output patterns: python3 fix_wrong.py <file.docx> 4. Verify with check_it.py: python3 ${ARCHIVE_BASE}/check_it.py <your-fixed-file.docx> 5. Open in Word and test: click each TOC entry to confirm internal jump works. COMMON ERROR TABLE Bookmark format wrong → rename using fix_attrs.py anchor contains # → strip # prefix in fix_final2.py TOC uses external rels → switch to w:instr field or Word anchor Duplicate headings → fix_wrong.py XML namespace error → use full namespace URI in elem.set() `); process.exit(0); } if (!name && !tocFix) { console.error('Usage: node scripts/docx-plan.js <creator-name> [--mode comparison]'); console.error(' node scripts/docx-plan.js --toc-fix'); process.exit(1); } const isComparison = mode === 'comparison'; const mdFile = isComparison ? `选题公式学习-综合版.md` : `${name}-结构化总结报告.md`; const docxFile = isComparison ? `选题公式学习-综合版.docx` : `${name}-结构化总结报告-商业版.docx`; console.log(` === xhsfenxi Word generation plan: ${name || 'comparison'} === Mode: ${isComparison ? 'comparison / 综合版' : 'single creator'} ───────────────────────────────────────────── STEP 1 — Confirm Markdown is complete ───────────────────────────────────────────── Make sure this file is finalized: ${mdFile} ───────────────────────────────────────────── STEP 2 — Run build_docx6.py (recommended) ───────────────────────────────────────────── Primary script (most stable): python3 ${ARCHIVE_BASE}/build_docx6.py ${mdFile} ${docxFile} If build_docx6.py fails, fall back to earlier versions: build_docx5.py → build_docx4.py → build_docx3.py ───────────────────────────────────────────── STEP 3 — Inspect the output ───────────────────────────────────────────── python3 ${ARCHIVE_BASE}/inspect_docx.py ${docxFile} Check for: ✓ Headings have _TocH1_XXX bookmarks ✓ TOC links are internal (not external URL rels) ✓ No duplicate heading text in body ✓ Page count and section structure look correct ───────────────────────────────────────────── STEP 4 — Fix if needed ───────────────────────────────────────────── Run: node scripts/docx-plan.js --toc-fix for the full TOC fix checklist. Most common fix: python3 ${ARCHIVE_BASE}/fix_final2.py ${docxFile} ───────────────────────────────────────────── STEP 5 — Final check ───────────────────────────────────────────── python3 ${ARCHIVE_BASE}/check_it.py ${docxFile} → Open in Word, click TOC entries, verify internal jumps work ───────────────────────────────────────────── EXPECTED OUTPUT FILES ───────────────────────────────────────────── ${docxFile} ${isComparison ? '' : ` ${name}-结构化总结报告.md (keep as source of truth)\n ${name}-爆款选题公式.md (if also produced)`} ───────────────────────────────────────────── ARCHIVE LOCATION ───────────────────────────────────────────── Scripts: ${ARCHIVE_BASE}/ Stable production script: build_docx6.py Inspection tool: inspect_docx.py `); FILE:scripts/intake.js #!/usr/bin/env node /** * xhsfenxi — intake.js * Purpose: Print a structured intake template for a Xiaohongshu creator-analysis task. * Usage: * node scripts/intake.js * node scripts/intake.js --json */ const args = process.argv.slice(2); const asJson = args.includes('--json'); const template = { task: 'xiaohongshu-creator-analysis', creatorNames: ['<creator-name>'], links: ['<homepage-or-search-url>'], screenshots: ['<optional-local-path-or-url>'], deliverables: ['structured-report'], outputDir: '<optional-output-dir>', needWord: false, needBusinessVersion: false, needComparison: false, notes: 'What exactly should be learned from this creator?' }; if (asJson) { console.log(JSON.stringify(template, null, 2)); process.exit(0); } console.log(` === xhsfenxi intake template === Provide or confirm the following before deep research: - creator name(s) - homepage/search URL(s) - screenshots or links for representative posts (optional but helpful) - requested deliverable(s): structured-report / topic-formula / comparison / business-word - output directory - any specific question to answer JSON template: ${JSON.stringify(template, null, 2)} `); FILE:scripts/report-plan.js #!/usr/bin/env node /** * xhsfenxi — report-plan.js * Purpose: Print recommended deliverables and filenames for a Xiaohongshu analysis request. * Usage: * node scripts/report-plan.js <creator-name> * node scripts/report-plan.js <creator-a> <creator-b> --mode compare */ const args = process.argv.slice(2); const modeIdx = args.indexOf('--mode'); const mode = modeIdx !== -1 ? args[modeIdx + 1] : 'single'; const names = args.filter((a, i) => { if (a === '--mode') return false; if (modeIdx !== -1 && i === modeIdx + 1) return false; return true; }); if (!names.length) { console.error('Usage: node scripts/report-plan.js <creator-name>'); console.error(' node scripts/report-plan.js <creator-a> <creator-b> --mode compare'); process.exit(1); } if (mode === 'compare') { if (names.length < 2) { console.error('Compare mode requires two creator names.'); process.exit(1); } const [a, b] = names; const files = [ `选题公式学习-综合版.docx` ]; console.log(` === xhsfenxi report plan === Mode: comparison Creators: ${a} / ${b} Recommended deliverables: - comparison report - hybrid topic-formula study - optional business Word version Recommended filenames: ${files.map(f => `- ${f}`).join('\n')} `); process.exit(0); } const name = names[0]; const files = [ `${name}-结构化总结报告.docx`, `${name}-爆款选题公式.docx` ]; console.log(` === xhsfenxi report plan === Mode: single creator Creator: ${name} Recommended deliverables: - structured report - viral topic formula - optional business Word version Recommended filenames: ${files.map(f => `- ${f}`).join('\n')} `); FILE:utils.py """ utils.py — 工具函数集包含：Cookie 健康检查 / 路径常量 / 输出工具 """ import json import os import time from typing