kg-extract-concepts

name	kg-extract-concepts
skill_type	skill
description	Knowledge Graph concept extraction atomic skill. Parses text and structures definitions and relations for Graph-OS.
domain	research
license	MIT
tags	["graph-os","semantic","parsing","concept","research"]
tier	core
wraps	["concept_registry"]
metadata	{"author":"Genius","version":"0.1.0"}
requires	["graph_write"]

Concept Node Extractor Skill

Condensed intent-surface note (Seam 8). Under the default intent surface (MCP_TOOL_MODE=intent), concept_registry is held back from the default tool list (nothing removed — REST + _execute_tool still reach it exactly as documented below). Two ways to use this skill unchanged: (1) load_tools(tools=["concept_registry"]) once per session (as below), then proceed exactly as documented; or (2) call the find intent verb with the same natural-language request — the resolver routes to concept_registry for you and returns the result plus a routing justification. Set MCP_TOOL_MODE=condensed/verbose/both to expose the granular tools eagerly instead.

Stateless atomic operation to ingest unstructured literature reviews, academic transcripts, or enterprise systems specifications, parse out core terms and relationships, map them to ontological definitions, and compile transaction-safe Cypher queries for ingestion into the Graph-OS Knowledge Graph.

Prerequisites

• graph_write — for transactional database writes and node/edge merging within the unified Graph-OS ecosystem.

Steps

Step 1: parse_unstructured_input

Analyze unstructured input text to locate relevant architectural or scientific domain concepts:

• Scan raw text inputs:
  • Process textual payloads (e.g., transcripts, academic papers, system design documents).
  • Use regex patterns, keyword match dictionaries, or natural language semantic parsing to identify candidates for new concepts, ontology items, systems definitions, or acronyms.
• Extract descriptive attributes:
  • Extract the term name, associated definition snippets, parent domain category, context references, and confidence rating.
• Output parameters:
  • extracted_terms: List of parsed concept dictionaries, each containing name, definition, domain, and raw_context.

Step 2: extract_conceptual_triples [depends_on: parse_unstructured_input]

Structure the extracted concepts into standardized ontological nodes and directed relationships:

• Define node schemas:
  • Concept Node:
    • id: Unique UPPERCASE term identifier string (e.g., CONCEPT:AU-ORCH.adapter.hot-cache-invalidation, CONCEPT:AU-KG.ontology.system-definition).
    • name: Clean term label.
    • description: Formal synthesized semantic definition.
    • domain: Research or technical area identifier.
  • Reference Node:
    • id: Hashed source context ID (e.g., REF:abc123xyz).
    • context: Original raw text snippet.
• Map directed edges/relationships:
  • Define taxonomic lineage: (Concept)-[:SUB_CLASS_OF]->(ParentConcept).
  • Define semantic support: (Concept)-[:DEFINED_IN]->(Reference).
• Output parameters:
  • nodes: List of formatted Node objects.
  • relationships: List of formatted Edge/Relationship objects.

Step 3: compile_cypher_transaction [depends_on: extract_conceptual_triples]

Transform the structured graph components into standardized, idempotent Cypher queries:

• Draft transaction queries:
  • For each concept node, construct a parameterized MERGE query: MERGE (c:Concept {id: $id}) ON CREATE SET c.name = $name, c.description = $description, c.domain = $domain ON MATCH SET c.description = $description
  • For each reference node, construct a parameterized MERGE query: MERGE (r:Reference {id: $id}) ON CREATE SET r.context = $context
  • For each relationship, construct a parameterized MERGE query: MATCH (a:Concept {id: $source_id}), (b:Concept {id: $target_id}) MERGE (a)-[r:SUB_CLASS_OF]->(b) (or other relationship types).
• Validate transaction security:
  • Verify queries are read-only-safe in execution templates and strictly construct parameter-backed bindings to prevent Cypher injection vulnerabilities.
• Output parameters:
  • status: "SUCCESS" or "FAILED"
  • cypher_payload: Structured payload containing lists of Cypher query strings and their corresponding argument parameters ready for execution via graph_write.