Structured guidance for composing, formatting, and presenting scientific research papers for top-tier venues (CVPR, ICCV, NeurIPS, ICML, ACL, ICLR, etc.). Covers the full workflow: section-by-section drafting, figure and table design, LaTeX typography, narrative logic, and submission-ready polishing. Use to write, revise, or format a manuscript into camera-ready or preprint PDF.
Guidance for producing publication-quality scientific research papers, extracted
from analysis of the ResNet (He et al., CVPR 2016 Best Paper) SOTA artifact and
generalized for computer vision, machine learning, and broader scientific domains.
Reference Source
Type: Uploaded PDF artifact (native PDF)
Artifact: ResNet paper - "Deep Residual Learning for Image Recognition"
Reference File Type: PDF (conference proceeding style, double-column LaTeX)
Primary Language: English (CJK-capable typography strategy defined for
mixed-script output)
Primary target for camera-ready submission via LaTeX
DOCX
Yes
For collaborative drafting and advisor review
PPTX
Yes
For oral-presentation slide decks derived from paper
When the user does not specify a format, default to PDF for final manuscripts
and PPTX for oral-presentation derivatives.
CRITICAL: When generating a full paper, produce all required sections with
all figures, tables, equations, and references - never produce a partial or
abbreviated paper. The reference artifact is 12 pages; match this scale for full
paper reproductions.
Workflow Overview
Plan - Define target venue, page limit, and section budget
Draft - Write ALL sections following the narrative-logic contract
Design - Create ALL figures, tables, and equations per the guidelines
Polish - Apply typography, cross-references, and submission formatting
Convert - Render to requested output format
Core Contracts
Load the relevant reference file when the user request touches that domain:
Structure & Narrative: See references/structure_contract.md
Page count: Matches target venue (8-12 pages), not compressed
1---2name: sci-paper3description: Structured guidance for composing, formatting, and presenting scientific research papers for top-tier venues (CVPR, ICCV, NeurIPS, ICML, ACL, ICLR, etc.). Covers the full workflow: section-by-section drafting, figure and table design, LaTeX typography, narrative logic, and submission-ready polishing. Use to write, revise, or format a manuscript into camera-ready or preprint PDF.4---56# SCI Paper Writing Skill78Guidance for producing publication-quality scientific research papers, extracted9from analysis of the ResNet (He et al., CVPR 2016 Best Paper) SOTA artifact and10generalized for computer vision, machine learning, and broader scientific domains.1112## Reference Source1314- **Type**: Uploaded PDF artifact (native PDF)15- **Artifact**: ResNet paper - "Deep Residual Learning for Image Recognition"16- **Reference File Type**: PDF (conference proceeding style, double-column LaTeX)17- **Primary Language**: English (CJK-capable typography strategy defined for18 mixed-script output)19- **Page count**: 12 pages (8 main + 1 references + 3 appendix)2021## Supported Outputs2223| Format | Default | Notes |24|--------|---------|-------|25| PDF | Yes | Primary target for camera-ready submission via LaTeX |26| DOCX | Yes | For collaborative drafting and advisor review |27| PPTX | Yes | For oral-presentation slide decks derived from paper |2829When the user does not specify a format, default to **PDF** for final manuscripts30and **PPTX** for oral-presentation derivatives.3132**CRITICAL**: When generating a full paper, produce **all required sections** with33**all figures, tables, equations, and references** - never produce a partial or34abbreviated paper. The reference artifact is 12 pages; match this scale for full35paper reproductions.3637## Workflow Overview38391. **Plan** - Define target venue, page limit, and section budget402. **Draft** - Write ALL sections following the narrative-logic contract413. **Design** - Create ALL figures, tables, and equations per the guidelines424. **Polish** - Apply typography, cross-references, and submission formatting435. **Convert** - Render to requested output format4445## Core