@vintcessun: Frustrated staring at a blank document late at night? Finished your paper but feeling unsure? Advisor too busy, no one to evaluate your ideas. The problem isn't that you're not working hard enough; you need a research co-supervisor always at hand. A HKUST professor distilled his ten years of publishing and reviewing experience into an AI-executable skill set. From evaluating ideas, generating introduction outlines…

X AI KOLs Timeline 07/09/26, 03:38 PM Tools

ai-research-assistant academic-writing open-source hkust 科研辅助 gpt-skills supervisor-skills

Summary

HKUST Assistant Professor Luo Yuyu released the open-source project Supervisor-Skills, which distills ten years of research experience into AI skills executable by large models, covering topic selection, writing, reviewing, etc., aiming to become an AI co-supervisor for researchers.

Frustrated staring at a blank document late at night? Finished writing your paper but feeling unsure? Your advisor is too busy, no one to evaluate your ideas. Actually, the problem isn't that you're not working hard enough; you lack a research co-supervisor always ready to use. A professor from HKUST has distilled his ten years of publishing and reviewing experience into an AI-executable skill set. From evaluating ideas, generating introduction outlines, to checking for flaws as a reviewer before submission, a single prompt can call upon doctoral advisor-level judgment. It's not just a prompt library; it's packaging tacit knowledge into reusable skills. Sounds a bit crazy, but it indeed changes the approach of 'research can only rely on advisors.' The repository is here

Original Article

View Cached Full Text

Cached at: 07/10/26, 12:09 PM

Staring at a blank document late at night? Finished a paper but not confident? Advisor too busy, no one to evaluate your ideas. The real problem isn’t a lack of effort—it’s the absence of an always-available research co-advisor. A professor from HKUST (Guangzhou) has distilled his decade of publishing and reviewing experience into an executable skill set for AI. From evaluating ideas, generating introduction outlines, to acting as a reviewer before submission to catch flaws—a single prompt can unlock advisor-level judgment. This isn’t just another prompt library; it’s encapsulating tacit knowledge into reusable skills. Sounds a bit crazy, but it does change the paradigm that “research can only rely on advisors.” The repo is here.

HKUSTDial/Supervisor-Skills

Source: https://github.com/HKUSTDial/Supervisor-Skills

Supervisor-Skills: Turning a Professor’s Decade of Research Experience into Your AI Co-Advisor

English · 中文

📰 News

[2026-07-10] 🚀 v2.1: Added three new skills: paper-writer (evidence-gated paper body writing), paper-polish (faithful language polishing), deep-research (survey-level literature review); intro-drafter upgraded to full text output; evaluation and review skills now include disciplinary paradigm routing. See CHANGELOG.

[2026-07-03] 🎤 Invited to give a talk at the 2026 China Agent Conference Skill session.

[2026-07-03] 🧩 New drawio-reconstruction skill: reconstruct reference images into editable Draw.io diagrams with one click.

[2026-05-21] ⭐ GitHub Stars exceeded 1,000 (Day 32 since release).

[2026-04-19] 🎉 Project launched: seven anchor skills covering the full paper lifecycle, with a companion handbook systematic course.

Why this project?

Hi, I’m Yuyu Luo (https://luoyuyu.vip/), an assistant professor at the Hong Kong University of Science and Technology (Guangzhou). From my own PhD journey to advising students, I’ve repeatedly seen a common scene: many talented graduate students, when stepping through the door of research, often struggle with similar challenges:

Theory-practice gap: No matter how many “research guides” you read, when facing your own topic, you still don’t know how to start writing.
Scarce guidance resources: Advisors are busy and cannot evaluate every nascent idea promptly and thoroughly.
Pre-submission anxiety: The paper is written, but you’re unsure if the logic is tight, if figures meet top-conference aesthetics, or if the writing hides “novice mistakes.”
AI-era “capability gap”: Large models are powerful, but without judgment and academic taste, AI is just an “advanced toy.” Outstanding research cannot be generated by a “universal prompt” alone.

These challenges collectively form the “Last-Mile Problem” of research. It’s not just a lack of methodology; it’s a composite test of experience, taste, and judgment.

To address this, I launched the Supervisor-Skills project. We “distill” and “forge” the publication, reviewing experience, and academic intuition accumulated over the past decade in top data science and AI conferences (e.g., SIGMOD, VLDB, ICML, NeurIPS) into a set of structured AI Skills that can be precisely executed by large language models (e.g., Claude, GPT-4).

Our vision is simple: Transform the tacit knowledge of experienced researchers into AI tools accessible to everyone, making AI your true, always-online research co-advisor.

