Methodology Origins Horizontal-Vertical Analysis was proposed by Digital Life Khazix, integrating diachronic-synchronic analysis in linguistics (Saussure), longitudinal-cross-sectional research design in social sciences, the case study method of business schools, and the core ideas of competitive strategy analysis, forming a general research framework applicable to products/companies/concepts/people. The core principle remains unchanged: the vertical axis pursues temporal depth, the horizontal axis pursues contemporaneous breadth, and ultimately they converge into a judgment.
You are conducting an in-depth research using Horizontal-Vertical Analysis. The final output is abeautifully typeset PDF research report。
Preliminary Preparation
Environment Preparation
- Confirm the PDF conversion script is available: This Skill includes
scripts/md_to_pdf.py(based on WeasyPrint), used to convert the final Markdown report into a beautifully typeset PDF. Ensure the dependencies are installed:pip install weasyprint markdown --break-system-packages。 - Writing Style: This Skill has a complete built-in writing style guide (see the "Writing Style" section below), no need to load other skills.
Clarify the Research Subject
After receiving the user input, confirm the following information. If the user has already provided enough clarity (for example, "help me research Hermes Agent using Horizontal-Vertical Analysis"), no need to ask further, just start:
- Research subject: the specific product name/company name/concept name/person name
- Type: product, company, concept, person, or other?
- Research motivation(optional): Why study it? What happened recently?
- Special focus(optional): Is there any direction you particularly want to go deeper into?
Step One: Online Information Collection
The quality of this methodology depends entirely on the richness and accuracy of the information.Must search online, cannot rely solely on existing knowledge. The value of a research report lies in its depth and completeness, so in the information collection stage, it is better to search more; do not let insufficient information cause the later analysis to be superficial.
Parallel Search Strategy
Use sub-agents to search in parallel to improve efficiency. Suggested division of labor:
- Sub-agent 1 — Vertical information: the research subject's origin, founder background, development history, key events, version iterations, financing, strategic pivots, crises
- Sub-agent 2 — Horizontal information: competitor identification, each competitor's characteristics and user reputation, industry comparative reviews, market share
- Sub-agent 3(only needed for complex subjects): supplementary information, such as founders' in-depth backgrounds, changes in the industry environment, user community discussions (GitHub issues, Reddit, Twitter/X, Zhihu, etc.)
Guide to Using Online Tools for Sub-agents(write directly into each sub-agent's prompt):
Each sub-agent's prompt must include the following online guidance:
You need to obtain information online. Use the following tools:
- WebSearch: Used to search and discover information sources, and obtain summaries and keyword results
- WebFetch: When a specific URL is known, used to extract content from the page in a targeted manner
- If the web-access skill is installed in the user's environment (check the path
/mnt/.claude/skills/web-access/SKILL.mdto see whether it exists), load it first and follow its guidance; it provides stronger browser CDP capabilities- Search strategy: first use WebSearch to discover information sources and leads, then after finding specific URLs use WebFetch for in-depth extraction
- Search multiple times and with multiple keyword combinations; do not give up after searching only once
- First-hand sources are better than second-hand sources: official blogs > original reporting by authoritative media > reprints/aggregations
- For academic research subjects, arxiv must be checked: If the research subject involves academic concepts, algorithms, AI models, technical paradigms, etc., relevant papers must be obtained through the arxiv API. How to call it:
curl -s "https://export.arxiv.org/api/query?search_query=all:关键词1+AND+all:关键词2&max_results=10", or use WebFetch to access the same URL. It returns XML format, including title, authors, abstract, publication date, and PDF link. Keyword combinations and the number of results can be adjusted as needed. After finding key papers, use WebFetch to read the paper pages (https://arxiv.org/abs/论文ID) to obtain more details.
The prompt should describe the goal ("obtain", "research", "understand"), and should not use verbs that imply specific methods ("search", "crawl"), so that the sub-agent can independently determine the best way to obtain information.
Information source priority
First-hand sources are better than second-hand sources; multiple media outlets citing the same error can create the illusion of circular corroboration:
| Information type | First-hand source |
|---|---|
| Product updates/technical decisions | Official blogs, GitHub Release Notes, founder tweets |
| Financing/business data | Official company announcements, SEC/business registration documents |
| User reputation | GitHub Issues, Reddit discussions, Twitter/X, Zhihu posts |
| Industry analysis | Original reporting by authoritative media (not reprints) |
| Academic/technical principles | arXiv papers (export.arxiv.org/api/query), Google Scholar, academic conference proceedings |
Information sufficiency self-check
After searching is complete, check:
- Vertical: Can a complete story be told? Are there obvious information gaps?
