HKUSTDial/Supervisor-Skills

What it solves

Supervisor-Skills addresses the "last-mile problem" in academic research, where students often struggle to translate theoretical guidelines into practical execution. It bridges the gap between having a powerful LLM and producing high-quality research by providing the academic intuition, taste, and judgment of a senior researcher, helping users avoid common pitfalls in idea generation, paper writing, and submission.

How it works

The project uses a dual-track architecture combining theoretical guides and executable AI skills:

• Handbook (Theoretical Guides): A systematic framework covering the entire research lifecycle, including how to evaluate paper quality, generate and refine ideas, write introductions, design scientific plots, and analyze top-tier conference papers.
• AI Skills (Executable Prompts): Distilled academic experience converted into structured prompts that can be imported into LLMs (like Claude, GPT-4, DeepSeek, or Kimi). These skills automate specific tasks such as idea evaluation, evidence-gated paper writing, language polishing, and deep literature research.

Who it’s for

It is primarily designed for graduate students and early-career researchers who need constant, high-quality guidance on how to conduct research and write papers for top-tier AI and data science conferences (e.g., SIGMOD, VLDB, ICML, NeurIPS).

Highlights

• Comprehensive Research Lifecycle: Covers everything from initial idea evaluation to final pre-submission review.
• Evidence-Gated Writing: A paper-writer skill that ensures factual claims are traceable to source materials or verified literature to prevent AI hallucinations.
• Academic-Grade Polishing: A paper-polish skill focused on removing "AI-sounding" prose and refining language while remaining faithful to the original meaning.
• Visual Aid Integration: Includes a drawio-reconstruction skill to convert reference images into editable Draw.io files.
• Deep Research Capability: A deep-research skill for survey-level literature reviews with multi-perspective retrieval and cross-comparison.