Skip to content

jasonhnd/makedocskill

Repository files navigation

make-doc-skill

AI-Powered Visual Document Generator


Transform complex information into beautifully designed, self-contained HTML documents.

License Claude Code agentskill.sh Version

18 visual templates x 28 content strategies x S/M/L/XL depth control. Brain -> Router -> Canvas architecture.

This is not template filling. This is consulting-grade visual intelligence.

See it in action · Install · How it works · Templates

Other Languages:

日本語 · 中文


Why This Skill Exists

Most AI tools generate documents the same way: dump text into a template and call it done. You get a wall of bullet points in a generic card layout, maybe with a gradient blob for decoration. It looks like every other AI-generated document on the internet.

The problem is not generation speed. The problem is that no thinking happens before the pixels hit the screen.

A McKinsey consultant does not open PowerPoint first. They spend 80% of their time on analysis — identifying core tensions, challenging assumptions, mapping causal chains — and only 20% on the final deliverable. The document is the last mile, not the first.

make-doc-skill works the same way. It enforces a Brain -> Router -> Canvas pipeline that separates thinking from rendering:

  • Brain reads your materials, extracts core arguments, evaluates evidence strength, and challenges your assumptions through expert-level sparring. A finance document gets questioned like a Goldman Sachs analyst would. A technical architecture gets stress-tested like a Google Staff Engineer would.
  • Router matches your content to the optimal visual structure from 18 templates and 28 content strategies. A comparison is not forced into a timeline. A tutorial is not crammed into a dashboard.
  • Canvas renders pixel-perfect HTML that strictly follows your brand CI — colors, fonts, border-radius, shadows — extracted automatically from your website.

The 8-step pipeline is not bureaucracy. Each step exists because skipping it produces measurably worse output. Step 4 (Sparring) alone typically identifies 3-5 blind spots that would have become embarrassing gaps in the final document. Step 6 (Review) enforces a 15-item quality checklist across structure, content depth, and CI compliance.

The result: documents that look like they came from a design agency, with analysis depth that reads like it came from a consulting firm. All from a single Claude Code skill.


Architecture

┌──────────────────────────────────────────────────────────────┐
│                      make-doc-skill                          │
│                                                              │
│   ┌──────────┐     ┌──────────┐     ┌──────────────────┐    │
│   │  Brain    │────>│  Router  │────>│     Canvas       │    │
│   │          │     │          │     │                  │    │
│   │ Logical   │     │ Template │     │ Visual           │    │
│   │ Analysis  │     │ Selection│     │ Rendering        │    │
│   │ Sparring  │     │ Strategy │     │ HTML Generation  │    │
│   │ Evidence  │     │ Mapping  │     │ CI Compliance    │    │
│   └──────────┘     └──────────┘     └──────────────────┘    │
│                                                              │
│   Steps 3-4           Step 5           Step 7               │
└──────────────────────────────────────────────────────────────┘

Core Principle: Depth comes from prompt quality, not the number of steps.

The three layers are decoupled by design. Even when Brain performs deep analysis, Canvas does not output analytical prose — it transforms analysis into visual structure. Router ensures this translation happens correctly.


18 Visual Templates

The optimal template is automatically recommended based on content type.

Template Use Case Visual Characteristics
Dashboard Data-dense overviews, KPIs Bento-box grid + CSS progress bars
Versus A vs B comparison Dual-column color separation + unified comparison table
Timeline History, phase breakdown Central axis + alternating cards
Article News, opinion pieces Hero + quote blocks + alternating text/image
Steps Tutorials, how-to guides Numbered circles + connector dashed lines
Report Research, finance Summary section + sidebar numbers + tables
Matrix Multi-dimensional evaluation Score grid + color scale + ranking
Profile Product, company intro Hero banner + metric bars + card grid
Flowchart Decision processes Nodes + CSS arrows + branching
Scorecard Evaluation, reviews Total score + ring progress + bars
Pitch Strategic proposals Full-width narrative + CTA
Infographic Data analysis Large numbers + icons + gradient sections
Comparison Table Product comparison Check/cross marks + recommended column highlight
FAQ Knowledge base Accordion Q&A
Slide Team presentations 16:9 fixed + page numbers
Iceberg Root cause analysis Surface (30%) + waterline + deep layer (70%)
Funnel Conversion analysis Progressively narrowing + elimination metrics
Bridge Transformation proposals Current state -> transition path -> target