import Dict, Any _BASE = os.path.dirname(__file__) # ── 路径常量 ────────────────────────────────────────────────── COOKIES_PRIMARY = os.path.join( _BASE, "..", "..", "..", "shopify-marketing", "xhs_cookies.json" ) COOKIES_ALT = os.path.join(_BASE, "..", "xhs_cookies.json") DATA_DIR = os.path.join(_BASE, "data") ARCHETYPES_DB = os.path.join(DATA_DIR, "archetypes.json") BLOGGERS_DB = os.path.join(DATA_DIR, "bloggers.json") HSWORD_DIR = os.path.join( _BASE, "..", "..", "..", "..", "openclaw_cosmo", "afa", "hsword" ) # ── Cookie 健康检查 ─────────────────────────────────────────── def check_cookies(cookies_file: str = None) -> Dict[str, Any]: """ 检查 Cookie 文件的有效性返回： { "valid": bool, "file": str, "total": int, "expired": list[str], "expiring_soon": list[str], # 24小时内过期 "message": str } """ path = cookies_file or _find_cookies() if not path or not os.path.exists(path): return {"valid": False, "file": path, "message": "Cookie 文件不存在"} try: with open(path, encoding="utf-8") as f: cookies = json.load(f) except Exception as e: return {"valid": False, "file": path, "message": f"读取失败: {e}"} now = int(time.time()) expired = [] soon = [] for c in cookies: exp = c.get("expiry") or c.get("expires", 0) if not exp: continue diff = exp - now if diff <= 0: expired.append(c["name"]) elif diff < 86400: soon.append(c["name"]) valid = len(expired) == 0 parts = [] if expired: parts.append(f"❌ 已过期：{', '.join(expired)}") if soon: parts.append(f"⚠️ 即将过期（24h内）：{', '.join(soon)}") if valid and not soon: parts.append("✅ 全部有效") return { "valid": valid, "file": path, "total": len(cookies), "expired": expired, "expiring_soon": soon, "message": " | ".join(parts), } def _find_cookies() -> str: """自动查找最新 Cookie 文件""" candidates = [ os.path.abspath(COOKIES_PRIMARY), os.path.abspath(COOKIES_ALT), ] # 返回最近修改的有效文件 valid = [p for p in candidates if os.path.exists(p)] if not valid: return "" return max(valid, key=os.path.getmtime) def get_best_cookies() -> str: """获取当前最佳 Cookie 路径（优先使用最新且有效的）""" for path in [os.path.abspath(COOKIES_PRIMARY), os.path.abspath(COOKIES_ALT)]: if os.path.exists(path): status = check_cookies(path) if status["valid"]: return path # 如果都过期，返回最新修改的（并给出警告） return _find_cookies() # ── 输出工具 ────────────────────────────────────────────────── def print_cookie_status(cookies_file: str = None): """打印 Cookie 状态（带颜色提示）""" status = check_cookies(cookies_file) print(f"Cookie 文件: {status['file']}") print(f"状态: {status['message']}") if not status["valid"]: print("⚡ 请重新运行：python3 xhs_login.py") return status["valid"] def save_md(content: str, path: str): """保存 Markdown 文件""" with open(path, "w", encoding="utf-8") as f: f.write(content) print(f"📄 Markdown 已保存：{path}") def parse_likes(v) -> int: """解析点赞数（支持'万'格式）""" if not v: return 0 s = str(v).replace(",", "").strip() if "万" in s: return int(float(s.replace("万", "")) * 10000) try: return int(s) except ValueError: return 0 FILE:xhsfenxi/SKILL.md --- name: xhsfenxi description: | 小红书博主分析全流程 skill — 拆解账号定位、提炼爆款选题公式、输出结构化报告和商业版 Word。内置三型博主分类体系（荒诞美学型 / 共鸣命名型 / 现实策略型），基于多位真实博主的深度分析沉淀。支持单账号深度拆解、多账号对比、自定义学习报告，证据分级（公开主页→截图→第三方公开资料→推断），所有交付物可一键转为 Markdown + 商业 Word。Use it when the user asks to analyze a Xiaohongshu blogger/account, extract viral topic formulas, compare creators, or produce business-grade deliverables. keywords: - xiaohongshu - RED - 小红书 - 小红书分析 - 博主分析 - 账号拆解 - 对标分析 - 爆款选题 - 爆款选题公式 - 内容策略 - 个人IP分析 - 创作者分析 - 三型博主 - 荒诞美学型 - 共鸣命名型 - 现实策略型 - 品牌符号 - 结构化总结报告 - Word报告 - 商业版报告 - 选题公式学习 - 博主拆解 - 小红书运营 - 内容账号分析 - 竞品分析 - 账号定位 - benchmark account - topic formula - creator research - content audit - account positioning - competitor analysis - archetype classification - viral topic - markdown report - word report - 小红书博主对比 - 爆款内容拆解 - 内容IP打法 metadata: openclaw: runtime: node: ">=18" --- # xhsfenxi — 小红书博主分析 Skill > 公开页分析 · 三型博主分类 · 爆款选题公式提炼 · 多账号对比 · Markdown/商业Word交付 --- ## When to Use Use this skill when the user asks to: - analyze a Xiaohongshu / RED creator account - identify which archetype a creator belongs to (荒诞美学型 / 共鸣命名型 / 现实策略型) - summarize an account's positioning, audience, style, or content pillars - extract a **viral topic formula** or reusable content method - compare two or more benchmark creators - learn what can be copied vs what must stay differentiated - turn the analysis into a **Markdown report**, **Word report**, or **business-style deliverable** Typical trigger phrases: - "分析这个小红书博主" - "拆一下这个账号" - "做爆款选题公式" - "帮我做对标分析" - "做成商业版 Word" - "这个博主是哪一型？" - "compare these two RED creators" - "extract the topic formula from this account" --- ## Three-Archetype Classification System Distilled from real analyses of multiple Xiaohongshu creators. ### Type A — 荒诞美学型 (Absurdist Aesthetics) **Representative archetype:** Core pattern: wraps serious or philosophical content in absurdist humor and high-quality visuals. | Dimension | Characteristics | |-----------|----------------| | Content kernel | Absurdist filter on everyday life; philosophical undertone | | What users get | Entertained + moved + aesthetically immersed | | Title mechanism | Contrast (grand setting × mundane action) + unified brand symbol | | Brand symbol | A single recurring tag, e.g. "（劲爆）" | | Expression style | Humorous, high-production, literary naming | | Commercial fit | High-end lifestyle, travel, photography, luxury brands | | Replication difficulty | High — depends on accumulated visual aesthetic sensibility | **Identifying markers:** - All content tied together by a signature phrase or visual symbol - Serious topics expressed lightly; absurd topics expressed earnestly - Titles feel like literary naming, not just descriptions --- ### Type B — 共鸣命名型 (Resonance & Naming) **Representative