Refine Graphify knowledge graphs through an evidence-based abduction and mutation loop.
npx skills add https://github.com/HKUST-KnowComp/DeepRefine-Skill --skill deeprefine
Please help me install this Agent Skill. Skill: deeprefine Source: https://github.com/HKUST-KnowComp/DeepRefine-Skill/tree/main Command: npx skills add https://github.com/HKUST-KnowComp/DeepRefine-Skill --skill deeprefine
This Skill addresses the degradation of LLM-Wiki knowledge graphs by providing a structured, evidence-aware refinement loop that fixes incomplete, incorrect, or redundant graph data.
Invoke the deeprefine skill to initiate the refinement loop for all pending queries in your current knowledge base project.
High-intent search queries and answers about installing and using this skill.
To refine a knowledge graph, you can execute an agent-native abduction and mutation loop that diagnoses errors, inserts missing edges, and disambiguates nodes based on evidence validation before applying updates.
Evidence-aware knowledge graph refinement is a process that labels all proposed graph mutations by evidence strength (HIGH/MEDIUM/LOW) and audits them before application, ensuring incomplete or incorrect graph data is safely corrected.
You can fix incorrect knowledge graph data without manual JSON editing by invoking an agent-native refinement loop that abducts errors, generates refinement actions, and safely updates the graph structure directly within your agent session.
Yes, agent-native refinement requires Graphify-generated knowledge graphs as input to execute its abduction and mutation loop for diagnosing errors, inserting edges, and disambiguating nodes across multiple agent frameworks.
The limitation of this abduction and mutation loop is that it requires strict adherence to the DeepRefine control flow and mandates evidence-based validation before applying any graph updates, preventing direct structural changes.
Yes, you can use OpenAI agents to diagnose errors and disambiguate nodes, as the refinement loop supports complex reasoning tasks across multiple agent frameworks while requiring an OpenAI dependency to operate.
costiashBootstrap knowledge graphs from video transcripts into structured graphs.
cookeyholderAnalyze git diffs against a knowledge graph to identify affected components.
EwallywGenerate competing hypotheses and rank evidence to trace complex issues.
kinnerkarmanishTraverse the knowledge graph to discover connections and synthesis opportunities.
