Diagramming

Diagramming Skill

Generate professional diagrams with semantic coloring (Cagle palette) and accessibility compliance.

CRITICAL: Always Use Mermaid/DOT Over ASCII Art

NEVER use ASCII art for diagrams. When you encounter ASCII art diagrams in existing documents or when you need to create any visual representation:

1. Replace existing ASCII diagrams with Mermaid or DOT equivalents
2. Create new diagrams using Mermaid or DOT, never ASCII
3. Convert on sight - if you see ASCII art representing a flow, architecture, sequence, or any visual concept, convert it to a proper diagram

ASCII art is:

• Not accessible (screen readers can't interpret it)
• Not maintainable (hard to modify)
• Not portable (breaks with font changes)
• Not professional (looks dated)

Mermaid/DOT diagrams are:

• Accessible (with accTitle/accDescr)
• Maintainable (text-based, version-controllable)
• Renderable (to PNG/SVG for documentation)
• Professional (clean, consistent styling)

Tool Selection

Default: Start with Mermaid. Switch to DOT when pure network topology is primary concern.

Diagram Type Router

Load only ONE guide per request. Match user intent to the most specific keywords:

Decision Logic (Specificity Order)

Match most specific first. Check in this order:

Default: If unclear, ask user to clarify or suggest options from 10-USE-CASE-SCENARIOS.md.

Essential Configuration (Always Available)

ELK Layout (Use for Complex Diagrams)

---
config:
  layout: elk
  elk:
    mergeEdges: false
    nodePlacementStrategy: BRANDES_KOEPF
---

When to use ELK: >10 nodes, dense connections, knowledge graphs, complex architectures

Theme Configuration (Base Theme Required)

%%{init: {
  "theme": "base",
  "themeVariables": {
    "primaryColor": "#E3F2FD",
    "primaryTextColor": "#0D47A1",
    "primaryBorderColor": "#1565C0",
    "lineColor": "#37474F"
  }
}}%%

CRITICAL: Only base theme supports customization. Only hex colors work.

Cagle Color Palette (Memorize)

Architecture Colors

Status Colors

Knowledge Graph Colors (Semantic)

classDef Template (Copy-Paste Ready)

classDef infra fill:#E3F2FD,stroke:#1565C0,stroke-width:2px,color:#0D47A1
classDef service fill:#E8F5E9,stroke:#2E7D32,stroke-width:2px,color:#1B5E20
classDef data fill:#FFF8E1,stroke:#F57F17,stroke-width:2px,color:#E65100
classDef user fill:#F3E5F5,stroke:#7B1FA2,stroke-width:2px,color:#4A148C
classDef process fill:#E1F5FE,stroke:#0277BD,stroke-width:2px,color:#01579B
classDef security fill:#E0F2F1,stroke:#00695C,stroke-width:2px,color:#004D40
classDef external fill:#ECEFF1,stroke:#455A64,stroke-width:2px,color:#263238
classDef success fill:#C8E6C9,stroke:#2E7D32,stroke-width:2px,color:#1B5E20
classDef warning fill:#FFF9C4,stroke:#F9A825,stroke-width:2px,color:#F57F17
classDef error fill:#FFCDD2,stroke:#C62828,stroke-width:2px,color:#B71C1C

Output Format

When generating diagrams, always:

1. Use backticks (``` ) not tildes (~~~) for code fences - required for renderer compatibility
2. Start with configuration if using ELK or custom theme
3. Declare classDef styles before using them
4. Use semantic colors based on entity type
5. Add comments for complex diagrams explaining structure
6. Include the complete diagram - never truncate

CRITICAL: Always use triple backticks (```) for mermaid/dot code blocks, never triple tildes (~~~). Renderers and export tools require backtick fences.

Document Structure (After Export)

After exporting diagrams to PNG, the document should have:

1. The rendered PNG image (using <img> tag for size control)
2. A collapsible <details> block containing the mermaid source

CRITICAL: For image sizing, MUST use <img> tag with width attribute. NEVER use CSS.

NOTE: Diagrams export to PNG by default for maximum compatibility with markdown previews (VS Code, GitHub, Zed). Use --format=svg only when SVG is specifically needed.

