Research Writing Assistant

Upgrade "paper writing" from one-off chat sessions into a trackable, recoverable, and reusable engineering-style collaboration workflow.
This Skill is designed for undergraduates, graduate students, and early-stage researchers with a clear goal: fewer detours, less rework, and more time spent on research that truly matters.

What This Is

This is not a prompt pack that only polishes sentences. It is a complete research writing collaboration system.
Before starting any task, it aligns goals and constraints through a brainstorming process to confirm paper type, research background, methods, and chapter structure, then routes to the appropriate skill module based on your discipline and task type.

If you are working on a thesis, a course project paper, or a submission draft, this Skill is more reliable than ordinary conversational writing tools — because it emphasizes process, documentation, and write-back, without depending on single-session memory.

Core Capabilities

• Brainstorming: 7-round Q&A to confirm paper type, discipline, title, research background, methods, chapter structure
• End-to-end collaboration: From topic development, body writing, and figure generation to pre-submission self-review, executed with stage gates
• De-AI writing: Does not treat polishing as compression; it preserves research objects, data scope, method conditions, metric meanings, and conclusion boundaries
• Figure support: Python handles reproducible data figures, while Gemini and similar image tools handle prompts for flowcharts, architecture diagrams, and conceptual figures
• Discipline-specific writing support: Routed modules for engineering, social sciences, medicine, and law
• Literature review support: Integration of English-language search and Chinese literature organization
• LaTeX template support: Provide your institution/journal template, auto-generate compilable LaTeX project
• Environment automation: Miniconda installation, virtual environment creation, plotting dependency setup

Supported Platforms

This Skill uses a directory-based design, adapted for the following platforms:

Platform	Configuration
Claude Code	.claude-plugin/plugin.json
Cursor	.cursor-plugin/plugin.json
Codex	.codex/INSTALL.md
OpenCode	.opencode/INSTALL.md
Gemini CLI	GEMINI.md
Others	AGENTS.md

What You Get

By default, Skill outputs are project files — not finished Word documents.

Output Type	Default Format	Notes
Written body	.md / plain text / .tex	Suitable for version control and further processing
Chapter files	chapters/*.md	One file per chapter
LaTeX project	chapters/*.tex + main.tex	Directly compilable
Figure scripts	.py	Reproducible figure generation logic
Prompt assets	.md	Reusable templates for translation, polishing, and de-AI-ification

De-AI Writing Boundaries

De-AI writing here does not mean shortening the text. Unless the user explicitly asks for a shorter version, the Skill should not remove facts, data, limiting conditions, or explanatory sentences. Research objects, data ranges, sample definitions, method conditions, metric meanings, experiment boundaries, conclusion limits, and domain-specific terms should remain intact. Language edits serve natural expression, clear logic, and stable wording.

Paper prose should stay in continuous paragraphs instead of becoming a stack of bullet points, and it should not rely on bold or italic styling to create emphasis. The writing avoids mechanical transitions such as "first", "second", "finally", "in addition", and "in conclusion", and it also avoids empty framing phrases such as "it is worth noting that" or "it should be pointed out that". If the original text is complete and naturally ordered but slightly wordy, the safer edit is a light cleanup rather than cutting useful information for the sake of looking concise.

Important Boundaries (Read First)

1. The Skill does not automatically generate or write directly to .docx files by default.
2. The Skill does not open Word and apply formatting on your behalf — you will need to copy manually or use a conversion tool.
3. The Skill can generate plain-text paragraphs ready to paste into Word, but final styling (heading levels, headers/footers, table of contents, reference fields) must be handled in Word.
4. References and data are never fabricated; all citations must be traceable. Please independently verify high-risk conclusions.

Installation

Option 1: Direct Download

Download the repository, extract it, and copy research-writing-skill/ into your paper writing directory.

Option 2: Git Clone

git clone https://github.com/Norman-bury/research-writing-skill.git
cd research-writing-skill

Platform-specific Installation

• Codex: See .codex/INSTALL.md
• OpenCode: See .opencode/INSTALL.md
• Others: Place the entire directory in your paper project root

Figure Examples

For data-driven result figures, the Skill can generate Python scripts first, then the figures are rendered locally. The two examples below show paper figures produced in that workflow, suitable for training curves, metric comparison, and experiment result checks.

For flowcharts, model architecture diagrams, and mechanism figures, figures-diagram can first generate the prompt, and then the prompt can be used with Gemini or another image-generation tool. The two examples below were drawn by Gemini from generated prompts.

Standard Collaboration Workflow (Recommended)

1. Brainstorming: Say "I want to write a paper", the Skill will guide you to confirm paper type, title, research background, etc.
2. Chapter planning: After confirming chapter structure, the Skill creates framework in chapters/
3. Chapter writing: Write chapter by chapter, one file per chapter
4. Figure generation: When data figures are needed, the Skill generates Python scripts
5. Self-review: Use the peer-review skill for pre-submission review
6. Delivery: Manually migrate to Word/LaTeX for final formatting

Quality Gates

Medium or full-paper tasks should go through skills/paper-orchestration/ first, with a persistent task packet and a capability-use audit in plan/progress.md. Introduction and Related Work require refs/evidence-map.md or plan/evidence-map.md before drafting. Experiment and Results sections require plan/experiment-protocol.md, tables/table-schema.md, and figures/data-manifest.md.