- Horizontal: Is the competitor list complete? Are any major players missing? Is there enough information on each competitor to make a comparison?
- Sources: Are key facts supported by reliable sources? Are any judgments based on only a single source?
If the information is insufficient, search again to fill the gaps. Do not settle for less.
Step 2: Diachronic / Longitudinal analysis
Along the timeline, fully reconstruct the overall development of the research subject from its birth to the present. This is the main body of the report, and its length should be the greatest.
Content requirements
Origin tracing: What was the background of its birth? What technology/idea/need was it based on? Who were the founding team or core promoters? What had these people done before, and why were they the ones to do this? What was the industry environment like at the time? Was there a key event or inspiration that directly led to its birth?
Birth milestone: The clear time of first release/founding/proposal, the initial form and positioning, and how it differs from now.
Evolution process: From its birth to the present, lay out all key milestones in chronological order. Including but not limited to: major version updates, financing events, team changes, strategic transformations, technical architecture changes, user-scale milestones, major partnerships or acquisitions, public relations crises or controversies.
Decision Logic: At each key milestone, reconstruct as much as possible the reasons behind the decisions. Why was A chosen instead of B? What constraints were faced at the time? Which early decisions "locked in" later directions of development, making them difficult to reverse? What mechanisms made it go deeper and deeper (network effects, ecosystem lock-in, technology stack choices, etc.)?
Phase Division: Naturally divide the entire journey into several phases (emergence phase, rapid growth phase, transformation phase, etc.), each with its core characteristics and core contradictions.
Length
6,000-15,000 words. The longer the history and the more milestones an object has, the closer to the upper limit; new things lean toward the lower limit. The core principle is to tell the story completely and thoroughly; every key milestone is worth expanding on, and important details should not be skipped just to compress. Better to write long and detailed than to touch lightly.
Step 3: Synchronic / Cross-sectional
Taking the current point in time as a cross-section, conduct a comprehensive comparison of the research object with competitors/similar players in the same track.
First, determine the competitive situation
Handle it in three scenarios:
Scenario A: No direct competitors. If the research object is a completely new category or a highly exclusive field, skip the one-by-one comparison and instead analyze: Why does it have no competitors? Is the category too new, are the barriers too high, or is the market too small? From which direction is a competitor most likely to emerge in the future? Are there indirect alternatives or previous-generation solutions that can be referenced?
Scenario B: A small number of competitors (1-2). Compare them one by one in depth, with detailed analysis of each competitor.
Scenario C: Sufficient competitors (3 or more). Select the 3-5 most representative ones for comparison, and briefly mention the rest.
Comparison Dimensions
Adjust flexibly according to the type of research object, but at least cover the following aspects:
Core Difference Comparison: Technical route/core methodology/underlying logic, product form/business model/organizational structure, target users/audience/use cases, core advantages and obvious shortcomings, pricing strategy/resource investment/scale.
User Perspective: What is the real user reputation of each competitor? What are the most frequently mentioned advantages and complaints in community reviews and user experience? Is there any deviation between users' actual usage and the official positioning? The comparison should not be written as a text version of a specification table; it should clearly explain what each competitor "has become" and what the real reasons are that users choose it.
Niche Analysis: In the overall map of the track, what position does the research object occupy? What gap does it fill, or who is it competing with head-on? Is the current landscape one of many players blooming, a rivalry between two strong players, or one dominant player?
Trend Judgment: Based on the horizontal comparison, what is the trajectory of the research object in the competitive landscape? What are the opportunities and risks?
Length
3,000-10,000 words. For Scenario A, keep it to about 3,000 words; for Scenario C, expand each major competitor into at least 1,500 words of independent analysis, and do not give them only a passing mention.
Step 4: Cross-dimensional Intersection Insights
This is the essence paragraph of the entire report. Combine the vertical development trajectory with the horizontal competitive landscape to offer a comprehensive, new judgment. Do not write it as an abridged version of the preceding content.
Core questions that need to be answered:
- How history shaped the current competitive position: Which decisions and events in the vertical journey determined its position in today's horizontal comparison?
- Vertical comparison of competitors: If the main competitors are also placed on the timeline, how do their origins and evolution paths differ? How did these differences lead to their respective characteristics today?
- Historical roots of advantages: Each of today's core advantages—which node or decision in history can it be traced back to?
- Historical roots of disadvantages: Each of today's core disadvantages—which historical decision can it be traced back to? Have the "good decisions" of the past become today's burdens?
- Future projections: Based on vertical trends and the horizontal competitive landscape, give three scenarios—the most likely, the most dangerous, and the most optimistic. Each scenario must have logical support.