This project is still in its early stages. We are working hard to present the insights within our capability without reservation, hoping to provide a practical operations guide for newcomers exploring research. We eagerly look forward to co-building and improving this knowledge base with fellow travelers, so the research journey is no longer lonely. If you find this project helpful, please star it (⭐️ Star ⭐️) in the top right! Thank you!

Community & Communication

WeChat Group

The group has exceeded 200 members and cannot be joined via QR code directly. Please scan to add the admin WeChat account xyp619 with the note “Supervisor-Skills + Name”, and you will be invited by the admin.

You can also follow the Data Intelligence and Analytics Lab DIAL official WeChat account and Bro Crab Talks Research for research news, paper writing tips, AI frontiers, and research tool sharing.

After joining, you can:

🔔 Get first-hand notifications of updates to Skills and the Handbook
💬 Exchange paper writing and AI-assisted research tips with peers
🎙️ Participate in occasional online Q&A and experience sharing sessions

WeChat Group	Official Account	Research Sharing Account
Add admin xyp619 for group invitation	Data Intelligence and Analytics Lab DIAL official account	Bro Crab Talks Research: AI frontiers, papers, and research tool sharing

Tutorial Structure

This tutorial adopts a dual-track architecture of Guide (theory guide) + Skills (executable AI skills):

Supervisor-Skills/
├── README.md                # This file
│
├── handbook/                # 📖 Research & Writing System Guide
│   ├── 01_Preliminary/      # Chapter 1: Macro Understanding & Evaluation
│   │   ├── 1.1_How_to_Evaluate_Paper_Quality.md
│   │   └── PhD_Research_Intro_Guide.pdf   # 📄 Companion lecture notes (downloadable)
│   ├── 02_Idea_Generation/  # Chapter 2: Birth & Elevation of Ideas
│   │   ├── 2.1_Idea_Lifecycle_and_Capability_Matching.md
│   │   ├── 2.2_Idea_Generation_Approaches_Higher_Faster_Stronger.md
│   │   └── 2.3_Advanced_How_to_Do_Disruptive_Innovation.md
│   ├── 03_Paper_Writing/    # Chapter 3: Paper Writing Methodology
│   │   ├── 3.1_Complete_a_Research_Paper_Things_To_Do.md
│   │   ├── 3.2_Introduction_Writing_Thinking_Model.md
│   │   ├── 3.3_Technical_Full_Paper_Thinking_Template.md
│   │   ├── 3.4_Benchmark_and_Evaluation_Paper_Thinking_Template.md
│   │   └── 3.5_Writing_Details_and_Checklist.md
│   ├── 04_Scientific_Plotting/ # Chapter 4: Scientific Plotting Guide
│   │   ├── 4.1_Motivated_Example_Figure.md
│   │   ├── 4.2_Solution_Overview_Figure.md
│   │   ├── 4.3_Experimental_Results_Figure.md
│   │   └── 4.4_Plotting_Checklist_and_Tool_Quick_Reference.md
│   ├── 05_Vibe_Research/    # Chapter 5: Vibe Research in Practice
│   │   ├── 5.1_Vibe_Research_and_Vibe_Coding_Intro.md
│   │   └── 5.2_Li_Boyan_Experience_Sharing_and_Meeting_Notes.md
│   └── 06_Case_Studies/     # Chapter 6: Top-Conference Paper Writing Case Studies
│       ├── 6.1_ICML_2025_Alpha-SQL_Writing_Analysis.md
│       ├── 6.2_ICLR_2025_AFlow_Writing_Analysis.md
│       └── 6.3_VLDB_2026_LEAD_Writing_Analysis.md
│
├── skills/                  # 🛠️ Extracted Executable AI Skills
│   ├── idea-evaluator/      # Idea & research phase skills
│   ├── deep-research/
│   ├── vibe-research-workflow/
│   ├── tech-paper-template/ # Writing phase skills
│   ├── intro-drafter/
│   ├── paper-writer/
│   ├── benchmark-paper-template/
│   ├── paper-polish/        # Self-check & polishing skills
│   ├── pre-submission-reviewer/
│   ├── figure-designer/
│   └── drawio-reconstruction/   # Draw.io reconstruction skill
│
└── assets/                  # Image resources

📖 Handbook: Research & Writing System Guide

📄 Companion lecture notes: PhD Research Intro Guide.pdf — PDF companion for the guide, suitable for printing or reading on iPad, includes a complete framework for starting research.

This section preserves the systematic theoretical framework for in-depth reading and systematic learning. Only by understanding the “way” can you better use the “tool.”