Common checks:

powershell -ExecutionPolicy Bypass -File scripts/check_skill_integrity.ps1
powershell -ExecutionPolicy Bypass -File scripts/research_quality_gate.ps1 -ProjectPath <paper-project>

Skill Map

Scenario	Skill
Entry and routing	skills/using-research-writing/
Medium/full-paper orchestration	skills/paper-orchestration/
Brainstorming	skills/brainstorming-research/
Evidence-driven Introduction/Related Work	skills/evidence-driven-writing/
Chapter writing	skills/writing-chapters/
Experiment and results planning	skills/experiment-results-planning/
LaTeX output	skills/latex-output/
General writing standards	skills/writing-core/
Humanities / social science writing	skills/writing-humanities/
Medical / biology writing	skills/writing-medical/
Law writing	skills/writing-law/
Literature review	skills/literature-review/
Translation / polishing / de-AI	skills/prompts-collection/
Pre-submission self-review	skills/peer-review/
Statistical analysis	skills/statistical-analysis/
Python figures	skills/figures-python/
Flowcharts / architecture diagrams	skills/figures-diagram/
Environment setup and troubleshooting	skills/environment-setup/

Discussion Group

If you want to discuss paper structure, literature organization, figure prompts, de-AI writing, or template adaptation while using this Skill, you are welcome to join the research writing discussion QQ group.

QQ group: 649198361

LaTeX Template Usage

If you have a LaTeX template from your institution or journal:

1. Place template files (.cls, .sty, .tex, etc.) in latex-templates/ directory
2. Tell the AI "use my LaTeX template"
3. The AI will parse template structure and generate corresponding .tex chapter files

See latex-templates/README.md for details.

Delivering Markdown to Word

Option A: Manual Copy (Default Recommendation)

1. Ask the Skill to output a "plain-text paragraph version" (avoiding Markdown markers)
2. Copy the body in your editor and paste it into Word
3. Apply your institution's template styles in Word (headings, body text, captions)
4. Manually check equations, references, figure/table numbers, and cross-references

Option B: Pandoc Conversion (Optional)

If Pandoc is installed locally, first confirm that the command is available:

pandoc --version

The simplest conversion is:

pandoc draft.md -o draft.docx

If you already have a Word style template from your institution or journal, use it as a reference document:

pandoc draft.md --reference-doc=template.docx -o draft.docx

Pandoc handles format conversion and style inheritance, but it does not replace final manual checking. After conversion, check heading levels, figure and table numbers, equations, references, headers, footers, and table-of-contents fields.

FAQ

Why are the default outputs not Word files?

Research collaboration needs text assets that are trackable, reusable, and version-controlled. Markdown is better suited for iterative work. Word is appropriate for final delivery, so it is handled as the last step.

Can it generate the final submittable version for me?

It can produce content close to a final draft, but your institution's template, table of contents fields, page numbers, reference fields, and formatting details should still be finalized in Word.

Will this Skill fabricate references?

No. The rules explicitly prohibit fabricating references or data. All citations must be traceable.

Repository Structure

research-writing-skill/
├── SKILL.md                    # Main entry (legacy platform compatible)
├── AGENTS.md                   # General agent configuration
├── GEMINI.md                   # Gemini CLI configuration
├── CHANGELOG.md                # Version history
├── .claude-plugin/             # Claude Code configuration
├── .cursor-plugin/             # Cursor configuration
├── .codex/                     # Codex configuration
├── .opencode/                  # OpenCode configuration
├── hooks/                      # Session start scripts
│   ├── session-start
│   ├── hooks.json
│   └── hooks-cursor.json
├── img/                        # README example images
├── skills/                     # Skill modules directory
│   ├── using-research-writing/
│   ├── paper-orchestration/
│   ├── brainstorming-research/
│   ├── evidence-driven-writing/
│   ├── writing-chapters/
│   ├── experiment-results-planning/
│   ├── latex-output/
│   ├── literature-review/
│   ├── figures-python/
│   ├── figures-diagram/
│   ├── peer-review/
│   ├── statistical-analysis/
│   ├── verification/
│   ├── environment-setup/
│   ├── prompts-collection/
│   ├── writing-core/
│   ├── writing-humanities/
│   ├── writing-medical/
│   └── writing-law/
├── latex-templates/            # User LaTeX templates directory
├── modules/                    # Legacy modules (kept for compatibility)
├── templates/                  # Code templates
├── plan-template/              # Plan templates
└── scripts/                    # Utility scripts

Version

• Version: 3.1.0
• Updated: 2026-05-10
• Maintenance goal: Stable workflow, traceable content, deliverable outputs, multi-platform support