DeepRefine-Skill

HKUST KnowCompResearch skill collectionsListed

About this tool

An agent skill from HKUST KnowComp, Yangqiu Song's team. /deeprefine uses query history to repair incomplete, incorrect or redundant content in graphify knowledge graphs and LLM-Wiki, improving later retrieval and Q&A. It syncs history, proposes refinements, performs dry-run checks and assigns confidence levels. Changes are written only after user confirmation. Associated paper: arXiv 2605.10488.

Official DeepRefine project logo
Project logo · View original asset
  • Feature 1: /deeprefine refines graphify knowledge graphs and LLM-Wiki using query history.
  • Feature 2: Dry-run checks classify confidence as HIGH, MEDIUM or LOW; LOW actions are not applied by default.
  • Feature 3: Write changes and refresh the Wiki after user confirmation.
  • Feature 4: Support for Claude Code, Codex, Cursor, Gemini CLI, Copilot CLI, OpenCode and terminal CLI.
  • Feature 5: Associated paper: arXiv 2605.10488.

Resource information

Maintainer
HKUST-KnowComp, Associate Professor Yangqiu Song's team; university laboratory.
Type
Skill
Category
Research skill collections
Runs on
Agent Skills
Who it's for
Researchers maintaining personal or group knowledge bases and literature graphs, using past queries to repair missing, incorrect or redundant links.
License at time of listing
MIT
Source
GitHub

TITU Work Market1

Available in the TITU Work Market. Search for the names below in the app to install.

deeprefine

Publisher
HKUST-KnowComp
Version
1.0.0
License
MIT
Last updated
10/28/2026
Installs
0