<img src="diagrams/doc/diagram-name.png" alt="Diagram Name" width="90%">

<details>
<summary>Mermaid Source</summary>

` ``mermaid
flowchart LR
    A --> B
` ``

</details>

Size Guidelines:

This preserves the source for future editing while keeping the document clean.

Standard Output Structure

---
config:
  layout: elk  # Only if needed
---
%%{init: {"theme": "base", "themeVariables": {...}}}%%
diagramType
    accTitle: Brief diagram title for screen readers
    accDescr: One-line description of what the diagram shows

    %% Style definitions
    classDef type1 fill:...,stroke:...

    %% Diagram content
    Node1[Label]:::type1
    Node1 --> Node2

Accessibility Directives

Always include accessibility metadata for screen readers:

flowchart LR
    accTitle: User Authentication Flow
    accDescr: Shows login request from client through API gateway to auth service

    %% For multi-line descriptions:
    accDescr {
        Detailed description spanning
        multiple lines for complex diagrams
    }

    Client --> Gateway --> AuthService

MANDATORY: Image sizing — generate high-res, display small via CSS, never upscale

Load-bearing rule for any document with diagrams:

1. Generate PNGs at HIGH NATIVE RESOLUTION (the renderer at ~/.claude/tools/mermaid-renderer/render-mermaid.js does this by default — viewport 4800×3200, deviceScaleFactor: 6, producing PNGs typically 1920px+ wide depending on diagram bbox). Do NOT lower the resolution defaults.

2. Display via CSS or <img> width attribute, never via browser upscaling. A small PNG stretched to fill a container produces blur. A large PNG displayed at smaller width via CSS produces a crisp image AND a high-res asset available for click-to-fullscreen.

3. The HTML export's lightbox/pan-zoom viewer opens images at NATIVE resolution on click. Combined with high-res PNGs, this means a user sees the full-fidelity diagram when they zoom in — without the browser ever upscaling.

4. Display sizing hierarchy (priority order):

   • Best: <img src="..." alt="..." width="800"> — explicit width in HTML, scales proportionally, predictable across renderers.
   • Acceptable: CSS rules max-width: 100%; max-height: 700px on the img selector — what the markdown-export skill's HTML template uses by default.
   • Avoid: width: 100% without max-width (can upscale on wide screens).
   • Never: style="image-rendering: pixelated" or any forced upscaling.

5. Recommended widths:

   Diagram type	<img> width
   Simple flowcharts	800 (≈70% column)
   Complex architectures	1100 (≈90% column)
   Sequence diagrams	900 (≈80% column)
   State machines	700 (≈60% column)
   ER diagrams	1000 (≈85% column)

6. CSS in HTML exports must include:

   img { max-width: 100%; max-height: 700px; height: auto; }
   .image-viewer img { max-width: none; max-height: none; }
   

   This shrinks display, lets the viewer show native size.

7. Same rule applies to DOT/Graphviz exports (~/.claude/tools/dot-renderer/) — render at high DPI, display via CSS, never upscale at the browser.

If the rendered PNG looks blurry in the document, the cause is almost always browser upscaling, not the renderer. Verify with identify <png> — the PNG should be ≥1500px wide for any architecture diagram.

MANDATORY: Line Breaks — ALWAYS <br/>, NEVER \n

Mermaid does NOT interpret \n as a line break. It renders as the literal characters \n in the output. This applies everywhere: node labels, edge labels, subgraph titles, and any quoted string.

ALWAYS use <br/>. NEVER use \n. No exceptions.

WRONG:  A["First line\nSecond line"]         → shows literal \n
WRONG:  A -->|"line one\nline two"| B        → shows literal \n
CORRECT: A["First line<br/>Second line"]      → actual line break
CORRECT: A -->|"line one<br/>line two"| B     → actual line break

Before writing ANY mermaid code, mentally replace every \n with <br/>. After writing, search the block for \n — if found in any label or string, replace it.