archetype:** Core pattern: translates personal experience into universally relatable life-stage propositions, named in memorable ways. | Dimension | Characteristics | |-----------|----------------| | Content kernel | Youth growth, worldview expression, life-stage naming | | What users get | Understood + named + given a new perspective | | Title mechanism | Proposition feeling + metaphor + judgment | | Brand symbol | Signature conceptual phrases and naming patterns | | Expression style | Warm, perceptive, aesthetically refined | | Commercial fit | Growth/education/creative platforms, mid-range lifestyle | | Replication difficulty | Medium — method learnable, requires genuine observation | **Identifying markers:** - Content "names" vague emotional states people couldn't articulate - Private experience → universal proposition - Titles read like thoughtful questions or redefinitions --- ### Type C — 现实策略型 (Reality & Strategy) **Representative archetype:** Core pattern: breaks down unspoken real-world rules and provides executable strategies for ordinary people to move upward. | Dimension | Characteristics | |-----------|----------------| | Content kernel | Ordinary woman's upward mobility, reality rule-breaking | | What users get | Validated + activated + empowered with actionable moves | | Title mechanism | Conflict words + counter-intuitive framing + strong stance | | Brand symbol | Self-labeling with perceived weakness (先夺走羞耻感) | | Expression style | Sharp, realistic, strong-attitude conclusions | | Commercial fit | Mass consumer, career, e-commerce, practical tools | | Replication difficulty | Medium — framework learnable; don't copy the surface aggression | **Identifying markers:** - Titles feel slightly improper but undeniably accurate - Each piece breaks an unspoken rule or collapses an information gap - Users feel "刺但有用" — uncomfortable but useful --- ### Mixed Formula (Advanced) The most powerful content often combines two archetypes: > **Type B resonance/naming × Type C reality/strategy** > = Content that both "understands you" and "tells you what to do next" Use `scripts/archetype.js` to quickly identify which archetype(s) an account fits. --- ## Core Operating Rules 1. **Use public evidence first.** Prefer search pages, homepage/profile pages, and other publicly visible surfaces. 2. **Do not fake detail-page access.** If post detail pages are blocked by risk control, say so clearly. 3. **Grade evidence quality.** Treat sources in this order: - Level A: visible Xiaohongshu public pages / user screenshots - Level B: other public sources (podcasts, interviews, encyclopedias, analytics sites) - Level C: synthesis / inference based on repeated visible patterns 4. **Do not present public estimates as backend truth.** Third-party follower or pricing data are supporting clues, not audited facts. 5. **Ask for 3–10 representative post links or screenshots** when the user wants deep post-level analysis. 6. **Deliver something useful even when blocked.** If detail pages fail, still produce a homepage-level strategic report. 7. **Always identify the archetype.** Every analysis should name which of the three types (or hybrid) the account belongs to — this frames the entire report. --- ## Default Workflow ### 1) Clarify the task shape Lock the requested output mode before heavy research: - Structured account report - Viral topic formula - Archetype identification - Two-or-more-account comparison - "How to learn from this creator without copying them" - Business-style Word deliverable Use `scripts/intake.js` for a fast intake template. ### 2) Gather source inputs Collect whatever the user already has: - account name - Xiaohongshu homepage/search URL - screenshots - external links (podcasts, news, interviews) - target output path - whether they need Word / clickable TOC / business styling ### 3) Research with the source ladder Read `references/workflow.md` for the full ladder. Preferred tool pattern: - `web_search` for public discovery - `browser` or `browser-use` for homepage/search-page inspection - `web_fetch` for readable public pages - `image` for screenshot analysis - `read` for any local Markdown reports already created ### 4) Classify the archetype Before deep analysis, make a preliminary archetype call: - Run `node scripts/archetype.js <creator-name>` to print the classification prompt - Answer: Type A (荒诞美学) / Type B (共鸣命名) / Type C (现实策略) / Mixed - The archetype frames the entire lens of the analysis ### 5) Build the account model Extract and write down the five layers: 1. **Identity** — who the creator is framed as 2. **Audience contract** — why people follow them 3. **Topic system** — what recurring problems/desires they cover 4. **Expression system** — how titles and openings work, including any brand symbol 5. **Transferability** — what another account can actually learn ### 6) Produce the deliverables Always output the three standard DOCX files directly — no intermediate Markdown step needed: | Output file | When to produce | |-------------|----------------| | `账号名-结构化总结报告.docx` | Single creator analysis | | `账号名-爆款选题公式.docx` | Single creator topic formula | | `选题公式学习-综合版.docx` | Multi-creator comparison | Use `scripts/docx-plan.js <creator-name>` to get the exact build command. See `references/workflow.md` for the Word TOC fix checklist if needed. --- ## Deliverables Three standard output files. Always produce as `.docx` directly. ### 账号名-结构化总结报告.docx Single creator full breakdown. Sections: 1. Account snapshot 2. Archetype classification + rationale 3. Strategic positioning 4. Audience and emotional contract 5. Content pillars 6. Title / hook system + brand symbol analysis 7. Narrative structure 8. Commercialization clues 9. What to learn 10. What not to blindly copy 11. Conclusion ### 账号名-爆款选题公式.docx Single creator topic formula. Sections: 1. Why this creator produces strong topics (archetype-based) 2. Total formula 3. 