28 Content Strategies

Each strategy has dedicated analysis angles, prohibited outputs, and recommended templates.

Strategy Primary Template Analysis Angles
breaking-news Article 5W1H, source credibility, timeline, impact assessment
opinion Article Core thesis, evidence chain, opposing views, logical gaps
product Profile Core features, differentiation, target users, pricing
tech-doc Steps Architecture, API, failure modes, TCO
research Report Research questions, methodology, data support, limitations
finance Report Core metrics, trends, risks, peer comparison
startup Pitch Team, problem, solution, market size, business model
case-study Report Background -> challenge -> solution -> results -> lessons
data-analysis Infographic Data sources, trends, outliers, correlations, predictions
review Scorecard Scoring dimensions, pros/cons, horizontal comparison
tutorial Steps Prerequisites, step breakdown, common errors
ecommerce Comparison Table Product attributes, price comparison, user reviews
interview Article Subject background, core insights, key quotes
event Timeline Agenda, speakers, key statements, action items
whitepaper Report Abstract, problem definition, methodology, conclusions
legal-policy Report Scope, core provisions, compliance requirements
press-release Article Publisher, core message, market impact
social-thread Article Core viewpoints, engagement heat, controversy points
job-posting Profile Role profile, core requirements, team culture
recipe Steps Ingredient list, step breakdown, key techniques
travel Dashboard Destination overview, itinerary, cost estimates
health Steps Symptoms, possible causes, treatment options
real-estate Dashboard Property overview, location analysis, price trends
podcast Article Guest introduction, topic timeline, core insights
announcement Article Core changes, scope of impact, required actions
sports Scorecard Event overview, key data, turning points
video Article Theme extraction, key timestamps, core insights
general Dashboard Core tension, multi-layer causation, So What validation

Depth Control (S / M / L / XL)

S M L XL
Analysis Depth Core conclusions + actions Standard analysis + evidence evaluation Deep analysis + explanatory frameworks Complete report + multi-perspective review
Sections 3-5 5-8 8-12 12+
Depth Visualization Basic Standard (confidence marks, contrast) Full (causal chains, trends, multi-source) Full + interactive (collapsible, citations)
Use Case Quick grasp of key points Team sharing Decision support Strategic reports
Estimated Time 15-30 min 30-60 min 60-120 min 120+ min

Installation

Platform Installation
Claude Code cp SKILL.md ~/.claude/skills/make-doc-skill/SKILL.md
Gemini CLI / Antigravity npx skills add jasonhnd/makedocskill
OpenAI Codex CLI npx skills add jasonhnd/makedocskill
Claude.ai Upload SKILL.md to Project Knowledge
Cursor / VS Code Copilot npx skills add jasonhnd/makedocskill
ChatGPT / Gemini Web Paste SKILL.md content into System Instructions

Quick Install

# One-liner
mkdir -p ~/.claude/skills/make-doc-skill && curl -sL https://raw.githubusercontent.com/jasonhnd/makedocskill/main/SKILL.md -o ~/.claude/skills/make-doc-skill/SKILL.md

# Or clone and copy
git clone https://github.com/jasonhnd/makedocskill.git
cp makedocskill/SKILL.md ~/.claude/skills/make-doc-skill/SKILL.md

Verify

# In Claude Code
/make-doc-skill

Usage

Simple Request

You: I want to make a document about our product architecture
Skill: What's the project name? (e.g., acme_product_arch)
You: acme_product_arch
Skill: [Creates output/acme_product_arch/, starts CI extraction...]
Skill: Do you have a company website I can crawl for brand CI?
You: https://acme.com
Skill: [Crawls site, extracts colors/fonts/shadows, saves 01_design.md]
Skill: Now 8 quick questions about audience, depth, format...