Length
1500-3000 words.
Writing style
This is not a cold consulting report, but an in-depth study that people can read from beginning to end. The writing style needs to find a balance between "the rigor of a research report" and "the readability of Khazix."
Core elements borrowed from Khazix's style
The following stylistic elements are directly applied to report writing (for detailed definitions, please refer to the khazix-writer skill):
Sense of rhythm: Sentences vary in length, and transitions between paragraphs are natural. Do not make every paragraph the same length; the technique of making a single sentence its own paragraph to create a sense of weight can be used. Good rhythm is like a wave—each time it drifts a little off the main line, then a "tie-back-to-the-main-line sentence" pulls it back.
Narrative-driven, not list-driven: The vertical section should have a story arc, with setup, development, turn, and resolution. For example, why a product suddenly exploded at a certain point in time, what the buildup behind it was, and what the turning point was. Do not write it as a running account like "released A in January 2023, released B in March 2023."
Knowledge is "pulled out casually while chatting": Naturally bring in background knowledge during the telling, not "now let me explain it to everyone."
Dare to make judgments: Encouraged to give opinions and insights, but every opinion must be supported by facts. First present the facts, then give the judgment. Clearly label speculation. When expressing judgments, use a stance that acknowledges subjectivity, such as "I think" or "my judgment is," rather than a condescending definitive conclusion.
Layer-by-layer rhetorical peeling: Do not state the conclusion directly; unfold it through "phenomenon → surface explanation → deeper questioning → core insight." Let readers participate in the thinking process.
Cultural elevation: In the intersection insights section, connect to larger cultural/philosophical/historical references. It should not be a forced elevation, but feel like "naturally thinking of it while chatting."
Circular echo: Details and hooks planted at the beginning or in the vertical section should be called back in the intersection insights or the ending. The sense of closure in cause and effect before and after is the key to turning the report from an "information stream" into a "work."
Elements not borrowed from Khazix's style
The following elements suit WeChat public account articles but not research reports, and need restraint:
- Excessive colloquialism: The report can have a conversational feel, but don't fill it with "this thing," "bro, no way," "so badass." A sprinkling here and there is fine, but the density should be much lower than in a WeChat official account article.
- Remove subheadings: WeChat official account articles aim to flow smoothly without adding subheadings. Research reports are different; if a 10,000-30,000-word piece has no clear structure and navigation, readers will get lost. Reports need a clear chapter structure.
- The punctuation ban can be relaxed: WeChat official account articles ban colons and em dashes. In research reports they can be used normally, because what a report needs is information delivery efficiency. But the habit of using 「」 can be retained.
- Fixed tail: Do not add the WeChat official account triple/star tail.
Absolute no-go zones (still apply)
The following AI-smell markers must be avoided no matter the style:
- Stock phrases: "First... second... finally," "In summary," "It is worth noting," "It is not hard to find"
- Empty adjectives: "empower," "grip point," "build a closed loop"
- Textbook openings: "In today's era of rapid AI development," "With the continuous advancement of technology"
- High-frequency landmine words: "to put it bluntly," "what does it mean?" "this means," "essentially," "in other words," "undeniably"
- Vague tool names: do not say "AI tool" or "some model"; say the specific name
- Fabricated scenarios: if some information cannot be found, honestly mark it as "this information is temporarily unavailable"; never fabricate
Write in plain human language
Avoid consulting-company-style stock phrases and empty generalizations. Replace general statements with concrete details and examples. For example, do not write "the company achieved rapid growth at this stage," but write "from $10 million ARR in mid-2024 to $1 billion by the end of 2025, the growth curve was almost vertical."
Step 5: Generate the PDF report
After the report is written, use the scripts/md_to_pdf.py script included with this Skill to convert the Markdown into a beautifully typeset PDF.
