# Supervisor-Skills: HKUST(GZ) research skills packaged for Claude and GPT-4

> Supervisor-Skills is a repository of executable AI skills and a handbook that distils a decade of supervision experience at HKUST(GZ) into prompts for idea evaluation, paper writing and pre-submission review. It is a prompt library, not a runtime, and that shapes who can use it.

**HKUSTDial/Supervisor-Skills** — 将博导十年科研经验炼化为可直接调用的 AI 技能。从 Idea 构思到论文投稿，你的 AI 科研副导师。

- Repository: https://github.com/HKUSTDial/Supervisor-Skills
- Stars: 7,760 · Forks: 480
- Language: Python
- License: NOASSERTION
- Published: 2026-09-22 · Updated: 2026-09-22 · Language: en
- Canonical page: https://hysenlabs.com/projects/hkustdial-supervisor-skills

## The last-mile problem Supervisor-Skills targets

The README frames the project around what it calls the "last-mile problem" in research: a student can read every methodology guide and still not know how to evaluate their own idea, structure an Introduction, or judge whether a figure meets a top-conference standard. The author, Luo Yuyu, an assistant professor at HKUST(GZ), states that the skills distil ten years of publishing and reviewing experience from venues such as SIGMOD, VLDB, ICML and NeurIPS into prompts a large language model can execute. The intended user is a graduate student who already has a topic but lacks timely feedback. The project is explicit that a generic prompt does not produce research taste, which is why it ships a handbook alongside the skills rather than prompts alone. That pairing is the actual product: theory in handbook/, executable checklists in skills/.

## How the skills and handbook are organised

The repository has two tracks. The handbook/ directory holds Markdown chapters in six groups, from evaluating paper quality through idea generation, writing, plotting, Vibe Research and case studies of specific papers (Alpha-SQL at ICML 2025, AFlow at ICLR 2025, LEAD at VLDB 2026). The skills/ directory holds twelve named skill folders, including idea-evaluator, deep-research, intro-drafter, paper-writer, paper-polish, pre-submission-reviewer, figure-designer, rebuttal-guidance and drawio-reconstruction. The v2.1 release notes say paper-writer, paper-polish and deep-research were added, that intro-drafter was upgraded to produce body text, and that evaluation and review skills gained subject-paradigm routing. There is no server component described. Each skill is a directory of instructions the model reads; the model does the work, and the repository supplies the structure, the checklist and the routing logic.

## Installing Supervisor-Skills and running idea-evaluator

There is no package published to PyPI or npm in the README, and no installer script is documented. The repository is consumed by cloning it and pointing your model at the relevant skill directory. The README's own layout shows the two directories you need:

```
Supervisor-Skills/
├── handbook/                          # 科研与写作系统指南
├── skills/                            # 提炼出的可执行 AI Skills
│   ├── idea-evaluator/                # 构思与调研阶段技能
│   ├── deep-research/
│   ├── paper-writer/
│   └── pre-submission-reviewer/
└── assets/
```

Once the repository is on disk, open the skill you want and load its instructions into your model session. The README gives no wrapper command, so the mechanics depend on your client: paste the skill file as context, or attach the directory if your tool supports it. The README also points to a companion PDF, handbook/01_Preliminary/博士生科研入门辅导.pdf, for offline reading, and to the matching handbook chapters under handbook/02_Idea_Generation/ for idea work and handbook/03_Paper_Writing/ for drafting. Note that the README does not document version pinning, so a copy taken today and a copy taken after the next release may behave differently.

## Where Supervisor-Skills stops being the right tool

The repository is a knowledge artefact, not a system. Nothing in the README describes an API, a hosted endpoint, a database, a test suite or a CI pipeline that runs the skills. If your requirement is an automated review step in a submission system, or a service your team calls over HTTP, this project does not provide it, and the absence is structural rather than a gap awaiting a patch. The second limit is model dependence: the skills assume a capable chat model such as Claude or GPT-4, and their output quality tracks that model's judgement. A weaker or heavily quantised model will follow the checklist and still produce weak reasoning, and the repository offers no evaluation harness to detect that. Third, the project is early. The README says so directly, and the release history runs from April 2026 to July 2026, with the last push on 2026-09-05. There is no retrieved release on a package index, so upgrade tracking is manual.

## How it differs from a literature tool or a paper generator

The closest familiar alternative is a literature-search or paper-generation tool, and the difference is where the judgement sits. A search tool retrieves and summarises; a generation tool produces text from a prompt. Supervisor-Skills inserts an explicit evaluative layer between the two, with named skills such as idea-evaluator and pre-submission-reviewer that are meant to reject weak work rather than produce more of it. The handbook chapters on evaluating paper quality and on the Introduction thinking model are the clearest expression of that stance: they are checklists for criticism, not templates for volume. If you want a tool that drafts a survey from a query, this is not it. If you want something that tells you your motivation figure does not land, the figure-designer and pre-submission-reviewer skills are aimed at exactly that, though the README does not quantify how well they do it.

## Licence, maintenance and upgrade cost

The README badge identifies the licence as CC BY-NC-SA 4.0, while the repository metadata reports NOASSERTION. Those two signals disagree, and the LICENSE file is the one to read before you rely on either. The practical consequence of CC BY-NC-SA 4.0 is that commercial use is restricted and derivatives carry the same terms; that matters if you plan to embed these skills in a paid product or an institutional service. This is a description of the licence text, not legal advice. On maintenance, the last push was on 2026-09-05, and the repository is not archived, so the project is current as of that date. There is no release artefact to pin, so your upgrade path is a fresh copy of the repository and a diff against skills/ and handbook/, plus a read of CHANGELOG.md. Budget time for that after any update, because a changed skill file changes model behaviour with no version number to warn you.

## Conclusion

Adopt Supervisor-Skills if you are a PhD student or early-career researcher who already works inside a chat model such as Claude or GPT-4 and wants structured, opinionated checklists for idea evaluation, drafting and pre-submission review. Skip it if you need a runnable pipeline, an API or a self-hosted service, because the repository ships Markdown skills and a handbook, not executable code, and the README does not describe a hosted endpoint. Before you commit, read the LICENSE file to confirm the CC BY-NC-SA 4.0 terms and check whether the non-commercial clause fits your institution, then open one skill directory under skills/ and confirm its prompt structure matches the model you plan to use.

## FAQ

### What is Supervisor-Skills and who is it for?

It is a repository of AI skills and a handbook that packages research methodology for idea evaluation, paper writing, plotting and pre-submission review. The README describes it as aimed at graduate students who need structured feedback between supervisor meetings.

### Does Supervisor-Skills work with Claude and GPT-4?

The README states that the skills are structured to be executed by large language models such as Claude and GPT-4. There is no runtime in the repository, so the model you use supplies the execution.

### How do I set up Supervisor-Skills?

The README documents no package on an index and no installer script. The documented path is to obtain the repository and load the relevant directory under skills/ into your model session.

### What licence does Supervisor-Skills use?

The README badge shows CC BY-NC-SA 4.0, while the repository metadata reports NOASSERTION. The LICENSE file is the authoritative source, and the non-commercial clause is the part to check for your use case.

## Sources

- [HKUSTDial/Supervisor-Skills on GitHub](https://github.com/HKUSTDial/Supervisor-Skills)
- [Issues](https://github.com/HKUSTDial/Supervisor-Skills/issues)
- [README](https://github.com/HKUSTDial/Supervisor-Skills/blob/main/README.md)

---

Hysen Labs editorial analysis, written from the project's own repository and release notes. Cite the canonical page: https://hysenlabs.com/projects/hkustdial-supervisor-skills