3–6 recurring topic models 4. Title formulas 5. Body structure formulas 6. Distribution / comment triggers 7. How to migrate to another account 8. 10–30 ready-to-use topic directions ### 选题公式学习-综合版.docx Multi-creator comparison. Sections: 1. Why compare these creators together 2. Archetype map 3. Shared foundations + key differences 4. Hybrid formula 5. Which path fits which use case 6. 30 combined topic directions 7. Final recommendation --- ## Word Output Run `node scripts/docx-plan.js <creator-name>` to get the exact build command. Standard build: `python3 build_docx6.py <creator-name>` If TOC / bookmark issues appear: run `inspect_docx.py` then `fix_final2.py`. See `references/workflow.md` for the full fix checklist. --- ## Output Files Three files, always `.docx`: - `账号名-结构化总结报告.docx` - `账号名-爆款选题公式.docx` - `选题公式学习-综合版.docx` (comparison / multi-creator) Use `scripts/report-plan.js <creator-name>` to confirm filenames. --- ## Scripts | Command | Purpose | |--------|---------| | `node scripts/intake.js` | Print a structured intake template for a new analysis task | | `node scripts/intake.js --json` | Same as above, JSON format | | `node scripts/archetype.js <creator-name>` | Print the archetype classification prompt | | `node scripts/archetype.js <creator-name> --quick` | Print a quick-read archetype cheatsheet | | `node scripts/report-plan.js <creator-name>` | Print recommended deliverables and filenames | | `node scripts/report-plan.js <A> <B> --mode compare` | Print a comparison deliverable plan | | `node scripts/docx-plan.js <creator-name>` | Print Word generation plan using build_docx*.py | --- ## Limitation Policy Always state which situation applies: - **Homepage-level only**: enough for strategic analysis, not enough for exact post-body claims - **Mixed-source analysis**: Xiaohongshu public pages + third-party public clues - **Deep-dive mode**: based on user-provided screenshots / links / transcripts If blocked by risk control, say so and continue with the strongest available public evidence. --- ## What Good Output Looks Like A strong output is: - structured - archetype-aware (names and reasons the type) - evidence-aware (grades sources) - strategically useful - honest about access limits - directly reusable in future reports Avoid vague praise, empty creator admiration, or unsupported claims about hidden metrics. --- ## References Read only when needed: - `references/workflow.md` — detailed evidence ladder, execution runbook, Word TOC fix guide - `references/templates.md` — report templates, archetype-specific section structures, naming patterns --- ## Proven Analyses Archive Three full analyses have been completed and archived in: ``` openclaw_cosmo/afa/小红书分析与工作流归档/01-分析报告与选题公式/ ``` | Archetype | Type | Key Insight | |-----------|------|-------------| | 荒诞美学型 | Type A | 品牌符号 + 反差标题 + 哲思轻量化输出 | | 共鸣命名型 | Type B | 私人经历→公共命题，给模糊状态命名 | | 现实策略型 | Type C | 困境→说破→规则→策略→爽感 | These archetypes are the ground truth for the classification system. --- *Version: 2.1.0 · Created: 2026-04-08 · Updated: 2026-04-09* FILE:xhsfenxi/_meta.json { "ownerId": "kn79bebfnwg15sb0g7cj5z5nyd83gxh0", "slug": "xhsfenxi", "version": "2.1.0", "publishedAt": null } FILE:xhsfenxi/package.json { "name": "xhsfenxi", "version": "2.1.0", "description": "xhsfenxi — Xiaohongshu creator-account analysis workflow with three-archetype classification (荒诞美学型/共鸣命名型/现实策略型), viral topic formula extraction, comparison studies, and Markdown/Word deliverables.", "keywords": [ "xiaohongshu", "RED", "小红书", "博主分析", "账号拆解", "爆款选题", "三型博主", "荒诞美学", "共鸣命名", "现实策略", "topic-formula", "creator-research", "content-strategy", "benchmark-analysis", "archetype-classification", "word-report", "business-report" ], "author": "cosmofang", "license": "MIT", "scripts": { "intake": "node scripts/intake.js", "intake:json": "node scripts/intake.js --json", "plan": "node scripts/report-plan.js", "archetype": "node scripts/archetype.js", "archetype:quick": "node scripts/archetype.js --quick", "docx-plan": "node scripts/docx-plan.js", "toc-fix": "node scripts/docx-plan.js --toc-fix" } } FILE:xhsfenxi/references/templates.md # Templates Reference — xhsfenxi v2.0 --- ## Output Files Always produce as `.docx` directly. Three standard files: | File | When | |------|------| | `账号名-结构化总结报告.docx` | Single creator analysis | | `账号名-爆款选题公式.docx` | Single creator topic formula | | `选题公式学习-综合版.docx` | Multi-creator comparison | --- ## Structured Report Skeleton (v2 — Archetype-Aware) ```markdown # 账号名结构化总结报告 > 说明：本报告基于[数据来源说明]。适合用于[账号定位/内容策略/竞品研究/爆款机制提炼]。 ## 一、执行摘要 [2–4段: 博主是谁，核心特征，最值得记住的一句话] ## 二、账号基础信息 [表格: 昵称/小红书号/粉丝/获赞收藏/笔记数/内容类型/账号阶段] [补充: 全平台历史数据（如有）] ## 三、博主背景（公开资料） [人物背景、职业起点、创作理念（来源于公开访谈）] ## 四、账号类型判断 **类型:** Type A/B/C 或混合型 **判断依据:** [2–3条可见信号] **核心内核:** [一句话总结] ## 五、博主核心内核拆解 [外壳是什么 / 真正的内核是什么 / 人设三层结构 / 最强能力] ## 六、内容结构拆解 [内容支柱分布（表格）/ 内容类型比例 / 统一主题] ## 七、标题机制与爆点分析 [最重要特征 / 品牌符号分析 / 高热内容分析（表格）/ 标题公式归纳] ## 八、与其他博主的横向对比（可选） [对比表格: 多维度 × 多博主] ## 九、商业化判断 [当前阶段 / 为什么品牌会找他 / 适合的合作方向 / 不适合的方向] ## 十、最值得学习的核心能力 [3–5个能力，每个含: 说明 + 可学习点] ## 十一、后续建议 [路径A: 深入单篇分析 / 路径B: 制作爆款选题公式 / 路径C: 跨博主综合版] ## 十二、最终总结 [1–2段核心结论] ## 附录：笔记完整列表（可选） [表格: 排名 / 类型 / 标题 / 点赞] ``` --- ## Viral Topic Formula Skeleton (v2) ```markdown # 账号名爆款选题公式 > 说明：本文件不是复刻[账号名]的表面语气，而是提炼真正有效的选题逻辑、标题机制、传播结构、人设打法与可迁移方法。 ## 一、先说结论：为什么能做出爆款？ [账号类型 + 5个叠在一起的核心能力 + 总公式一句话] ## 二、总爆款公式 [5步骤公式 + 压缩版关键词] ## 三、最常见的 3–6 类爆款选题模型 [每个模型: 公式 / 典型表达方向 / 为什么有效 / 可套用句式 / 适合主题] ## 四、标题公式：可以直接拆用 [5–6个标题公式，每个含: 结构说明 + 套用模板] ## 五、内容结构公式 [2–3个结构公式，每个含: 步骤 / 适合做的主题] ## 六、隐形公式 / 最值钱的底层打法 [3–4个] ## 七、如何迁移到自己的内容里 [4个步骤: 找现实母题 / 改写生活故事 / 用冲突词 / 4问检验] ## 八、可直接复用的 20 个选题方向 [编号列表] ## 九、哪些能学，哪些不能硬抄 [可以学的 / 不能硬抄的] ## 十、最终总结 [一句话核心结论] ``` --- ## Comparison Skeleton (v2 — Archetype Map) ```markdown # 选题公式学习（综合版） ## ——结合[博主A]与[博主B]（以及[博主C]）的内容方法论 > 说明：这份文档不是简单拼接报告，而是把各博主最有效的内容能力抽取出来，重新整理成可学习、可迁移、可直接应用的选题方法。 ## 一、为什么把这几位博主放在一起学？ [各博主关键词 / 各博主强项一句话 / 为什么组合更完整] ## 二、账号类型对照表 [表格: 博主 / 类型 / 核心方向 / 用户获得感 / 内容气质 / 标题风格 / 爆点来源] ## 三、共同底层：真正的选题不是题材，而是命题 [5个共同点 / 核心结论] ## 四、双系统（或多系统）选题公式 [系统A（共鸣命名型）: 适用场景 / 核心公式 / 压缩版 / 标题类型] [系统B（现实策略型）: 同上] [（如有）系统C（荒诞美学型）: 同上] ## 五、混合公式 [混合公式定义 + 例子 + 为什么更强] ## 六、标题公式综合版 [6–8个跨类型标题公式] ## 七、内容结构综合版 [结构A / B / C，各含步骤和用户获得感] ## 八、可直接复用的 30 个综合选题方向 [A组（偏共鸣命名）/ B组（偏现实策略）/ C组（偏荒诞美学）/ D组（混合）] ## 九、结论与行动建议 [最重要的学习提醒 + 最终公式] ``` --- ## Archetype Comparison Table Template Use this table when comparing 2–3+ accounts: ```markdown | 维度 | [博主A] | [博主B] | [博主C] | |---|---|---|---| | 账号类型 | Type A 荒诞美学 | Type B 共鸣命名 | Type C 现实策略 | | 核心方向 | | | | | 用户获得感 | | | | | 内容气质 | | | | | 标题风格 | | | | | 品牌符号 | | | | | 爆点来源 | | | | | 商业化方向 | | | | | 可复制难度 | | | | ``` --- ## Evidence Label Reference Use these labels in your internal working notes (do not need to show all in final report): | Label | Meaning | |-------|---------| | `A1` | Xiaohongshu homepage visible fact | | `A2` | Xiaohongshu screenshot fact | | `B1` | Third-party public clue (interview, news, etc.) | | `C1` | Synthesis / interpretation | --- ## Business Word Checklist When building a business-style Word file: - [ ] Cover page - [ ] TOC with `_TocH1_XXX` bookmarks (internal jump, not external link) - [ ] Page numbers - [ ] Conclusion page - [ ] Restrained typography (no loud decorative color blocks unless requested) - [ ] Tested in Word: TOC entries jump correctly - [ ] No duplicate headings accidentally inserted in body --- ## Brand Symbol Analysis Template For Type A accounts or any account with a unified recurring symbol: ```markdown ### 品牌符号分析 **符号:** [（劲爆）/ 其他] **出现频率:** [几乎所有 / 大部分 / 部分] **品牌效果:** - 识别度: [描述] - 反差机制: [描述] - 先发制人效果: [描述] **可学习点:** [其他账号如何建立自己的品牌符号] ``` FILE:xhsfenxi/references/workflow.md # Workflow Reference — xhsfenxi v2.0 --- ## Source Ladder Use sources in this order whenever possible: ### Level A — Primary public evidence Highest trust for visible claims: - Xiaohongshu search result pages - creator homepage/profile pages - user-provided screenshots from Xiaohongshu - user-provided post links that are actually accessible What you can safely claim from Level A: - visible follower / note counts shown on page - visible account ID / self-introduction - visible recent titles / visible public engagement numbers - posting cadence clues visible on page - obvious content categories - brand symbols and signature phrases (like "（劲爆）") ### Level B — Secondary public clues Use as support, never as backend truth: - podcast interviews - encyclopedia entries - public creator bios elsewhere - analytics websites / public creator databases - public brand collaboration listings What Level B is good for: - filling identity / background context - confirming repeated themes - spotting public commercialization clues - triangulating a bigger narrative ### Level C — Synthesis / inference Allowed only when clearly framed as interpretation. Examples: - "This account behaves more like a Type A 荒诞美学 creator than a Type B 共鸣命名 creator." - "The recurring topic logic appears to be: contrast × literary naming × unified symbol." - "The transferable lesson is not the tone, but the framing system." Never present Level C synthesis as raw observed fact. --- ## Standard Execution Runbook ### Step 1 — Confirm the deliverable Choose one or more: - structured report - viral topic formula - archetype classification - comparison report - customized learning report - Word / business version ### Step 2 — Capture the minimum task brief Minimum useful brief: - creator name - available links or screenshots - target output format - whether a business Word version is required ### Step 3 — Archetype pre-classification Before deep analysis, make a preliminary archetype call using visible signals: | Signal | Likely archetype | |--------|-----------------| | Unified brand symbol/tag on every post | Type A 荒诞美学型 | | High-production video, absurdist or philosophical content | Type A 荒诞美学型 | | Posts name vague emotions or life stages | Type B 共鸣命名型 | | Titles feel like thoughtful questions or redefinitions | Type B 共鸣命名型 | | Titles contain conflict words ("骗子", "不要脸", "装") | Type C 现实策略型 | | Content breaks unspoken workplace/relationship/money rules | Type C 现实策略型 | | Mix of resonance + strategy | Mixed B+C | ### Step 4 — Research in layers 1. Find the account and public homepage/search data 2. Save visible account facts (follower count, note count, visible titles, engagement) 3. Pull representative visible titles — note brand symbols, title patterns 4. Add external public clues only where helpful 5. Record limitations before interpretation ### Step 5 — Extract the account system (Five Layers) Always map these five layers: 1. **Identity** — who the creator is framed as; any self-labeling strategy 2. **Audience contract** — why people follow them; what emotional/strategic need is met 3. **Topic system** — what recurring problems/desires they cover; 3–6 topic models 4. **Expression system** — title formulas, brand symbols, opening patterns 5. **Transferability** — what another account can actually learn; what must not be copied ### Step 6 — Produce the right file #### Structured report Recommended top-level sections: 1. Account snapshot 2. Archetype classification + rationale 3. Positioning judgment 4. Audience and demand 5. Content pillars 6. Title / hook system (including brand symbol analysis) 7. Narrative structure 8. Commercialization clues 9. What to learn 10. What not to copy 11. Conclusion 12. (Optional) Full note list sorted by engagement #### Viral topic formula Recommended top-level sections: 1. Why this creator produces strong topics (archetype-based) 2. Total formula 3. 