schema（schema 搜索）

Schema: Knowledge Ontology Generator

Transform any input into structured knowledge schemas with rich semantic metadata.

Overview

Generate ontologies from:

• Plain text and natural language
• Structured data (JSON, YAML, CSV, XML)
• Markdown files with metadata
• Code repositories

Output formats:

• Primary: Obsidian markdown (YAML frontmatter, wikilinks, tags, callouts, mermaid)
• Secondary: JSON-LD, RDF/OWL, Neo4j Cypher, GraphQL schemas

Modes

Fractal Mode

Strict hierarchical constraints for self-similar structures:

• 2-3 children per non-leaf node
• Homonymic inheritance (child labels contain parent stem)
• Uniform relation types per parent
• Topology score ≥4.0

Free Mode

Flexible generation optimized for semantic coherence:

• Variable branching factor
• Relaxed naming constraints
• Focus on meaningful relationships

Installation

Quick Install (recommended):

cd ~/.claude/skills/hm-skills/schema
./install.sh

This will:

• Install Python dependencies (jinja2, pyyaml)
• Download spaCy language model (en_core_web_sm)
• Create global schema-gen wrapper in ~/bin
• Configure PATH in your shell rc file

Manual Setup:

pip3 install --user jinja2 pyyaml
python3 -m spacy download en_core_web_sm

Architecture

Four-layer pipeline with graceful degradation:

1. Layer 1: Structural Extraction - AST parsing (markdown-oxide, tree-sitter, pandas)
2. Layer 2: Semantic Analysis - NLP (spaCy, networkx)
3. Layer 3: LLM Enrichment - Optional deep analysis (Claude API, MCP tools)
4. Layer 4: Output Generation - Template-based formatting (Jinja2)

Usage

Recommended: Use the global wrapper schema-gen (automatically installed to ~/bin/schema-gen)

# Basic usage - text to Obsidian markdown
schema-gen --input "text content" --output schema.md

# Fractal mode with deep analysis
schema-gen --input data.json --mode fractal --deep

# Export to multiple formats
schema-gen --input code/ --format obsidian,jsonld,cypher

Alternative: Direct invocation from skill directory

cd /Users/mikhail/.claude/skills/hm-skills/schema
python3 -m scripts.schema_cli --input "text content" --output schema.md

Usage Examples

Input Formats

Text to Schema

# Simple text input
schema-gen --input "AI analyzes data" --output ai-schema.md

# From text file
schema-gen --input document.txt --output schema.md --mode fractal

JSON to Schema

# Structured data
schema-gen --input data.json --output schema.md

Markdown to Schema

# Extract hierarchy from markdown headings
schema-gen --input notes.md --output schema.md --verbose

Code to Schema

# Parse Python code structure (classes, functions, methods)
schema-gen --input mymodule.py --output code-schema.md --verbose

Output Formats

Single Format

# Obsidian markdown (default)
schema-gen --input data.json --output schema.md

Multiple Formats

# Export to all formats
schema-gen \
  --input document.txt \
  --format obsidian,jsonld,cypher,graphql \
  --output output/schema.md \
  --verbose

# Generates:
# - output/schema.md (Obsidian markdown)
# - output/schema.jsonld (JSON-LD linked data)
# - output/schema.cypher (Neo4j graph database)
# - output/schema.graphql (GraphQL schema)

Advanced Usage

Fractal Mode with Deep Analysis

schema-gen \
  --input complex.txt \
  --mode fractal \
  --deep \
  --verbose

Custom Output Directory

schema-gen \
  --input data.json \
  --output custom/path/schema.md \
  --format obsidian,cypher

Output Example

Generated Obsidian markdown includes:

---
created: 2025-01-05T12:00:00
tags: [knowledge, schema]
ontology_type: free
node_count: 10
edge_count: 15
topology_score: 2.50
---

# Schema Title

> [!info] Schema Overview
> Description of the schema

## Structure

graph TD root["Root"] --> child1["Child 1"] root --> child2["Child 2"] class root,child1,child2 internal-link

## Entities

### Root

**Properties:**
- `category`: example
- `weight`: 0.8

**Relationships:**
- parent of → [[Child 1]]
- parent of → [[Child 2]]

Configuration

Edit mode configuration files:

• [[config/fractal-mode.yaml]] - Strict hierarchical constraints
• [[config/free-mode.yaml]] - Flexible generation settings

Customize output templates:

• [[config/templates/obsidian.md.j2]] - Obsidian markdown
• [[config/templates/jsonld.json.j2]] - JSON-LD
• [[config/templates/cypher.cypher.j2]] - Neo4j Cypher

Features

Property Inheritance

Child nodes automatically inherit properties from ancestors, following breadcrumb-plugin patterns.

Multi-Dimensional Navigation

Generate alternate navigation paths:

• Temporal (creation order, lifecycle)
• Conceptual (domain hierarchies, abstraction)
• Spatial (containment, proximity)
• Functional (purpose-based, process flows)

Implicit Relationship Inference

Automatically detect relationships based on:

• Co-occurrence in context
• Tag overlap (>50% shared)
• Structural proximity
• Semantic similarity

Error Handling

Graceful degradation ensures output even on failures:

• Layer 1 fails → Plain text fallback
• Layer 2 fails → Heuristic relationships
• Layer 3 fails → Skip enrichment
• Template fails → Raw JSON output

References

• [[references/ast-parsing-guide|AST Parsing Guide]]
• [[references/semantic-patterns|Semantic Pattern Reference]]
• [[references/template-syntax|Template Variable Reference]]
• [[references/mcp-integration|MCP Tool Integration]]

Examples

• [[examples/text-to-schema|Text to Schema]]
• [[examples/json-to-schema|JSON to Schema]]
• [[examples/markdown-to-schema|Markdown to Schema]]
• [[examples/code-to-schema|Code to Schema]]