SKOS Distillation Skill

A Claude Code skill that converts document collections into structured, navigable knowledge bases using SKOS semantic relationships.

What is SKOS?

SKOS (Simple Knowledge Organization System) is a W3C standard for representing knowledge organization systems such as thesauri, taxonomies, and classification schemes. It provides a common data model for sharing and linking knowledge using three core relationship types:

• Broader — a concept is more general (e.g., "Animal" is broader than "Dog")
• Narrower — a concept is more specific (e.g., "Dog" is narrower than "Animal")
• Related — concepts are associated but not hierarchically (e.g., "Dog" is related to "Pet Care")

SKOS builds on RDF (Resource Description Framework) and is widely used in libraries, government data portals, and enterprise knowledge management to create machine-readable concept hierarchies.

What Does This Skill Do?

Instead of traditional RAG (chunking + vector retrieval), this skill distills a corpus of 3+ documents into canonical concept summaries organized by a SKOS taxonomy. The result is a standalone knowledge skill that Claude can navigate with minimal context overhead (~2-4K tokens per query vs. 4-16K for RAG).

The 4-phase pipeline:

Source Documents
  → Phase 1: Concept Extraction (extraction-log.json)
  → Phase 2: Taxonomy Construction (taxonomy.json)
  → Phase 3: Concept Distillation (concepts/*.md)
  → Phase 4: Skill Assembly (output knowledge skill)

What it produces:

File	Purpose
taxonomy.json	Flat concept map with SKOS broader/narrower/related links
concepts/*.md	~600-word summary per concept with YAML frontmatter
concept-map.md	Navigable tree view of the full taxonomy
SKILL.md	The generated knowledge skill for future Claude sessions

Each phase has a quality gate (Python validation script) that must pass before proceeding.

Prerequisites

• Python 3.10+ (no pip dependencies required)
• 3+ source documents in markdown or text format on a single domain
• Claude Code CLI installed

Installation

This is a Claude Code skill distributed as a plugin. You need the Claude Code CLI installed to use it.

From GitHub

# Install for your user (available in all projects)
claude install seriouscoderone/skos-skill

# Or install for a specific project only
claude install seriouscoderone/skos-skill --scope project

From inside Claude Code

If you're already in a Claude Code session, use the /plugin slash command:

# Add this repo as a marketplace source
/plugin marketplace add seriouscoderone/skos-skill

# Install the plugin
/plugin install skos-distillation-skill@skos-skill

Or use the interactive plugin browser:

/plugin

This opens a UI where you can browse the Discover tab, find the skill, and install it.

From a local clone

git clone https://github.com/seriouscoderone/skos-skill.git
claude install ./skos-skill

Verify installation

After installing, check that the skill is available:

/skos-distillation-skill

The skill should appear in the / slash command menu.

Usage

Once installed, start a Claude Code session and invoke the skill:

/skos-distillation-skill

Or describe what you want and Claude will recognize when to use it:

"I have a folder of documents about Kubernetes networking. Can you distill them into a knowledge base?"

Claude will guide you through each phase interactively, running validation gates between steps.

Project Structure

skos-skill/
├── .claude-plugin/
│   └── plugin.json                           # Plugin manifest for claude install
├── skills/
│   └── skos-distillation-skill/
│       ├── SKILL.md                          # Skill definition and workflow
│       ├── references/
│       │   ├── concept-template.md           # Template for concept files
│       │   ├── extraction-schema.json        # JSON Schema for Phase 1 output
│       │   ├── taxonomy-schema.json          # JSON Schema for Phase 2 output
│       │   ├── output-skill-template.md      # Template for generated SKILL.md
│       │   ├── skos-reference.md             # SKOS vocabulary quick reference
│       │   └── serialization-formats.md      # SKOS serialization formats guide
│       └── scripts/
│           ├── validate_extraction.py        # Gate 1 validator
│           ├── validate_taxonomy.py          # Gate 2 validator
│           ├── validate_concepts.py          # Gate 3 validator
│           └── generate_concept_map.py       # Concept map generator
└── README.md

When to Use

• You have 3+ documents on a domain you want to make queryable
• You want structured knowledge navigation, not keyword search
• You need consistent, deduplicated answers across a document corpus
• You want a reusable knowledge skill for future Claude sessions

When NOT to Use

• Single document summarization (just read the doc)
• Documents that change frequently (produces a static snapshot)
• Fewer than 3 source documents (not enough for a useful taxonomy)