Conversion process
- First complete the Markdown manuscript: Write the complete report in standard Markdown format and save it as
[研究对象]_横纵分析报告.md - Install dependencies(if not installed):
pip install weasyprint markdown --break-system-packages - Run the conversion script：
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" - The script will automatically generate an intermediate HTML file (for debugging convenience) and the final PDF
The script's built-in typesetting specifications
md_to_pdf.py already has a complete CSS typesetting scheme built in, no manual adjustment needed:
- Page: A4, margins top 25mm/left and right 20mm/bottom 20mm
- Cover page: Automatically generated, including title (28pt dark blue), subtitle "Horizontal and Vertical Analysis Method In-Depth Research Report," author information, decorative divider
- Color scheme: H1 title = #1a5276 dark blue, H2 = #1e8449 green, H3 = #2e86c1 light blue, H4 = #5b2c6f purple, body text = #2c3e50 dark gray
- Fonts: CSS fallback chain
"Droid Sans Fallback", Helvetica, Arial, sans-serif, automatically handles mixed Chinese and English typesetting - Body text: 10.5pt, line spacing 1.75, justified, orphan/widow control
- Blockquote: 3pt dark blue vertical line on the left + light gray background
- Table: full width, dark blue header with white text, zebra-striped rows
- Header: "Report Title | Horizontal and Vertical Analysis Method In-Depth Research Report" (not displayed on the first page)
- Footer: "Page X" (not displayed on the first page)
- The first H1 in Markdown will be automatically extracted as the cover title and will not appear again in the body
Markdown Writing Notes
To ensure the script parses correctly and generates the best PDF results:
- Use
# 标题on the first line as the report title (automatically used for the cover) - Immediately after the title, you can use the
> 研究时间：... | 所属领域：... | 研究对象类型：...format to write metadata lines, which will be extracted to the cover - Use
##as the main section headings (vertical analysis, horizontal analysis, vertical-horizontal intersection, etc.) - Use
###and####as subsections - Use standard Markdown table syntax for tables
- Use
>syntax for quotes - Use
**文本**
End Content
Add at the end of the Markdown manuscript:
- Information Sources: A list of all cited sources, with URLs and access times noted
- Methodology Note: Briefly explain the source of the vertical-horizontal analysis method (1-2 sentences is enough)
Report Structure Template
封面页
目录
一、一句话定义
[用一句话说清楚这个东西是什么]
二、纵向分析：从诞生到当下
[完整的纵向叙事，6000-15000字]
三、横向分析：竞争图谱
[横向对比分析，3000-10000字]
四、横纵交汇洞察
[交叉分析和未来推演，1500-3000字]
五、信息来源
[所有引用的来源列表]
File Naming and Delivery
Name the PDF file [研究对象名称]_横纵分析报告.pdf, and save it to the user's working directory.
Adaptation for Different Types of Research Subjects
The core principle remains unchanged (the vertical axis pursues temporal depth, the horizontal axis pursues breadth in the same period), but the emphasis differs:
When researching a product: The vertical axis focuses on version iterations, evolution of the technology roadmap, user growth curves, and key product decisions; the horizontal axis focuses on feature comparisons, performance comparisons, user experience, and pricing.
When researching a company: The vertical axis focuses on the founding team, financing history, strategic pivots, organizational changes, and key personnel changes; the horizontal axis focuses on business model differences, market share, revenue comparisons, and organizational structure differences.
When researching a concept(technological paradigm, business model, cultural phenomenon): The vertical axis focuses on the origin of the concept (who proposed it, based on what theory/need), how it became popular, and what debates and evolutions it underwent; the horizontal axis focuses on differences from similar concepts, their respective applicable scenarios, and the arguments of different camps.
When researching a person: The vertical axis focuses on personal experience, career trajectory, key decisions, growth curve, and changes in public statements; the horizontal axis focuses on comparisons with other people in the same field (differences in approach, style, achievements, influence, and path choices).
Length Overview
| Section | Word Count Range | Description |
|---|---|---|
| Vertical Analysis | 6,000 - 15,000 words | The main body of the report; do not skim the surface |
| Horizontal Analysis | 3,000 - 10,000 words | Adjust according to the number of competitors |
| Vertical-Horizontal Intersection | 1,500 - 3,000 words | The highlight section, providing new judgments |
| Full Text Total | 10,000 - 30,000 words | Do not be afraid of length; depth and completeness are where the value lies |
Quality Inspection Checklist
Self-check before delivery:
- Is the vertical axis a narrative storytelling style? Does it read with causal logic and the context of the times? Is it not a chronological ledger?
- Is the background and motivation of the founder/initiator sufficiently deep?
- Is every key node written out in full, without skipping important details for the sake of compression?
- Is the decision logic reconstructed? Not just "what happened," but also "why this choice was made"?
- Is the judgment of the competitor scenarios on the horizontal axis correct (A/B/C)? Is the competitor analysis deep enough?
- Does the user reputation section cite real users' voices? Not just official propaganda?
- Does the intersection of horizontal and vertical produce new judgments, not a condensed version of the earlier content?
- Do all three scenarios in the future projection have logical support?
- Does the writing style have rhythm and readability? Not a cold consulting report?
- Has none of the absolute red lines been violated?
- Are all key facts annotated with information sources?
- Is information that cannot be found honestly marked as "temporarily unavailable," without fabrication?
- Is the PDF layout aesthetically pleasing, clearly structured, and highly readable?
- Is the total word count within the range of 10,000-30,000 words?