plantuml（plantuml 命令行）

PlantUML Diagram Generation and Conversion

Purpose

This skill enables PlantUML diagram creation and conversion. Use for: diagram generation from descriptions, code-to-diagram conversion,.puml file conversion, markdown processing, syntax validation, and Unicode symbol enrichment.

Intent Classification & Routing Decision Tree

Step 1: Identify User Intent

IF user mentions:

• "create [diagram type]" OR "generate diagram" OR "draw diagram" → Route to: Resilient Workflow (PRIMARY) - references/workflows/resilient-execution-guide.md → Script: python scripts/resilient_processor.py file.md [--format png] → Backup Route: Diagram Creation Workflow (Step 2) if script unavailable
• "convert.puml" OR "convert file" OR ".puml to image" → Route to script: python scripts/convert_puml.py file.puml [--format png|svg]
• "process markdown" OR "extract diagrams" OR "markdown with puml" → Route to script: python scripts/process_markdown_puml.py file.md [--format png|svg] [--validate]
• "upload to Confluence" OR "upload to Notion" AND markdown contains PlantUML diagrams → ⚠️ CRITICAL: Route to script: python scripts/process_markdown_puml.py file.md [--format png|svg] FIRST → Action: Convert diagrams to images before upload → Reason: Confluence/Notion do not render PlantUML code blocks natively
• "from code" OR "analyze [Spring Boot|FastAPI|ETL|Node.js|React]" → Route to: examples/[framework]/
• "syntax error" OR "fix diagram" OR "error message" OR "not working" → Route to: references/troubleshooting/toc.md
• "style" OR "theme" OR "colors" OR "appearance" → Route to: references/styling_guide.md
• "symbols" OR "icons" OR "emojis" OR "unicode" → Route to: references/unicode_symbols.md

Step 2: Classify Diagram Type (if creating new diagram)

IF user describes:

• Interactions/messages over time: "API call", "message", "participant", "sequence" → Load references/sequence_diagrams.md
• Class structure: "class", "inheritance", "OOP", "extends", "implements" → Load references/class_diagrams.md
• Database schema: "database", "entity", "table", "schema", "ER diagram" → Load references/er_diagrams.md
• Workflow/process: "workflow", "process", "activity", "decision", "fork/join" → Load references/activity_diagrams.md
• State machine: "state", "transition", "state machine" → Load references/state_diagrams.md
• Project timeline: "gantt", "timeline", "schedule", "milestones" → Load references/gantt_diagrams.md
• System architecture: "component", "module", "architecture" → Load references/component_diagrams.md
• Deployment: "deployment", "server", "infrastructure", "cloud" → Load references/deployment_diagrams.md
• Use cases: "actor", "use case", "features" → Load references/use_case_diagrams.md
• Ideas/brainstorm: "mindmap", "brainstorm", "ideas" → Load references/mindmap_diagrams.md

IF diagram type unclear: → Load references/toc.md (156 lines, ~780 tokens) → Present options to user → Route to selected workflow

Step 3: Load Supporting Resources (On-Demand Only)

Load only when explicitly needed:

• Syntax reference: Load references/[diagram_type]_diagrams.md ONLY IF:

- First-time creation for user - Complex/advanced features needed - User requests "how to" or "syntax" - Cost: 1,500-3,000 tokens

• Styling: Load references/styling_guide.md ONLY IF:

- User mentions "style", "color", "theme" - Workflow reaches styling step - Cost: ~1,000 tokens

• Unicode symbols: Load references/unicode_symbols.md ONLY IF:

- User building code-to-diagram - User mentions "symbols" or "icons" - Cost: ~1,250 tokens

• Examples: Load examples/[framework]/ ONLY IF:

- User explicitly requests examples - Workflow references specific example - Cost: ~750 tokens

• Troubleshooting: Load troubleshooting guides ONLY IF:

- Error occurs during generation - User reports syntax issue - Start with: references/troubleshooting/toc.md (~400 tokens) - Error decision tree - Then load specific guide: references/troubleshooting/[category]_guide.md (~500-1,000 tokens) - 12 focused guides covering 215+ common errors - Cost: ~400-1,400 tokens total (decision tree + specific guide)

Token Budget Management

Budget Limits

• Tier 1 (metadata): 100 tokens (loaded)
• Tier 2 (this file): ~2,000 tokens (loaded)
• Tier 3 (per workflow): 1,000-2,000 tokens (load on-demand)
• Target total: <10,000 tokens per typical request
• Maximum: <15,000 tokens for complex requests

Loading Cost Estimates

Budget Scenarios

Simple Request (user familiar with syntax):

• Tier 1 + Tier 2 + Workflow: 100 + 2,000 + 1,500 = 3,600 tokens ✅

Standard Request (need syntax reference):

• Tier 1 + Tier 2 + Workflow + Syntax: 100 + 2,000 + 1,500 + 2,500 = 6,100 tokens ✅

Complex Request (syntax + styling + examples):

