Skip to content

README.md

Latest commit

 

History

History
237 lines (179 loc) · 8.24 KB

README.md

File metadata and controls

237 lines (179 loc) · 8.24 KB
Raw

OntoStore

The official skill registry for OntoSkills.

Browse, install, and publish validated OWL 2 skill packages for deterministic AI agent behavior.

Documentation • Browse Skills • Publish

16 Packages 409 Skills OWL 2 BM25 + Semantic Search MIT License

What is OntoStore?

OntoStore is the versioned skill registry at the heart of the OntoSkills ecosystem. It hosts pre-compiled OWL 2 ontology packages that AI agents query via SPARQL for deterministic, verifiable behavior.

Every package contains .ttl artifacts validated against the OntoSkills SHACL constitution — no probabilistic interpretation, no ambiguity.

ontoskills install obra/superpowers          # Install a package
ontoskills install obra/superpowers --with-embeddings  # With semantic search files
ontoskills install anthropics                # Install all from an author

Available Packages

Author Package Skills Category
anthropics claude-office-skills 137 Office & productivity
anthropics knowledge-work-plugins 113 Knowledge management
anthropics financial-services-plugin 27 Finance & compliance
anthropics claude-plugin-official 19 Plugin development
anthropics claude-code 10 Code assistance
coreyhaines31 marketingskills 34 Marketing
pbakaus impeccable 21 Quality & testing
coinbase agentic-wallet-skills 9 Blockchain & wallets
n8n-io n8n 10 Workflow automation
obra superpowers 14 Developer tooling
nextlevel-builder ui-ux-pro-max-skill 7 UI/UX design
kotlin kotlin-agent-skills 4 Kotlin development
remotion-dev skills 1 Video creation
vercel-labs vercel-react-best-practices 1 React development
vercel-labs agent-browser 1 Browser automation
mareasw greeting 1 Demo / hello world

Trust Tiers

Every package declares a trust tier that affects search ranking and visibility:

Tier Multiplier Description
official 1.2x Maintained by the OntoSkills team
verified 1.0x Reviewed and validated from known vendors
community 0.8x Community-contributed, use at your own discretion

Skill Discovery

Installed skills are discoverable by the OntoMCP server through two search engines:

BM25 Keyword Search (default, always available)

Every installed skill is searchable via BM25 keyword matching — no extra dependencies, no model downloads. The BM25 index is built in-memory at server startup from skill intents, aliases, and descriptions.

  • Zero setup — works out of the box after ontoskills install
  • English stemming and stop words
  • Results ranked by keyword relevance x trust-tier quality multiplier

Semantic Search (optional)

For large skill catalogs where keyword matching may miss relevant results, OntoStore provides pre-computed embeddings. Install them with --with-embeddings:

ontoskills install obra/superpowers --with-embeddings

When available, the MCP server uses ONNX inference for semantic similarity matching. BM25 remains the default; semantic search activates only when embeddings are installed.

Registry Structure

ontostore/
├── index.json                         # Registry index (packages list + embedding model)
├── embeddings/
│   └── tokenizer.json                 # Shared tokenizer (for semantic search)
├── packages/
│   ├── anthropics/
│   │   ├── claude-code/
│   │   │   ├── package.json           # Package manifest
│   │   │   └── <skill>/
│   │   │       ├── ontoskill.ttl      # Compiled OWL 2 ontology
│   │   │       └── intents.json       # Pre-computed embeddings (optional)
│   │   └── ...
│   ├── obra/
│   │   └── superpowers/
│   │       └── ...
│   └── ...
└── README.md

index.json

The registry index lists all available packages and the embedding model metadata:

{
  "embedding_model": {
    "model_name": "sentence-transformers/all-MiniLM-L6-v2",
    "dimension": 384,
    "model_file": "model.onnx",
    "tokenizer_file": "tokenizer.json"
  },
  "packages": [
    {
      "package_id": "obra/superpowers",
      "trust_tier": "verified",
      "source_kind": "ontology",
      "manifest_path": "packages/obra/superpowers/package.json"
    }
  ]
}

package.json (per-package manifest)

Each package has a manifest declaring its skills, modules, and metadata:

{
  "package_id": "obra/superpowers",
  "version": "0.11.0",
  "trust_tier": "verified",
  "source_kind": "ontology",
  "modules": ["tdd/ontoskill.ttl", "brainstorming/ontoskill.ttl"],
  "skills": [
    {
      "id": "tdd",
      "path": "tdd/ontoskill.ttl",
      "default_enabled": true,
      "aliases": ["test-driven-development"],
      "intents": ["write tests before code", "practice TDD"]
    }
  ],
  "embedding_files": ["tdd/intents.json", "brainstorming/intents.json"]
}

Installing Packages

From the CLI

# Install a single package
ontoskills install obra/superpowers

# Install with semantic search embeddings (optional)
ontoskills install obra/superpowers --with-embeddings

# Install all packages from an author
ontoskills install anthropics

# Enable/disable specific skills
ontoskills enable obra/superpowers tdd brainstorming
ontoskills disable obra/superpowers tdd

Adding a Registry Source

ontoskills registry add-source official https://ontoskills.sh/ontostore/index.json

Publishing a Package

1. Compile your skills

ontocore compile -i skills/ -o ontoskills/

2. Create a package.json

Place it in your skill directory alongside the compiled .ttl files. See the package spec for the full schema.

3. Submit a PR

Fork this repository, add your package under packages/<author>/<package>/, update index.json, and open a pull request. Packages are validated against the OntoSkills SHACL constitution before merge.

Requirements:

  • All .ttl files must pass SHACL validation
  • package.json must conform to the package manifest schema
  • Skill IDs must be unique across the registry
  • Trust tier will be set to community by default; verified requires review

Related

© 2026 MareaSW