A PhD supervisor's decade of know-how, bottled as AI prompts

What it does

Supervisor-Skills is a dual-track knowledge base: a handbook of research methodology and copy-pasteable skills—structured prompts for Claude, GPT-4, or DeepSeek. The skills cover the full paper lifecycle, from scoring an idea against a five-dimensional framework to pre-submission review from a top-conference referee’s perspective. It is essentially an attempt to codify the implicit judgment calls that separate a workshop sketch from a SIGMOD full paper.

The interesting bit

The project treats “academic taste” as a configuration problem. Instead of vague advice like “find a novel problem,” it offers literal prompt templates—idea-evaluator, pre-submission-reviewer, figure-designer—that encode a specific professor’s mental checklists. Case studies of actual ICML, ICLR, and VLDB papers show how those checklists map to acceptance-worthy work.

Key highlights

• Seven skills, including an intro-drafter built on a flowchart thinking model and a pre-submission-reviewer that hunts for “low-level errors.”
• A six-chapter handbook on idea generation, scientific plotting, and “Vibe Research” (the author’s term for AI-assisted workflows).
• Real paper autopsies: Alpha-SQL (ICML ‘25), AFlow (ICLR ‘25), and LEAD (VLDB ‘26).
• The quality rubric boiled down to four Ns: Novel Problem, Novel Method, Nice Story, Nice Presentation.

Caveats

• Explicitly early-stage; the TODO list still lacks rebuttal-writing and collaboration guides.
• The handbook and case studies are in Chinese.
• The skills are prompt templates, not agents—you bring your own API keys and judgment.

Verdict

Grab this if you are a first-year PhD student in data systems or AI who needs a sanity check between reading guides and hitting submit. Skip it if you already have a hands-on advisor with unlimited office hours.