Pipeline

How it Works

From SKILL.md to deterministic execution in 4 simple steps

1
1
SKILL.md
Human-authored
2
2
OntoCore
The Compiler
3
3
.ttl Files
The Ontology
4
4
OntoMCP
The Runtime
1

SKILL.md

Human-authored

Skills are authored in Markdown with YAML frontmatter. Developers write clear instructions, define intents, and structure content that AI agents can understand and execute.

  • YAML frontmatter for metadata
  • Markdown body for instructions
  • Version-controlled with your code
  • Human-readable, AI-friendly
SKILL.md 
---
name: tdd-workflow
description: Test-driven development
intents:
  - write_tests_first
---

## Instructions
1. Write the test first
2. Run the test (it fails)
3. Write minimal code
4. Run the test (it passes)
5. Refactor
2

OntoCore

The Compiler

OntoCore compiles SKILL.md files into OWL 2 DL ontologies. The compiler validates structure, extracts semantic relationships, and generates machine-readable RDF triples.

  • Parses and validates SKILL.md
  • Extracts semantic relationships
  • Generates OWL 2 DL ontology
  • SHACL validation built-in
terminal 
$ npx ontoskills compile

  Parsing tdd-workflow/SKILL.md...
  Validating structure...
  Generating ontology...
  Writing triples...

  ✓ Compiled to ontoskills/tdd-workflow.ttl
3

.ttl Files

The Ontology

The compiled Turtle (.ttl) files contain the ontology — a formal representation of skills, intents, and their relationships using RDF triples and OWL semantics.

  • Standard RDF Turtle format
  • OWL 2 DL definitions
  • SPARQL-queryable graph
  • Semantic skill relationships
output.ttl 
@prefix oc: <https://ontoskills.dev/ns#> .

oc:tdd-workflow a oc:Skill ;
  oc:name "tdd-workflow" ;
  oc:description "Test-driven dev" ;
  oc:resolvesIntent "write_tests_first" ;
  oc:hasStep oc:write-test,
             oc:run-test,
             oc:implement .
4

OntoMCP

The Runtime

OntoMCP is the Rust-based MCP runtime that loads ontologies and serves skill queries to AI agents through the Model Context Protocol with sub-millisecond response times.

  • Native Rust performance
  • MCP protocol implementation
  • Sub-millisecond SPARQL
  • Intent-based discovery
mcp-query 
# Search by intent
search_intents("create_pdf")
  → pdf-generator
  → report-exporter

# Get skill context
get_skill("pdf-generator")
  → { name, description, instructions }

Ready to compile your first skill?

Get started with OntoSkills in minutes. Write your first SKILL.md, compile it to an ontology, and query it through MCP.

Get Started Browse OntoStore