3–6 recurring topic models 4. Title formulas 5. Body structure formulas 6. Distribution / comment trigger clues 7. How to migrate it to another account 8. 10–30 topic directions #### Comparison report (2–3+ accounts) Recommended top-level sections: 1. Why compare these creators together 2. Archetype map (each creator's type) 3. Shared foundations 4. Key differences 5. Hybrid formula 6. Which path fits which use case 7. Suggested topic directions 8. Final recommendation --- ## Archetype-Specific Analysis Lenses ### Type A — 荒诞美学型 Focus questions: - What is the brand symbol? (unified tag, phrase, or visual marker) - How does the title create contrast? (grand × mundane, serious × absurd) - Where does the humor come from? (earnest treatment of absurd things) - What is the underlying philosophical/emotional core? - Why is replication hard? (visual sensibility, accumulated aesthetic) Key learning for other accounts: - Find your own "brand symbol" — a unifying tag that creates recognition - Train "contrast naming" — don't describe the place, name the feeling - Practice "absurdist framing" — pick topics that are ridiculous when taken seriously --- ### Type B — 共鸣命名型 Focus questions: - What life stages or emotional states does the account name? - How does private experience become universal proposition? - What are the signature conceptual phrases? - What is the "new way to understand yourself" users get? Key learning for other accounts: - Practice "命题化" — reframe personal experience as a universal human situation - Build a personal vocabulary of named states ("Odyssey时期", etc.) - The formula: experience → proposition → naming → judgment → resonance --- ### Type C — 现实策略型 Focus questions: - What real-world rules does the account break open? - How does the self-labeling strategy work? (antifragile identity) - What is the conflict word in each title? - What is the "can-do action" users walk away with? Key learning for other accounts: - Find your "reality母题" — the real困境 you speak to - Practice rewriting weak framings into strong rule-breaking framings - The formula: 困境 → 说破 → 规则 → 策略 → 爽感 --- ## Limitation Language Use direct wording like: - "This round is based primarily on public homepage/search-page evidence." - "Single-post detail pages were not stably accessible due to platform risk control." - "Third-party figures are used as directional clues, not audited backend data." - "Archetype classification is based on visible patterns; the creator may operate differently at the full content level." --- ## Deep-Dive Upgrade Path If the user wants finer analysis, ask for 3–10 representative posts, ideally with: - screenshots of cover + title - screenshots of body pages / subtitles - comments screenshots - transcript or summary if video-based Then upgrade from account-level to post-level analysis. --- ## Word Output Workflow ### Overview Always generate Markdown first, then convert to Word. Iterating Markdown is cheaper than re-generating DOCX. ### Script Selection Guide Use the workflow archive scripts at: ``` openclaw_cosmo/afa/小红书分析与工作流归档/02-Word生成与目录修复脚本/ ``` | Script | Use when | |--------|----------| | `build_docx6.py` | Default — most stable, use this first | | `build_docx.py` → `build_docx5.py` | Earlier iterations, kept for reference/debugging | | `inspect_docx.py` | Inspect internal XML, bookmarks, link structure | | `fix_final2.py` | Final TOC / attribute fix after build | | `fix_attrs.py` | Fix XML namespace attribute writing errors | | `fix_wrong.py` | Targeted corrections for known wrong output | | `check_it.py` | Verify output correctness | ### Word TOC Fix Checklist When a generated .docx has broken TOC links: 1. Run `inspect_docx.py` — check `word/document.xml` for bookmark names and link types 2. Verify heading bookmarks use format `_TocH1_XXX` (not `#_TocH1_XXX`) 3. Verify internal links use Word-native anchor style, not external URL rels 4. If `elem.set()` errors appear, use full namespace: `elem.set('{' + W_NS + '}id', value)` 5. Run `fix_final2.py` after any XML edits 6. Test by opening in Word — click TOC entries to verify jump ### Common Errors | Error | Cause | Fix | |-------|-------|-----| | Bookmark name wrong format | Missing `_TocH1_` prefix | Rename via `fix_attrs.py` | | `anchor` contains `#` | Wrong anchor format | Strip `#` prefix | | TOC uses external link rels | Used `http://` link type for internal jump | Switch to `w:instr` field or anchor | | Duplicate headings in body | TOC headings accidentally inserted into body | Use `fix_wrong.py` to clean | | Namespace attribute error | Missing full namespace URI in `elem.set()` | Use `'{' + W_NS + '}id'` form | ### Best Practice - Keep only `build_docx6.py` as the production script - Keep all earlier versions as debugging reference, not for production - One Markdown → one `build_docx6.py` call → inspect → fix if needed FILE:xhsfenxi/scripts/archetype.js #!