Contracts4647Load the relevant reference file when the user request touches that domain:4849- **Structure & Narrative**: See `references/structure_contract.md`50 - Section hierarchy, narrative arc, paragraph-level patterns, citation strategies.51- **Visual Style & Typography**: See `references/style_contract.md`52 - Page layout, font system, color usage, mathematical notation, header/footer rules.53- **Figures, Tables & Equations**: See `references/figure_table_guidelines.md`54 - Caption conventions, numbering, chart styling, table formatting, equation layout.5556## Complete Paper Requirements5758When producing a full paper reproduction or draft, the output MUST include:5960### All Sections (in order)611. Title page: Title, Authors, Affiliations, Abstract, Keywords622. Section 1: Introduction (1-1.5 pages)633. Section 2: Related Work (0.5-1 page)644. Section 3: Method/Approach (2-3 pages) with subsections 3.1, 3.2, 3.3, 3.4655. Section 4: Experiments/Results (2.5-3.5 pages) with subsections 4.1, 4.2, 4.3666. Section 5: Conclusion (0.3-0.5 pages)677. References (1 page, numbered [1]-[N])688. Appendices A-C (optional, 2-3 pages)6970### All Visual Elements71- **Figures**: ALL referenced figures with proper captions below (Fig. 1-N)72- **Tables**: ALL referenced tables with proper captions above (Table 1-N)73- **Equations**: ALL numbered equations with sequential numbering (1), (2), ...74- **Footnotes**: Numbered superscript footnotes at page bottom7576### Page Count Rules77- Full paper: 8-12 pages depending on content78- Minimum: Each major section must have enough content (Introduction >= 5 paragraphs)79- Never produce a paper compressed into 1-2 pages8081## Venue-Specific Adaptation8283| Parameter | CVPR/ICCV/ECCV | NeurIPS/ICML/ICLR | ACL/EMNLP |84|-----------|---------------|-------------------|-----------|85| Page limit | 8 + refs | 9 + refs | 8 + refs |86| Columns | 2 | 1 (NeurIPS/ICML), 2 (ICLR) | 2 |87| Abstract | 1 para, ~150 words | 1 para, ~200 words | 1 para, ~150 words |88| Fig/Table per page | 1-2 | 1-2 | 1-2 |8990## Quick Checklist (Pre-Submission)9192- [ ] **Title**: Concise, specific, no unnecessary jargon93- [ ] **Abstract**: Contains background, problem, method, key results, significance94- [ ] **Introduction**: Domain -> sub-domain -> specific gap -> contribution list -> roadmap95- [ ] **Related Work**: Organized by theme, not by paper; explicit differentiation96- [ ] **Method**: Mathematical formulation before implementation; all subsections present97- [ ] **Experiments**: ALL subsections with datasets, metrics, baselines, results, ablations98- [ ] **Figures**: All referenced before appearing; captions self-contained; BELOW figure99- [ ] **Tables**: All referenced; best results in **bold**; captions ABOVE table100- [ ] **Equations**: Numbered sequentially; all symbols defined in surrounding text101- [ ] **References**: Complete numbered bibliography [1]-[N]102- [ ] **Page count**: Matches target venue (8-12 pages), not compressed
Run npx skillmds@latest add serejaris/sci-paper in your terminal (requires Node.js), paste this page's agent-chat prompt into Claude, Cursor, or any MCP-connected agent, or download the SKILL.md file and copy it into your agent's skills directory.
Structured guidance for composing, formatting, and presenting scientific research papers for top-tier venues (CVPR, ICCV, NeurIPS, ICML, ACL, ICLR, etc.). Covers the full workflow: section-by-section drafting, figure and table design, LaTeX typography, narrative logic, and submission-ready polishing. Use to write, revise, or format a manuscript into camera-ready or preprint PDF. It is listed under Design & Media on SkillMD.
This skill has not completed SkillMD's automated safety review yet. SkillMD never runs a skill's scripts for you; review the SKILL.md before installing.
This skill is tagged as working with Claude Code, Claude.ai, OpenAI Codex. SKILL.md is an open format, so most agents that read a skills directory can load it too.
Yes. Installing skills from SkillMD is free, and the skill stays under its author's original license.
serejaris (@serejaris) published this skill. Their other Agent Skills are listed on their SkillMD profile.