MANDATORY: Reserved Keywords — Never Use as classDef Names

Mermaid reserves these keywords. Using them as classDef names causes silent parse failures (diagram renders as "Syntax error in text" with no useful message):

WRONG:  classDef class fill:#E1BEE7,stroke:#6A1B9A
WRONG:  classDef graph fill:#FFF9C4,stroke:#F9A825
CORRECT: classDef cls fill:#E1BEE7,stroke:#6A1B9A
CORRECT: classDef namedGraph fill:#FFF9C4,stroke:#F9A825

Validation Gate (MANDATORY before export)

After creating or modifying mermaid diagrams in a document, ALWAYS validate before rendering:

bash ~/.claude/tools/mermaid-renderer/validate-diagrams.sh <document-path>

This runs every mermaid block through the mermaid parser and reports OK/FAIL with error details. Do not render or export until all blocks pass.

If a block fails:

1. Read the error message — it shows the line number and unexpected token
2. Common causes: reserved keyword as classDef name, unescaped special characters, malformed YAML frontmatter
3. Fix and re-validate

MANDATORY: Common Errors — Avoid These

These are the most frequently encountered mermaid errors. Each causes silent rendering failure ("Syntax error in text"):

1. Angle brackets in comments parsed as HTML

When mermaid code is embedded in <pre class="mermaid"> for rendering, angle brackets in comments like %% @prefix ex: <http://example.org/> are parsed as HTML tags, destroying the mermaid source before the parser sees it.

The renderer handles this automatically, but be aware:

• <br/>, <b>, <i> in labels are safe (explicitly preserved)
• All other < characters in comments are escaped automatically
• If rendering fails silently, check for unescaped < in comments

2. \n in labels (renders as literal text)

See the MANDATORY: Line Breaks section above.

3. Reserved words as classDef names

See the MANDATORY: Reserved Keywords section above.

4. Unescaped HTML special characters in labels

<, >, & in node labels are interpreted as HTML:

WRONG:  A["Latency <15min"]       → <15min parsed as HTML tag
CORRECT: A["Latency &lt;15min"]   → renders correctly
CORRECT: A["Latency under 15min"] → avoid the issue entirely

5. Unquoted labels with special characters

Labels containing (, ), [, ], {, }, #, &, or : must be quoted:

WRONG:  A[Count: 42]       → colon breaks the parser
CORRECT: A["Count: 42"]    → quoted label is safe

6. Missing diagram type declaration

Every mermaid block must start with a diagram type (or YAML frontmatter followed by a diagram type):

WRONG:  A --> B             → no diagram type
CORRECT: flowchart LR       → diagram type declared
            A --> B

7. Subgraph end keyword conflicts

The word end closes subgraphs. Never use it as a node ID:

WRONG:  end[End State]       → conflicts with subgraph end
CORRECT: endState[End State]  → safe

Quality Checklist

Before returning any diagram:

• Correct diagram type for the use case?
• ELK enabled if >10 nodes or complex relationships?
• Semantic colors applied consistently?
• All nodes labeled clearly (≤30 characters for readability)?
• Line breaks use <br/> not \n? (search the block!)
• No reserved words as classDef names? (class, graph, end, default, etc.)
• Relationships have meaningful labels where needed?
• Configuration block properly formatted?
• No syntax errors — run validation gate?
• Accessibility: accTitle and accDescr included?

Loading Secondary Guides

When you need detailed syntax or patterns beyond this entry point:

1. Match user request to the Decision Logic table above
2. Use the Read tool to load the guide from the same directory as this SKILL.md file
3. Apply patterns from the guide to user's specific need
4. Use Cagle colors from this file (always available)

Load only ONE guide per request to minimize context usage.

Example Interactions

User: "Create a database schema for a blog"

Action: Load 06-ER-DIAGRAM-GUIDE.md, generate erDiagram with USER, POST, COMMENT, TAG entities

User: "Show me how a REST API request flows"

Action: Load 03-SEQUENCE-DIAGRAM-GUIDE.md, generate sequenceDiagram with Client, Gateway, Service, Database

User: "Architecture diagram for a microservices platform"

Action: Load 12-SOLUTION-ARCHITECTURE-GUIDE.md, generate flowchart with ELK, using service/data/infra colors

User: "What diagram should I use for showing order states?"

