OntoSkills

The deterministic skill platform for AI agents.

OntoCore • OntoMCP • OntoSkills Registry • ontoskills CLI

Ecosystem • Use Cases • Installation • CLI • Registry • MCP Server • Documentation

---

Ecosystem

OntoSkills is a neuro-symbolic platform that turns SKILL.md sources into queryable OWL 2 ontologies, ships a local MCP runtime, and distributes compiled skills through an official registry.

flowchart LR
    CORE["OntoCore<br/>SKILL.md -> ontoskill.ttl"] --> COMPILED["OntoSkills<br/>compiled ontologies"]
    COMPILED --> MCP["OntoMCP<br/>local MCP runtime"]
    REG["OntoSkills Registry<br/>official compiled registry"] --> CLI["ontoskills<br/>CLI"]
    CLI --> MCP
    CLI --> COMPILED
    CLI --> CORE

Loading

What Ships

Component	Purpose
ontoskills	User-facing CLI for installs, updates, registry operations, source imports, and managed local state
ontocore	Compiler for turning SKILL.md sources into validated ontologies
ontomcp	Local MCP server for semantic skill discovery, context retrieval, and planning
OntoSkills Registry	Official compiled skill registry, built in by default

---

Use Cases

Use Case	How OntoSkills Helps
Enterprise AI agents	Deterministic skill discovery via ontology queries instead of file reading
Edge and small-model deployments	Query only the relevant skill context instead of loading large prompt bundles
Multi-agent systems	Shared ontology as capability graph, planning graph, and epistemic layer
Auditable automation	Provenance, skill metadata, and explicit enable/disable state
Skill distribution	Published compiled skills can be installed remotely from the official registry

---

Why OntoSkills

The problem

Raw skill files are hard to distribute, expensive to load, and ambiguous to query. As the number of skills grows, agents either read too much context or make poor decisions from incomplete context.

The approach

OntoSkills turns skills into OWL 2 knowledge artifacts and serves them through a narrow runtime interface:

• OntoCore compiles source skills
• OntoSkills Registry distributes compiled skills
• ontoskills manages install/update/enable/disable flows
• OntoMCP exposes the enabled ontology set to MCP clients

The result

• deterministic retrieval instead of fuzzy file scanning
• package-aware provenance and versioning
• smaller MCP tool surface
• installable skill ecosystems instead of repo-only artifacts

---

Installation

Runtime-only install

npx ontoskills install mcp
npx ontoskills search hello
npx ontoskills install marea.greeting/hello
npx ontoskills enable marea.greeting/hello

Install the compiler too

npx ontoskills install core
ontoskills init-core
ontoskills compile

Everything is managed under:

~/.ontoskills/
  bin/
  core/
  ontoskills/
  skills/
  state/

---

CLI Commands

Product commands

ontoskills install mcp
ontoskills install core
ontoskills update mcp
ontoskills update core
ontoskills search hello
ontoskills install marea.greeting/hello
ontoskills enable marea.greeting/hello
ontoskills disable marea.greeting/hello
ontoskills remove marea.greeting/hello
ontoskills rebuild-index
ontoskills uninstall --all

Compiler-oriented commands

ontoskills init-core
ontoskills compile
ontoskills compile my-skill
ontoskills query "SELECT ?s WHERE { ?s a oc:Skill }"

Source import

ontoskills import-source-repo https://github.com/nextlevelbuilder/ui-ux-pro-max-skill

This clones the source repo into ~/.ontoskills/skills/vendor/ and compiles the results into ~/.ontoskills/ontoskills/vendor/.

---

Registry And Packages

Official registry

The official registry is built into the CLI and does not need manual setup.

Current demo package:

• marea.greeting/hello

Registry repository:

• ontoskills-registry

Third-party registries

Third-party registries are opt-in:

ontoskills registry add-source acme https://example.com/index.json
ontoskills registry list

Enable/disable model

• install puts a compiled skill in the local store
• enable exposes it to OntoMCP
• disable removes it from the enabled runtime index

---

Local MCP Server

OntoMCP is the runtime that loads enabled ontologies and exposes them through MCP.

The current MCP tool surface is:

• search_skills
• get_skill_context
• evaluate_execution_plan
• query_epistemic_rules

For local development from the repository:

cargo run --manifest-path mcp/Cargo.toml -- --ontology-root ./ontoskills

For product install:

ontoskills install mcp

More in mcp/README.md.

---

Repository Layout

core/        compiler
mcp/         MCP runtime
docs/        long-form documentation
registry/    official registry blueprint
site/        public site
skills/      source skills
ontoskills/  compiled ontology artifacts
specs/       SHACL and ontology constraints

---

Documentation

• Overview
• Getting Started
• Architecture
• Knowledge Extraction
• Registry
• MCP Runtime
• MCP With Claude Code
• MCP With Codex
• Roadmap
• MCP Guide
• Claude Code MCP Guide

---

Related Links

• Official Registry Repo
• Project Site