• Tier 1 + Tier 2 + Workflow + Syntax + Styling + Examples: 100 + 2,000 + 1,500 + 2,500 + 1,000 + 750 = 7,850 tokens ✅

Advanced Request (all resources):

• Tier 1 + Tier 2 + Multiple workflows + Resources: ~10,000-15,000 tokens ⚠️

Budget-Conscious Strategies

1. Skip syntax reference if user provides clear diagram description
2. Use inline minimal examples instead of loading full example guides
3. Load basic styling instead of comprehensive guide
4. Reference examples directory without loading guide files
5. Load only relevant section of troubleshooting guide (navigate to specific diagram type)

Resource Loading Policy

MANDATORY LOADING RULES

1. Workflow Guides

• When: User intent classified
• Path: references/[type]_diagrams.md
• Cost: ~1,500 tokens
• Rule: Load ONLY the relevant diagram reference, NEVER all references

2. Syntax References

• When: First-time creation OR complex features OR user requests
• Path: references/[diagram_type]_diagrams.md
• Cost: 1,500-3,000 tokens
• Rule: SKIP for simple diagrams with clear user description

3. Styling Resources

• When: User mentions styling OR workflow reaches styling step
• Path: references/styling_guide.md
• Cost: ~1,000 tokens
• Rule: NEVER preload, only on explicit need

4. Examples

• When: User requests examples OR workflow explicitly references
• Path: examples/[framework]/
• Cost: ~750 tokens
• Rule: NEVER preload "just in case"

5. Code-to-Diagram Patterns

• When: User has codebase OR mentions framework
• Path: examples/[framework]/
• Cost: ~500 per example
• Rule: Load only the matching framework examples

6. Troubleshooting

• When: Error occurs OR user reports issue
• Path: references/troubleshooting/toc.md → references/troubleshooting/[category]_guide.md
• Cost: ~500-1,000 tokens
• Rule: Load only when problem identified

7. Unicode Symbols

• When: Code-to-diagram OR user mentions symbols
• Path: references/unicode_symbols.md
• Cost: ~1,250 tokens
• Rule: Load only relevant category section

How to Load Resources

Use Read tool with specific file path:

Read tool: [skill-root]/[path]

NEVER:

• Use Glob to load entire directories
• Load "just in case" resources
• Preload comprehensive guides when focused guide suffices
• Load multiple diagram type guides simultaneously

Setup Verification

Before creating diagrams, verify setup:

python scripts/check_setup.py

Checks: Java, Graphviz, plantuml.jar

If missing components → Load references/plantuml_reference.md (Installation section)

Core Workflows (Routing Only)

1. Diagram Creation

Trigger: User requests new diagram Route: Classify type → Load references/[type]_diagrams.md Supporting: Syntax reference (if needed), styling (if requested)

2. Standalone Conversion

Trigger: Convert.puml file(s) to images Route: Direct script command Script: python scripts/convert_puml.py file.puml [--format png|svg]

3. Markdown Processing

Trigger: Markdown with PlantUML blocks or linked.puml files ⚠️ CRITICAL Trigger: User uploading markdown to Confluence or Notion with PlantUML diagrams Route: Direct script command Script: python scripts/process_markdown_puml.py file.md [--format png] Output: file_with_images.md + images/ directory Note: Confluence/Notion require diagram conversion to images before upload

4. Code-to-Diagram

Trigger: Analyze codebase and generate diagrams Route: examples/[framework]/ Supporting: Framework examples from examples/[framework]/

5. Troubleshooting

Trigger: Syntax error or generation failure Route:

1. Load references/troubleshooting/toc.md - Error decision tree
2. Identify error category
3. Load specific guide: references/troubleshooting/[category]_guide.md Resources: 12 guides covering 215+ common errors (installation, syntax, arrows, styling, diagrams, performance)

6. Styling

Trigger: User wants to customize appearance Route: references/styling_guide.md Supporting: references/styling_guide.md

Default Visual Policy (sequence-first readability):

1. Use !theme plain as neutral baseline
2. For sequence diagrams, group participants by architecture layer using box "..." #Color
3. Use soft background colors for layer boxes; keep participant/arrow styling mostly default
4. Add <style> only for focused readability improvements, not full recoloring

7. Resilient Workflow (PRIMARY)

Trigger: ALL diagram creation requests (recommended default) Route: references/workflows/resilient-execution-guide.md Script: python scripts/resilient_processor.py file.md [--format png|svg] Steps:

1. Identify type → Load reference
2. Create file → Structured naming: ./diagrams/<md>_<num>_<type>_<title>.puml
3. Convert → Handle errors with troubleshooting guides (max 3 retries)
4. Validate → Integrate into markdown Error Fallback: Perplexity → Brave Search → Gemini → WebSearch Benefits:

• Consistent file naming for organization
• Automatic error recovery with troubleshooting guides
• External search fallback for unresolved errors
• Validation before markdown integration

Quick Reference (Minimal)

Universal Structure:

@start[type]
' Your content
@end[type]