From Files

You: @report.pdf Turn this into a strategic overview for the leadership team
Skill: [Reads PDF, extracts key findings, identifies content type as "research"]
Skill: [Crawls company website for CI, runs expert sparring]
Skill: I found 3 blind spots in the analysis: ...
Skill: [Generates 16:9 slide deck with full CI compliance]

Quick Summary

You: Summarize this URL as a one-pager (Size S)
Skill: [Fetches URL, identifies content type, selects template]
Skill: [Fast 3-5 section summary, core conclusions + action items]
Skill: [Outputs single-page scroll HTML in under 20 minutes]

Expert Questioning Mode

During sparring (Step 4), expert-level questioning chains are automatically injected based on content type. These are not generic "you are an analyst" prompts — they are domain-specific interrogation patterns.

Finance

You are a TMT sector analyst (CFA Level 3) with 10 years at Goldman Sachs. "Is revenue growth organic or M&A-driven? What's the FCF vs NI divergence? How reliable is management guidance? Any hidden time bombs on the balance sheet?"

Technology

You are a Staff Engineer with 12 years at Google. Experienced 3 major tech migrations. "What's the core abstraction? Failure modes at 10x scale? Lock-in risks? TCO comparison?"

News

You are a senior editor at Reuters' investigative unit. Processing 200+ stories daily. "Primary evidence? Speculation? Source bias? Risk of reversal within 72 hours?"


Depth Visualization

When materials contain certain data structures, corresponding visualization methods are automatically applied. Degradation to text lists is prohibited.

Type Visualization Rule
Causal Chains Vertical flow + arrows + mechanism at each node Parallel cards prohibited
Confidence Horizontal progress bar + percentage + rationale 4-level color scale
Contrast Increments GAP column + directional arrows + color-coded differences Up/down/right arrows required
Information Quality Green/yellow/red circles for data reliability [DATA_GAP] shown as warning callout
Trend Direction CSS mini bar + directional arrows + derivative calculation Rate of change required
Uncertainty Independent section + dashed border + info gap list Cannot be mixed with confirmed data

Pipeline Transparency

Every document goes through 8 steps. Each step produces a visible artifact. You can restart from any step at any time.

Step Action Output
0 Project Name output/[name]/ directory
1 CI Extraction 01_design.md (colors, fonts, shadows, spacing)
2 Pre-Interview (8 questions) 02_interview.md (audience, purpose, format, depth)
3 Material Organization 03_source.md (structured sources with gap markers)
4 Sparring 04_sparring.md (hypothesis challenge + blind spots + reframing)
5 Composition 05_composition_[lang].md (section-by-section design spec)
6 Review 06_review.md (3-layer, 15-item quality check)
7 HTML Generation [title]_[lang].html (final self-contained output)

Output Formats

Single-Page Scroll

For online viewing, internal wikis, and knowledge bases. Vertical scroll layout with responsive design.

  • Responsive (mobile-first)
  • Glass morphism cards
  • Micro-interactions (hover effects)
  • Fade-in animations
  • Print-optimized with @media print

16:9 Slide Deck

For projection, team presentations, and PDF output. Fixed at 1280x720px.

  • Page numbers (bottom-right X / N)
  • CI color bar (bottom 4px accent bar)
  • Cover page with CI dark background
  • Auto page break on print
  • One-click PDF export via browser print

CI Fidelity

Auto-Extraction

In Step 1, the skill crawls your website and extracts: brand colors (primary, secondary, background, text), fonts (Google Fonts declarations), border-radius values, box-shadow patterns, spacing rhythm, and overall tone.

