Skills System

​

Overview

​

The Five Sub-Skills

​

1. PRD Parser (prd-parse)

​

Purpose

​

Inputs

​

Outputs

​

Key Features

• Tokenization by heading levels (H1, H2, H3)
• Semantic classification (feature, story, requirement, entity, flow)
• Named Entity Recognition (NER) for domain identification
• Relationship mapping between entities
• Dependency graph calculation between features
• Completeness scoring with warnings if score < 0.6

const prdParse = require('prd-parse');

const result = await prdParse.run({
  prd_content: fs.readFileSync('./my-prd.md', 'utf-8')
});

console.log(`Completeness: ${result.completeness_score}`);
console.log(`Features found: ${result.parsed_prd.features.length}`);
console.log(`Entities found: ${result.parsed_prd.entities.length}`);

​

2. Mermaid Generator (mermaid-gen)

​

Purpose

​

Inputs

​

Outputs

​

Key Features

• Smart PRD element → diagram type mapping
• Syntax validation before emission (parser check)
• Multi-language label support
• Automatic diagram splitting if >50 nodes
• Feature/story ID references in Mermaid comments
• Coverage percentage calculation

​

Diagram Type Mapping

const mermaidGen = require('mermaid-gen');

const result = await mermaidGen.run({
  parsed_prd: parsedPrdObject,
  diagram_types: ['flowchart', 'sequence', 'erDiagram', 'stateDiagram'],
  locale: 'en-US'
});

result.diagrams.forEach(diagram => {
  console.log(`${diagram.type}: ${diagram.coverage_pct}% coverage`);
  fs.writeFileSync(`${diagram.type}.mmd`, diagram.code);
});

​

3. Logic Validator (logic-validate)

​

Purpose

​

Inputs

​

Outputs

​

Validation Criteria

​

Validation Modes

• interactive: Present each diagram with score, await user decision per diagram
• batch: Present all diagrams with consolidated report, await bulk decision
• auto: Auto-approve if score >= validation_threshold, reject otherwise

const logicValidate = require('logic-validate');

const result = await logicValidate.run({
  parsed_prd: parsedPrdObject,
  diagrams: diagramsArray,
  validation_mode: 'auto',
  validation_threshold: 0.85
});

if (result.validation_report.status === 'approved') {
  console.log(`Validation passed with score: ${result.validation_report.overall_score}`);
} else {
  console.log('Validation failed. Suggestions:');
  result.validation_report.suggestions.forEach(s => console.log(`- ${s}`));
}

​

4. Figma Generator (figma-gen)

​

Purpose

​

Inputs

​

Outputs

​

Key Features

• Figma REST API v1 integration
• Auto-layout frame creation
• Design token application (colors, typography, spacing)
• Responsive variant generation (mobile, tablet, desktop)
• Development annotation embedding
• Component library generation
• Rate limiting with exponential backoff
• Component caching to minimize API calls

const figmaGen = require('figma-gen');

const result = await figmaGen.run({
  diagrams: validatedDiagrams,
  figma_file_key: 'abc123XYZ',
  figma_access_token: process.env.FIGMA_TOKEN,
  design_system: 'material-3',
  project_name: 'My SaaS App'
});

result.figma_assets.forEach(asset => {
  console.log(`${asset.type}: ${asset.figma_url}`);
});

​

5. Asset Delivery (asset-deliver)

​

Purpose

​

Inputs

​

Outputs

​

Generated Artifacts

​

Key Features

• Token and secret sanitization in logs
• Sensitive information redaction ([REDACTED])
• Complete orchestration timeline
• Developer handoff documentation generation
• Metrics aggregation (timing, quality scores, coverage)

const assetDeliver = require('asset-deliver');

const result = await assetDeliver.run({
  parsed_prd: parsedPrdObject,
  diagrams: diagramsArray,
  validation_report: validationReportObject,
  figma_assets: figmaAssetsArray
});

console.log('Delivery package created:');
console.log(`- Diagrams: ${result.delivery_package.diagrams_count}`);
console.log(`- Figma assets: ${result.delivery_package.figma_assets_count}`);
console.log(`- Package location: ${result.delivery_package.output_path}`);

​

Orchestration Flow

1. Sequential execution of sub-skills in order
2. Data passing between phases via structured outputs
3. Error handling and retry logic per phase
4. Feedback loops when validation fails
5. State management for resumable execution
6. Logging and metrics across all phases

​

Dependency Management

​

Benefits of Modular Architecture

​

Reusability

• Use prd-parse alone to analyze PRD quality
• Use mermaid-gen to generate diagrams from any structured data
• Use logic-validate to validate any Mermaid diagrams
• Use figma-gen to generate Figma assets from any diagram source

​

Testability

• Faster test execution
• Easier debugging
• Clear failure attribution

​

Independent Development

​

Flexibility

• Swap figma-gen for a sketch-gen implementation
• Add a new validation criterion to logic-validate
• Extend mermaid-gen to support new diagram types

​

Configuration Inheritance

# .omni-architect.yml
locale: "pt-BR"
validation_threshold: 0.85

# Override for specific sub-skill
skills:
  mermaid-gen:
    locale: "en-US"  # Override parent locale
  logic-validate:
    validation_threshold: 0.90  # Stricter threshold

​

Next Steps