Action: Recommend State Diagram, load 05-STATE-DIAGRAM-GUIDE.md

User: "I need to show project milestones"

Action: Load 07-GANTT-GUIDE.md, generate gantt with milestones

User: "Visualize this RDF ontology"

Action: Load 17-LINKED-DATA-GUIDE.md, generate flowchart LR with ELK, class/instance/literal colors

User: "Show a Neo4j social network model"

Action: Load 18-PROPERTY-GRAPH-GUIDE.md, generate flowchart with labeled relationships (:KNOWS, :FOLLOWS)

User: "Generate a DOT graph from this Turtle data"

Action: Load 19-DOT-GRAPHVIZ-GUIDE.md, generate DOT digraph with semantic colors

User: "I need a pure network graph visualization"

Action: Load 19-DOT-GRAPHVIZ-GUIDE.md, generate DOT with layout attributes

Default Behaviors

• ELK layout: Enable for >10 nodes or dense connections
• LR direction: Use for semantic/knowledge graphs (S-P-O structure)
• Semantic colors: Apply consistently from Cagle palette above
• Accessibility: Always include accTitle and accDescr
• Namespaces: Document as comments (%% @prefix ex: <...>)
• Ambiguity: Ask user to clarify rather than guess

Post-Creation Export Workflow (MANDATORY)

CRITICAL: After creating or editing ANY mermaid/DOT diagram in a markdown document, you MUST run the export command to render the diagram to an image. This is NOT optional.

Automatic Export Rule

ALWAYS run the export command immediately after:

• Adding a new mermaid or DOT code block to a document
• Modifying an existing diagram in a document
• Replacing ASCII art with a mermaid/DOT diagram

DO NOT leave mermaid code blocks unrendered in documents. The final document must contain:

1. The rendered image (PNG/SVG)
2. The source code in a collapsible <details> block

Export Commands

For Mermaid diagrams (```mermaid blocks):

node ~/.claude/tools/mermaid-renderer/process-document.js <document-path> --verbose

For DOT/Graphviz diagrams (```dot, ```graphviz, ```gv blocks):

node ~/.claude/tools/dot-renderer/process-document.js <document-path> --verbose

Result Structure

document.md
diagrams/
└── document/
    ├── diagram-1.png
    ├── architecture-overview.png
    └── data-flow.png

What Export Does

1. Extracts all diagram code blocks from the document
2. Renders each to PNG in diagrams/{document-name}/
3. Replaces code blocks with image references: ![name](diagrams/doc/name.png)
4. Preserves original code in <details> block for future editing

Post-Export Document Format

<!-- Before -->
` ``mermaid
flowchart LR
    A --> B
` ``

<!-- After (use <img> tag for sizing, NEVER CSS) -->
<img src="diagrams/document/diagram-1.png" alt="Diagram 1" width="70%">

<details>
<summary>Mermaid Source</summary>

` ``mermaid
flowchart LR
    A --> B
` ``

</details>

Re-editing Workflow

1. Find the <details> block containing the mermaid source
2. Copy the mermaid code block above the image reference
3. Edit the diagram
4. Re-run the export command
5. New SVG replaces the old one, <details> block is updated

HTML/PDF Export (After Diagram Rendering)

After rendering diagrams to PNG, export the full document to HTML/PDF:

node ~/.claude/tools/markdown-export/convert.js <document-path> --verbose

What the HTML Export Does

1. Embeds images inline as base64 data URIs for self-contained HTML
2. Renders Mermaid blocks client-side via Mermaid.js with ELK layout
3. Applies max-height (80vh) to images so they fit on screen without scrolling
4. Adds pan/zoom viewer — clicking any image opens an interactive overlay with:
   • Mouse wheel zoom (centered on cursor)
   • Click-drag pan
   • Double-click to toggle fit/1:1
   • Touch: pinch-zoom and drag
   • Keyboard: +/- zoom, 0 fit, 1 actual size, Esc close
   • Control bar with zoom buttons and percentage display
5. Generates PDF via Puppeteer with rendered diagrams
6. Copies diagrams to export/html/diagrams/ for reference

HTML Export Features

HTML Renderer: Angle Bracket Escaping

The mermaid renderer (render-mermaid.js) automatically escapes angle brackets in HTML <pre> context to prevent them being parsed as HTML tags. This handles:

• %% @prefix ex: <http://...> comments — < escaped to <
• Preserves allowed HTML tags in labels: <br/>, <b>, <i>, <u>, <em>, <strong>, <sub>, <sup>
• Without this, mermaid blocks with URL comments silently fail (rendering as 53,854-byte error PNGs)