Skip to content

SKILL.md

Latest commit

 

History

History
68 lines (53 loc) · 2.1 KB

SKILL.md

File metadata and controls

68 lines (53 loc) · 2.1 KB
Raw

name: kragen-knowledge-graph description: Graph-RAG Solver keywords:

  • knowledge-graph
  • RAG
  • reasoning
  • graph-of-thoughts
  • biomedical-qa measurable_outcome: Return a reasoning path and an answer supported by ≥3 knowledge graph nodes for complex biomedical questions with <5s latency. license: MIT metadata: author: Bioinformatics Oxford version: "1.0.0" compatibility:
  • system: Python 3.9+ allowed-tools:
  • run_shell_command
  • web_fetch

KRAGEN (Knowledge Graph Enhanced RAG)

A knowledge graph-enhanced Retrieval-Augmented Generation system for biomedical problem solving, using Graph-of-Thoughts (GoT) reasoning.

When to Use

  • Complex Reasoning: Questions requiring multi-hop deduction (e.g., "How does gene A influence disease B via protein C?").
  • Hypothesis Verification: Checking if a proposed mechanism is supported by existing knowledge graphs.
  • Literature Synthesis: Combining facts from structured DBs and unstructured text.

Core Capabilities

  1. Graph Retrieval: Query biomedical knowledge graphs (e.g., PrimeKG, SPOKE).
  2. Graph-of-Thoughts: structured reasoning over retrieved nodes.
  3. Vector DB Integration: Combines graph data with vector embeddings for hybrid search.

Workflow

  1. Input: Natural language question.
  2. Retrieval: Fetch relevant sub-graph and similar text chunks.
  3. Reasoning: LLM traverses the graph to find connecting paths.
  4. Answer: Generate response with citation of graph nodes.

Example Usage

User: "Explain the mechanism connecting BRCA1 mutations to ovarian cancer."

Agent Action:

python -m kragen.solve --question "BRCA1 mutations to ovarian cancer mechanism"