/usr/bin/env node /** * xhsfenxi — archetype.js * Purpose: Print the archetype classification prompt for a Xiaohongshu creator. * Helps identify whether an account is Type A (荒诞美学), Type B (共鸣命名), * Type C (现实策略), or a hybrid — based on the three proven archetypes * distilled from real analyses of multiple Xiaohongshu creators. * Usage: * node scripts/archetype.js <creator-name> * node scripts/archetype.js <creator-name> --quick */ const args = process.argv.slice(2); const quick = args.includes('--quick'); const name = args.filter(a => !a.startsWith('--'))[0]; if (!name) { console.error('Usage: node scripts/archetype.js <creator-name> [--quick]'); process.exit(1); } if (quick) { console.log(` === xhsfenxi archetype cheatsheet === TYPE A — 荒诞美学型 (Absurdist Aesthetics) Example: 荒诞美学博主（vlogger型） Signals: - Unified brand symbol/tag on every post (e.g. "（劲爆）") - Absurdist or philosophical topics treated earnestly - High-production video; literary title naming - Serious × absurd contrast in every piece Formula: 荒诞滤镜下的生活哲学 TYPE B — 共鸣命名型 (Resonance & Naming) Example: 成长世界观表达博主 Signals: - Posts name vague emotional states or life stages - Private experience → universal proposition - Titles feel like thoughtful questions or redefinitions Formula: 经历 → 命题 → 命名 → 判断 → 共鸣 TYPE C — 现实策略型 (Reality & Strategy) Example: 普通女孩上行策略博主 Signals: - Titles contain conflict words ("骗子", "不要脸", "装") - Content breaks unspoken workplace/relationship/money rules - Self-labels with perceived weakness to pre-empt criticism Formula: 困境 → 说破 → 规则 → 策略 → 爽感 MIXED B+C — Most powerful hybrid Combines: resonance/naming × reality/strategy = "You understand me" + "Now I know what to do" `); process.exit(0); } console.log(` === xhsfenxi archetype classification: ${name} === TASK Classify the Xiaohongshu account "${name}" into one of these three archetypes (or a hybrid), based on visible public evidence. ───────────────────────────────────────────── THREE ARCHETYPES ───────────────────────────────────────────── TYPE A — 荒诞美学型 (Absurdist Aesthetics) Core: Wraps philosophical/serious content in absurdist humor + high visual quality Proven example: 荒诞美学vlogger型 (unified brand symbol; grand × mundane contrast; literary naming) Key signals to look for: - Does every post share a unified recurring symbol or phrase? - Do titles contrast a grand/serious setting with a mundane/absurd action? - Is the visual/production quality noticeably higher than peers? - Are serious topics treated with humor and humor treated earnestly? TYPE B — 共鸣命名型 (Resonance & Naming) Core: Names vague emotions and life stages so users feel "finally, someone said it" Proven example: 成长世界观表达型 (youth growth, worldview expression, concept naming) Key signals to look for: - Do posts give names or definitions to emotional states most people can't articulate? - Does the account turn personal stories into universal life propositions? - Do titles feel like thoughtful philosophical questions or redefinitions? - Is the tone warm, perceptive, aesthetically refined? TYPE C — 现实策略型 (Reality & Strategy) Core: Breaks unspoken real-world rules; provides executable strategies for ordinary people Proven example: 普通女孩上行策略型 (antifragile identity; rule-breaking; 普通人上行) Key signals to look for: - Do titles contain conflict words or counter-intuitive framing? - Does the account self-label with a perceived weakness to pre-empt criticism? - Does content expose hidden rules in workplace/relationships/money/consumption? - Does every piece end with a clear, actionable conclusion or stance? MIXED HYBRID Some accounts combine two archetypes. Most powerful is B+C: "I understand your situation (Type B) AND here's what you can actually do (Type C)" ───────────────────────────────────────────── CLASSIFICATION PROMPT ───────────────────────────────────────────── Based on publicly visible data for "${name}", answer: 1. Which archetype does this account most closely match? (A / B / C / Mixed) 2. What are 2–3 concrete visible signals that support this classification? 3. If mixed: what percentage is each type, and does one dominate? 4. What is the single sentence that describes this account's core "filter" on the world? 5. Is there a brand symbol or signature phrase? If yes, what is it and how does it function? Then proceed with the full analysis using this archetype as the lens. `); FILE:xhsfenxi/scripts/docx-plan.js #!/usr/bin/env node /** * xhsfenxi — docx-plan.js * Purpose: Print the Word generation plan for a completed Xiaohongshu analysis, * including which build_docx*.py script to use, the full workflow steps, * and TOC fix guidance from the workflow archive. * Usage: * node scripts/docx-plan.js <creator-name> * node scripts/docx-plan.js <creator-name> --mode comparison * node scripts/docx-plan.js --toc-fix */ const ARCHIVE_BASE = '~/Desktop/cosmocloud/Deeplumen/cosmowork/openclaw_cosmo/afa/小红书分析与工作流归档/02-Word生成与目录修复脚本'; const args = process.argv.slice(2); const tocFix = args.includes('--toc-fix'); const modeIdx = args.indexOf('--mode'); const mode = modeIdx !== -1 ? args[modeIdx + 1] : 'single'; const name = args.filter((a, i) => { if (a.startsWith('--')) return false; if (modeIdx !