hv-analysis
A deep research skill based on the Horizontal-Vertical Analysis method. Proposed by Digital Life Khazix, it integrates core concepts from Saussure's diachronic-synchronic analysis, social science longitudinal and cross-sectional research designs, business school case study methodologies, and competitive strategy analysis. Use this when a user wants to systematically research a product, company, concept, technology, or person. The core is a dual-axis analysis: the vertical axis traces the full lifecycle from origin to present (presented as a narrative story), the horizontal axis conducts a systematic comparative analysis against competitors/counterparts at the current cross-section of time, and finally, the intersection of both axes yields unique insights. The final output is an exquisitely formatted PDF research report. Trigger phrases include but are not limited to: horizontal-vertical analysis, research this, help me analyze, deep research, do a study, investigate, competitor analysis, see how this thing is, what is the deal with this product/company/concept, figure this out for me, help me understand, do a deep research for me. Even if the user simply says "help me learn about XX" or "what is the background of XX", it should trigger as long as the context implies a need for systematic deep research (rather than a simple concept explanation). It also applies when a user drops a product name, company name, or technical term saying "help me research this". DO NOT use for simple definition lookups (user just asking "what is XX"), DO NOT use for WeChat Official Account article writing (use khazix-writer for that), and DO NOT use for purely generating titles/summaries (use wechat-title).
1How It Works & Mechanism
The methodology is "Dual-Axis": the vertical axis reconstructs the subject's complete trajectory chronologically from inception to present, the horizontal axis benchmarks competitors across the current time cross-section, and the intersection of both axes generates fresh conclusions—each of the three sections enforces strict word count ranges to mandate depth over superficial summaries.
## 第二步：纵向分析（Diachronic / Longitudinal） 沿时间轴，完整还原研究对象从诞生到现在的发展全貌。这是报告的主体部分，篇幅应该最重。
Quality prerequisites are defined as hard requirements: drafting is prohibited if materials are insufficient, prioritizing over-searching during the information-gathering phase.
这个方法论的质量完全取决于信息的丰富度和准确性。**必须联网搜索**，不能仅靠已有知识。研究报告的价值在于深度和完整度，所以信息收集阶段宁可多搜，不要因为信息不够导致后面的分析浮于表面。
Concrete web-access actions are mapped to tools and endpoints: the primary agent dispatches sub-agents to search concurrently, embedding WebSearch/WebFetch usage guidelines and curl invocation examples for "mandatory arXiv checks for academic subjects" into each sub-agent's prompt.
**WebSearch**：用于搜索发现信息来源，获取摘要和关键词结果
Optional enhancement: If the user's environment has the web-access skill installed (including browser CDP capabilities), prioritize loading it.
如果用户环境中安装了 web-access skill（检查路径 `/mnt/.claude/skills/web-access/SKILL.md` 是否存在），优先加载它并遵循其指引，它提供更强的浏览器CDP能力
A source tiering system is codified in the text: primary sources take precedence over secondary sources (official blogs/Release Notes/founder tweets > authoritative media originals > syndicated aggregations), accompanied by a primary source mapping table for each information category.
### 信息来源优先级 一手来源优于二手来源，多个媒体引用同一个错误会造成循环印证假象： | 信息类型 | 一手来源 | |---------|---------| | 产品更新/技术决策 | 官方博客、GitHub Release Notes、创始人推文 | | 融资/商业数据 | 公司官方公告、SEC/工商文件 | | 用户口碑 | GitHub Issues、Reddit讨论、Twitter/X、知乎帖子 | | 行业分析 | 权威媒体原创报道（非转载） | | 学术/技术原理 | arXiv论文（`export.arxiv.org/api/query`）、Google Scholar、学术会议论文集 |
PDF generation relies on a built-in script: SKILL.md provides installation commands for dependencies (pip install weasyprint markdown) and execution commands, with the script embedding a complete CSS styling suite (A4, cover page, 4-level heading color schemes, headers/footers, zebra-striped tables).
python [skill目录]/scripts/md_to_pdf.py input.md output.pdf --title "研究对象名称" --author "数字生命卡兹克"
Writing style and "absolute taboos" are kept separate: maintaining Khazix's readability (rhythm, narrative arc, layer-by-layer unravelling, thematic echoes) alongside the structural navigation of a research report, while enumerating AI clichés and fabricated scenarios that must be avoided.