Color Discipline

During generation, the CI color palette is strictly enforced. Hierarchy is differentiated using color-mix() transparency variations. Introduction of hues outside the CI palette is prohibited.

CI Compliance Review

In Step 6, Layer 3, six items are forcibly checked before generation: border-radius / box-shadow / color palette / font-family / heading color / secondary color usage.


Technical Specifications

Spec Detail
Output Self-contained single HTML file
External Dependencies Tailwind CSS CDN, Google Fonts, Lucide Icons (inline SVG)
CSS Architecture Custom properties (--c-bg, --c-surface, --c-accent, etc.)
Print Support @media print + break-inside: avoid
Auto-Generated OG tags (SVG title card), favicon (SVG from CI), SEO (noindex, nofollow)
Multilingual Japanese / Chinese / English (auto-switch interaction)
Version Control Built-in history tracking, restart from any step

Token Usage

Scenario Estimated Tokens Estimated Cost
Size S (quick summary) ~15k ~$0.38
Size M (standard) ~25k ~$0.63
Size L (deep analysis) ~40k ~$1.00
Size XL (full report) ~60k ~$1.50
Restart from Step 4 ~20k ~$0.50

Claude Max/Pro subscribers are not billed per token.


Directory Structure

Repository

makedocskill/
├── SKILL.md              # Skill body (~2,360 lines)
├── README.md             # This file
├── LICENSE               # Apache License 2.0
├── .gitignore
├── i18n/
│   ├── ja/README.md      # Japanese
│   └── zh/README.md      # Chinese
└── docs/
    ├── README.md              # Document index
    ├── ARCHITECTURE.md        # Three-layer decoupled architecture
    ├── TEMPLATES.md           # 18 template reference
    ├── CONTENT_STRATEGIES.md  # 28 strategy details
    ├── DEPTH_VISUALIZATION.md # Depth visualization rules
    ├── SIZE_DEPTH_MATRIX.md   # Size-depth matrix
    ├── DESIGN_RULES.md        # Universal design rules
    ├── PIPELINE.md            # 8-step pipeline details
    └── PROMPT_ENGINEERING.md  # Prompt engineering standards

Generated Project

output/[project_name]/
├── README.md                  # Project index (auto-updated)
├── 00_assets/                 # Original files
├── 01_design.md               # CI design standards
├── 02_interview.md            # Pre-question records
├── 03_source.md               # Material organization
├── 04_sparring.md             # Sparring records
├── 05_composition_[lang].md   # Composition table
├── 06_review.md               # Review results
├── [title]_[lang].html        # Final output
└── *_history/                 # Version history

Documentation

Document Content
ARCHITECTURE.md Brain -> Router -> Canvas three-layer architecture
TEMPLATES.md Complete reference for 18 templates
CONTENT_STRATEGIES.md Details for 28 strategies
DEPTH_VISUALIZATION.md 6 depth visualization rules
SIZE_DEPTH_MATRIX.md Size-depth matrix
DESIGN_RULES.md Universal design rules
PIPELINE.md 8-step pipeline details
PROMPT_ENGINEERING.md Prompt engineering standards

Heritage

This skill is derived from the EIP (Explain In Picture) project — an AI-driven visual document generation web application (Next.js 16 + React 19 + Tailwind CSS 4) that accumulated the Engine 1.0 methodology: three-stage pipeline, 31 visual template skeletons, 28 content strategies, a 6-layer classification engine, depth visualization rules, and S/M/L/XL size control.

make-doc-skill condenses these insights into Claude Code skill format, achieving equivalent generation quality without a web application.


License

Apache License 2.0

About

AI-powered HTML visual document generator for Claude Code. 18 templates × 28 content strategies × S/M/L/XL depth control. Brain→Router→Canvas architecture.

Topics

Resources

License

Stars

Watchers

Forks

Packages

 
 
 

Contributors