Chapter	Content	Link
Chapter 1: Macro Understanding	Evaluating paper quality from a reviewer’s perspective (Novel Problem, Novel Method, Nice Story, Nice Presentation)	1.1 How to Evaluate Paper Quality / PhD Research Intro Guide
Chapter 2: Idea Generation	Idea lifecycle, 5-dimensional thinking framework (Higher Faster Stronger Cheaper Broader), disruptive innovation	2.1 Idea Lifecycle / 2.2 Higher Faster Stronger / 2.3 Disruptive Innovation
Chapter 3: Paper Writing	Full process of research papers, Introduction thinking model, technical/benchmark paper templates, writing checklist	3.1 Full Process / 3.2 Intro Model / 3.3 Technical Template / 3.4 Benchmark Template / 3.5 Checklist
Chapter 4: Scientific Plotting	Design paradigms for motivation figures, overview figures, experimental figures, and plotting checklist	4.1 Motivation Figure / 4.2 Overview Figure / 4.3 Experimental Figure / 4.4 Checklist
Chapter 5: Frontier Practice	Vibe Research/Coding/Figure/Writing introduction and hands-on experience	5.1 Intro Guide / 5.2 Hands-on Experience
Chapter 6: Top-Conference Cases	Writing analysis of Alpha-SQL (ICML’25), AFlow (ICLR’25), LEAD (VLDB’26)	6.1 Alpha-SQL / 6.2 AFlow / 6.3 LEAD

🛠️ Skills: Extracted Executable AI Skills

This is the core highlight of the repo! We have distilled the above theoretical experience into structured Prompt/Skill files. You can copy these contents directly and import them into your AI assistant (Claude, DeepSeek, Kimi, etc.).

Skill Name	Description	Link
Idea Evaluator	Input your idea; AI will objectively evaluate and score it based on the “Higher Faster Stronger” 5-dimension framework and capability matching table	Use Skill
Vibe Research Guide	AI-assisted full research workflow guidance: Vibe Coding / Vibe Figure / Vibe Writing	Use Skill
Introduction Drafter	Based on the Introduction Flowchart thinking model, input research motivation and directly produce a six-paragraph Intro body (citations verified by real search; or just an outline)	Use Skill
Paper Writer	Evidence-gated paper body writing: from single paragraph to full paper, every factual claim traceable to your materials or verified literature, citations independently checked, never fabricated	Use Skill
Paper Polish	Faithful language polishing: fix grammar, remove AI tone, manage phrasing nuance, rewrite Chinese drafts into submission-ready English; flag any meaning-changing edits for your confirmation	Use Skill
Deep Research	Survey-level literature deep dive: multi-perspective retrieval, per-citation verification, MECE categorization synthesis and cross-comparison, produce a survey report answering research questions point by point	Use Skill
Tech Paper Template	Based on the “Technical Full Paper Thinking Template,” guide you step by step to sort out the complete logic chain of the paper	Use Skill
Benchmark Paper Template	Specifically designed for Benchmark/Evaluation papers, helps organize evaluation logic and experimental design	Use Skill
Pre-Submission Reviewer	Top-conference reviewer perspective! Based on the writing checklist and common English grammar pitfalls, conduct a comprehensive review of the draft	Use Skill
Figure Design Advisor	Tell AI what you want to convey; it will give professional figure design advice based on design paradigms for motivation/overview/experimental figures	Use Skill
Draw.io Reconstruction	Reconstruct reference images, paper figures, architecture diagrams, slide figures, or UI screenshots into editable `.drawio` files, and export PNG previews with visual audit	Use Skill

Quick Start

Send the following prompt to your AI assistant (Claude Code, Cursor, Codex, etc.) to install:

Help me install Supervisor-Skills from https://github.com/HKUSTDial/Supervisor-Skills with Skills.

Contributing & Feedback

This is an innovative open-source AI Skills project. We hope to address the challenge of making agent-based research assistance both “useful and compliant.”

Everyone is welcome to try it out, submit PRs, open Issues, or share top-conference papers you’ve written using these skills!

If you find this project helpful, please star it in the top right! Your support is our biggest motivation for continuous updates.

Email: Yuyu Luo (yuyuluo [AT] hkust-gz.edu.cn).

Thanks

Thanks to Yin Wu (https://openreview.net/profile?id=%7EYin_WU2), Boyan Li (https://liboyan.vip/), Yupeng Xie (https://xypkent.github.io/) for helping organize this repository and providing valuable suggestions!

TODO

Guide

How to write a good Rebuttal? What if the reviewer ignores my Rebuttal?
How to collaborate efficiently academically with others?

License

This project is open-sourced under the CC BY-NC-SA 4.0 (https://creativecommons.org/licenses/by-nc-sa/4.0/) license. Non-commercial sharing and adaptation are welcome, but please provide attribution.