== -1 && i === modeIdx + 1) return false; return true; })[0]; if (tocFix) { console.log(` === xhsfenxi Word TOC Fix Guide === Archive scripts location: ${ARCHIVE_BASE}/ STEP-BY-STEP FIX PROCESS 1. Run inspect_docx.py to diagnose the issue: python3 ${ARCHIVE_BASE}/inspect_docx.py <your-file.docx> → Check: bookmark names, link types, heading structure 2. Verify the following are correct: - Bookmark names follow format: _TocH1_XXX (NOT #_TocH1_XXX) - Internal TOC links use Word-native anchor, NOT external http:// rels - No duplicate headings accidentally inserted in document body - Namespace attributes written as: elem.set('{' + W_NS + '}id', value) 3. Run the appropriate fix script: - For attribute/namespace errors: python3 fix_attrs.py <file.docx> - For final TOC + heading cleanup: python3 fix_final2.py <file.docx> - For known wrong output patterns: python3 fix_wrong.py <file.docx> 4. Verify with check_it.py: python3 ${ARCHIVE_BASE}/check_it.py <your-fixed-file.docx> 5. Open in Word and test: click each TOC entry to confirm internal jump works. COMMON ERROR TABLE Bookmark format wrong → rename using fix_attrs.py anchor contains # → strip # prefix in fix_final2.py TOC uses external rels → switch to w:instr field or Word anchor Duplicate headings → fix_wrong.py XML namespace error → use full namespace URI in elem.set() `); process.exit(0); } if (!name && !tocFix) { console.error('Usage: node scripts/docx-plan.js <creator-name> [--mode comparison]'); console.error(' node scripts/docx-plan.js --toc-fix'); process.exit(1); } const isComparison = mode === 'comparison'; const mdFile = isComparison ? `选题公式学习-综合版.md` : `${name}-结构化总结报告.md`; const docxFile = isComparison ? `选题公式学习-综合版.docx` : `${name}-结构化总结报告-商业版.docx`; console.log(` === xhsfenxi Word generation plan: ${name || 'comparison'} === Mode: ${isComparison ? 'comparison / 综合版' : 'single creator'} ───────────────────────────────────────────── STEP 1 — Confirm Markdown is complete ───────────────────────────────────────────── Make sure this file is finalized: ${mdFile} ───────────────────────────────────────────── STEP 2 — Run build_docx6.py (recommended) ───────────────────────────────────────────── Primary script (most stable): python3 ${ARCHIVE_BASE}/build_docx6.py ${mdFile} ${docxFile} If build_docx6.py fails, fall back to earlier versions: build_docx5.py → build_docx4.py → build_docx3.py ───────────────────────────────────────────── STEP 3 — Inspect the output ───────────────────────────────────────────── python3 ${ARCHIVE_BASE}/inspect_docx.py ${docxFile} Check for: ✓ Headings have _TocH1_XXX bookmarks ✓ TOC links are internal (not external URL rels) ✓ No duplicate heading text in body ✓ Page count and section structure look correct ───────────────────────────────────────────── STEP 4 — Fix if needed ───────────────────────────────────────────── Run: node scripts/docx-plan.js --toc-fix for the full TOC fix checklist. Most common fix: python3 ${ARCHIVE_BASE}/fix_final2.py ${docxFile} ───────────────────────────────────────────── STEP 5 — Final check ───────────────────────────────────────────── python3 ${ARCHIVE_BASE}/check_it.py ${docxFile} → Open in Word, click TOC entries, verify internal jumps work ───────────────────────────────────────────── EXPECTED OUTPUT FILES ───────────────────────────────────────────── ${docxFile} ${isComparison ? '' : ` ${name}-结构化总结报告.md (keep as source of truth)\n ${name}-爆款选题公式.md (if also produced)`} ───────────────────────────────────────────── ARCHIVE LOCATION ───────────────────────────────────────────── Scripts: ${ARCHIVE_BASE}/ Stable production script: build_docx6.py Inspection tool: inspect_docx.py `); FILE:xhsfenxi/scripts/intake.js #!/usr/bin/env node /** * xhsfenxi — intake.js * Purpose: Print a structured intake template for a Xiaohongshu creator-analysis task. * Usage: * node scripts/intake.js * node scripts/intake.js --json */ const args = process.argv.slice(2); const asJson = args.includes('--json'); const template = { task: 'xiaohongshu-creator-analysis', creatorNames: ['<creator-name>'], links: ['<homepage-or-search-url>'], screenshots: ['<optional-local-path-or-url>'], deliverables: ['structured-report'], outputDir: '<optional-output-dir>', needWord: false, needBusinessVersion: false, needComparison: false, notes: 'What exactly should be learned from this creator?' }; if (asJson) { console.log(JSON.stringify(template, null, 2)); process.exit(0); } console.log(` === xhsfenxi intake template === Provide or confirm the following before deep research: - creator name(s) - homepage/search URL(s) - screenshots or links for representative posts (optional but helpful) - requested deliverable(s): structured-report / topic-formula / comparison / business-word - output directory - any specific question to answer JSON template: ${JSON.stringify(template, null, 2)} `); FILE:xhsfenxi/scripts/report-plan.js #!/usr/bin/env node /** * xhsfenxi — report-plan.js * Purpose: Print recommended deliverables and filenames for a Xiaohongshu analysis request. * Usage: * node scripts/report-plan.js <creator-name> * node scripts/report-plan.js <creator-a> <creator-b> --mode compare */ const args = process.argv.slice(2); const modeIdx = args.indexOf('--mode'); const mode = modeIdx !== -1 ? args[modeIdx + 1] : 'single'; const names = args.filter((a, i) => { if (a === '--mode') return false; if (modeIdx !== -1 && i === modeIdx + 1) return false; return true; }); if (!names.length) { console.error('Usage: node scripts/report-plan.js <creator-name>'); console.error(' node scripts/report-plan.js <creator-a> <creator-b> --mode compare'); process.exit(1); } if (mode === 'compare') { if (names.length < 2) { console.error('Compare mode requires two creator names.'); process.exit(1); } const [a, b] = names; const files = [ `选题公式学习-综合版.docx` ]; console.log(` === xhsfenxi report plan === Mode: comparison Creators: ${a} / ${b} Recommended deliverables: - comparison report - hybrid topic-formula study - optional business Word version Recommended filenames: ${files.map(f => `- ${f}`).join('\n')} `); process.exit(0); } const name = names[0]; const files = [ `${name}-结构化总结报告.docx`, `${name}-爆款选题公式.docx` ]; console.log(` === xhsfenxi report plan === Mode: single creator Creator: ${name} Recommended deliverables: - structured report - viral topic formula - optional business Word version Recommended filenames: ${files.map(f => `- ${f}`).join('\n')} `); </article></body></html>

Xhsfenxi Pro

Comments (0)