### 绝对禁区（依然适用） 以下AI味标记无论什么文体都要避免： - 套话："首先...其次...最后"、"综上所述"、"值得注意的是"、"不难发现" - 空洞形容词："赋能"、"抓手"、"打造闭环" - 教科书开头："在当今AI快速发展的时代"、"随着技术的不断进步" - 高频踩雷词："说白了"、"意味着什么？"、"这意味着"、"本质上"、"换句话说"、"不可否认" - 空泛工具名：不说"AI工具"、"某个模型"，要说具体名字 - 编造场景：如果某个信息搜不到，诚实标注「该信息暂缺」，绝不编造
A pre-delivery 14-item quality checklist is enforced (narrative style, decision-making logic, competitor scenario classification, quoting authentic user feedback, evidence-backed three-scenario models, source citations, word count ranges, etc.).
## 质检清单 交付前自检： - [ ] 纵轴是叙事故事体？读起来有因果逻辑和时代脉络？不是年表流水账？ - [ ] 创始人/发起者的背景和动机有足够深度？ - [ ] 每个关键节点都展开写了，没有为了压缩而跳过重要细节？ - [ ] 决策逻辑有还原？不只是「发生了什么」，还有「为什么这么选」？ - [ ] 横轴的竞品场景判断正确（A/B/C）？竞品分析够深？ - [ ] 用户口碑部分引用了真实用户的声音？不只是官方宣传？ - [ ] 横纵交汇产出了新的判断，不是前面内容的缩写版？ - [ ] 未来推演的三个剧本都有逻辑支撑？ - [ ] 写作风格有节奏感、有可读性？不是冷冰冰的咨询报告？ - [ ] 没有触犯绝对禁区里的任何一条？ - [ ] 所有关键事实标注了信息来源？ - [ ] 搜不到的信息诚实标注了「暂缺」，没有编造？ - [ ] PDF排版美观、结构清晰、可读性好？ - [ ] 总字数在 10,000-30,000 字的范围内？
2Core Capabilities
3External Dependencies
| Kind | Dependency / Endpoint |
|---|---|
| network | arXiv API（学术类研究对象必查） |
| network | arXiv 论文页（精读） |
| cli | curl |
| package | WeasyPrint + markdown（Python 包，PDF 渲染） |
| api | 宿主联网工具 WebSearch / WebFetch（未点名具体搜索后端） |
4Recommended Governance & Evidence
- scripts/: Local script or command execution: scripts/md_to_pdf.py (local Python run by user or agent): reads Markdown, writes intermediate HTML and final PDF; no
- scripts/ 或 SKILL.md: Declares dynamic installation of external language package dependencies
- 网络端点: Declares external network access: GET https://export.arxiv.org/api/query?search_query=... — paper retrieval for academic research subjects (SKIL
- 本地工作目录: File I/O and artifact output: Research stage: `[Research_Subject]_Horizontal_Vertical_Analysis_Report.md` (Markdown draft saved to user's working directory)
5Second-Pass Verification
- [ok] arxiv API 调用点真实存在 — SKILL.md 子 agent 指引段给出 `curl -s "https://export.arxiv.org/api/query?search_query=all:关键词1+AND+all:关键词2&max_results=10"`，并说明返回 XML（标题/作者/摘要/日期/PDF 链接），调用点真实。
- [ok] md_to_pdf.py 无网络/无子进程 — 脚本 import 仅 sys/os/re/argparse/markdown；无 requests/urllib/socket/subprocess；行为＝读 md、写 html、WeasyPrint 输出 pdf。
- [ok] CSS 与正文所述排版规范对应 — CSS_TEMPLATE 含 @page A4 margin 25mm/20mm、「第 X 页」@bottom-center、@page :first 隐藏页眉页脚等，与 SKILL.md「脚本内置的排版规范」一致。
- [ok] 第三方依赖与安装命令 — SKILL.md 两处写 `pip install weasyprint markdown --break-system-packages`（环境准备与生成 PDF 步骤各一次），已标为 yellow 的附加风险点。
- [ok] web-access skill 路径探测 — 原文为「如果用户环境中安装了 web-access skill（检查路径 `/mnt/.claude/skills/web-access/SKILL.md` 是否存在），优先加载它并遵循其指引」——是本地路径存在性判断，不产生网络请求。
- [ok] 字数指标一致性 — 纵向 6,000-15,000 + 横向 3,000-10,000 + 交汇 1,500-3,000 与「全文总计 10,000-30,000 字」及质检清单末条吻合。
- [ok] 来源分级与防循环印证 — 「一手来源优于二手来源，多个媒体引用同一个错误会造成循环印证假象」+ 五行信息类型/一手来源表逐字核对通过。
- [ok] 无凭证读取 — API_KEY/token/secret/keychain 零命中；全文不涉及登录态或私有 API。
6Verdict & Scenarios
9659c7f314c71b58…