Common Arrows: -> solid, --> dashed, ..> dotted

Scripts:

• Resilient (PRIMARY): python scripts/resilient_processor.py file.md [--format png]
• Convert: python scripts/convert_puml.py file.puml [--format svg]
• Markdown: python scripts/process_markdown_puml.py doc.md [--validate]
• Setup check: python scripts/check_setup.py

For detailed syntax → Load references/[type]_diagrams.md

Sequence baseline with grouped layers (recommended):

@startuml
!theme plain

box "Client" #LightCyan
  actor User
end box

box "Application" #LightYellow
  participant App
  participant API
end box

User -> App : Login
App -> API : Authenticate
API --> App : Token
App --> User : Success
@enduml

Error Handling

Missing plantuml.jar

1. Check: python scripts/check_setup.py
2. Load: references/plantuml_reference.md (Installation section)

Syntax Error

1. Load: references/troubleshooting/toc.md (~400 tokens)
2. Use error decision tree to identify category
3. Load specific guide: references/troubleshooting/[category]_guide.md (~500-1,000 tokens)
4. Apply recommended fix from guide

Comprehensive troubleshooting coverage (215+ errors in 12 guides):

• Setup: Installation, Java, Graphviz, plantuml.jar (15 errors)
• Syntax: Delimiters, comments, basic structure (20 errors)
• Arrows: Relationship syntax across all diagram types (20 errors)
• Text: Quotes, special characters, encoding (20 errors)
• Styling: skinparam, style blocks, colors, fonts (20 errors) — If skinparam errors appear, migrate to modern <style> blocks.
• Preprocessor:!include,!define,!procedure (20 errors)
• Sequence diagrams: Participants, arrows, fragments (20 errors)
• Class diagrams: Relationships, visibility, generics (20 errors)
• ER diagrams: Entities, cardinality, keys (20 errors)
• Activity diagrams: Flow control, forks, partitions (20 errors)
• Image generation: Rendering, output formats (20 errors)
• Performance: Timeouts, memory, optimization (20 errors)

Unknown Diagram Type

1. Load: references/toc.md (~780 tokens)
2. Present options
3. Route to selected workflow

Setup Issues

1. Load: references/plantuml_reference.md (Installation section)
2. Follow setup instructions

Skill Integration Points

Invoke other skills when:

• Complex code analysis → Use general-purpose agent
• Documentation generation → Use docs-sync-editor after diagram creation
• Grammar in labels → Use grammar-style-editor
• Architecture analysis → Use code-explainer before code-to-diagram

Diagram Type Classification Keywords

Quick routing keywords:

• Sequence: "interaction", "API call", "message flow", "participant"
• Class: "class diagram", "inheritance", "OOP", "extends"
• ER: "database", "entity", "schema", "table", "relation"
• Activity: "workflow", "process", "decision", "activity"
• State: "state machine", "transition", "state"
• Component: "component", "module", "architecture"
• Deployment: "deployment", "infrastructure", "server"
• Gantt: "timeline", "schedule", "project", "gantt"
• Use Case: "actor", "use case", "features"
• MindMap: "mindmap", "brainstorm", "ideas"

For all 19 types → Load references/toc.md

Token Budget Tracking

Current session tracking:

• Report when approaching 10,000 tokens
• Suggest lighter alternatives if budget tight
• Document loaded resources

Budget escalation thresholds:

• <5,000 tokens: Excellent ✅
• 5,000-10,000 tokens: Normal ✅
• 10,000-15,000 tokens: Review necessity ⚠️
• 15,000 tokens: Minimize loading ❌

References (Load On-Demand)

Core Syntax (load when needed):

• references/toc.md - All diagram types navigation
• references/[type]_diagrams.md - Specific diagram syntax
• references/common_format.md - Universal elements
• references/styling_guide.md - CSS-like styling
• references/plantuml_reference.md - CLI and troubleshooting

Troubleshooting Resources (⚠️ load when errors occur):

• references/troubleshooting/toc.md - Error decision tree and navigation hub
• references/troubleshooting/*.md - 12 focused guides (215+ common errors)

- Installation, syntax, arrows, text, styling, preprocessor - Sequence, class, ER, activity diagrams - Image generation, performance

Other Resources (load when needed):

• references/common_syntax_errors.md - Legacy comprehensive guide
• references/unicode_symbols.md - Symbol enrichment
• examples/[framework]/ - Code-to-diagram patterns

Summary: Using This Skill Effectively

Decision Tree Process:

1. Classify intent using Step 1 routing
2. Load diagram references or run target script for identified task
3. Load supporting resources only as workflow requires
4. Track token budget throughout process
5. Use Read tool for surgical loading

Never:

• Load entire directories
• Preload "just in case"
• Load all diagram type guides
• Exceed budget without necessity

This PDA architecture:

• Reduces typical requests: 24-40% fewer tokens
• Reduces complex requests: 60-70% fewer tokens
• Loads only relevant content
• Maintains quality and completeness