student-write-paper - SKILL.md Agent Skill

name: student-write-paper description: Draft paper sections that stay aligned with the actual evidence and do not overclaim.

Student Write Paper

Student-specific extensions

Write around a claim-evidence structure:

claim
evidence
limitation
next question

Every draft should contain:

contribution bullets
explicit limitations
no result statement without a matching artifact or citation

Do not write a stronger story than the experiments can defend.

Purpose

Draft paper sections for eventual publication, following standard ML/AI conference format.

Paper Structure

{artifact_dir}/{your-name}/
├── paper_draft_v1.md
├── paper_draft_v2.md    (revised after review)
└── figures/
    ├── main_results.png
    └── architecture.png

Standard Sections

1. Abstract

Problem statement (1-2 sentences)
Approach (1-2 sentences)
Key results (1-2 sentences)
Significance (1 sentence)

2. Introduction

Motivation and problem context
Limitations of existing approaches
Our contribution (bulleted list)
Paper organization

3. Related Work

Organized by theme, not chronologically
Position our work relative to prior art
Note what we borrow vs. what is novel

4. Method

Formal problem definition
Proposed approach with mathematical notation
Algorithm pseudocode if applicable
Complexity analysis

5. Experiments

Experimental setup (datasets, baselines, metrics, implementation details)
Main results table
Ablation studies
Analysis and discussion

6. Conclusion

Summary of contributions
Limitations
Future work

Writing Rules

Use precise language — avoid vague claims ("significantly better" → "3.2% improvement in F1")
Every claim must be supported by evidence (experiments or citations)
Define all notation on first use
Tables and figures must be self-contained (readable without main text)
Use consistent notation throughout

Versioning

Save each major revision as a new version: paper_draft_v1.md, paper_draft_v2.md
After reviewer feedback, create a new version addressing all comments
Include a changelog at the top of each new version

Submission Checklist

All experiments are reproducible (seeds, configs logged)
Baselines are fair and up-to-date
Error bars or confidence intervals included
Limitations section is honest
All figures are high-resolution
References are complete and properly formatted