Research Writing Skill: Treat Paper Writing as Engineering, Not Chat
> Source: Norman-bury/research-writing-skill, https://github.com/Norman-bury/research-writing-skill
Why Your Paper Has Been Stuck on the Introduction for Three Months
A common scenario for students writing a thesis: open ChatGPT, ask it to "write a research background section about XX," copy-paste, submit to the advisor. The advisor says the logic is wrong, so you go back to the AI, copy-paste again. After ten cycles, the introduction has taken three months and the body hasn't started.
This isn't a writing-skill problem. It's treating paper writing as a one-off chat rather than a trackable, reusable engineering process. research-writing-skill targets exactly this problem.
Positioning: Not a Polishing Tool, but an Engineering Collaboration System
Most "AI writing assistants" just polish text in a single round — no memory, no process, no versioning. research-writing-skill takes a fundamentally different approach: it manages paper writing like software engineering.
- Align goals and constraints before starting
- 7 rounds of Q&A to confirm paper type, discipline, topic, methods, and chapter structure
- Stage-gated execution: topic selection → body writing → figures → self-review → delivery
- Deliverables are project files (.md/.tex) — trackable, recoverable, versionable
- Paper type (thesis / course project / submission draft)
- Discipline (engineering / social science / medicine / law)
- Research topic and background
- Core methods and data sources
- Chapter structure planning
- Preserve research subjects, data ranges, sample definitions, method conditions, metric meanings, experimental boundaries, conclusion limitations, and terminology
- Language adjustments serve naturalness, clarity, and consistent terminology
- Prefer continuous paragraphs; avoid bullet-point padding and bold/italic emphasis tricks
- Avoid mechanical connectives like "first, second, finally, moreover, in addition"
- Avoid empty shell phrases like "it's worth noting that" or "it must be emphasized that"
- writing-core: general writing standards
- writing-humanities: social sciences / humanities
- writing-medical: medicine / biology
- writing-law: law
- Claude Code (.claude-plugin/)
- Cursor (.cursor-plugin/)
- Codex (.codex/)
- OpenCode (.opencode/)
- Gemini CLI (GEMINI.md)
- Generic agents (AGENTS.md)
- Version control: Git can diff Markdown, not Word
- Reusability: Markdown is text — scriptable, replaceable, templatable
- Cross-platform: opens in any editor, no Office dependency
- Flexible conversion: Pandoc converts to Word, PDF, LaTeX; reverse conversion loses formatting
- Norman-bury/research-writing-skill, GitHub, https://github.com/Norman-bury/research-writing-skill
- Version 3.1.0, updated 2026-05-10
The Engineering Workflow
Step 1: Brainstorming (7 rounds of Q&A)
Instead of "write my paper," the system systematically confirms:
This front-loads decisions to reduce rework — many people discover halfway through that their methods or structure don't meet requirements, invalidating everything written so far.
Step 2: Chapter-based Writing
Each chapter is a separate file under chapters/. The introduction and related-work sections must first produce a literature evidence map (refs/evidence-map.md); experiment chapters must first produce an experiment protocol (plan/experiment-protocol.md), table schemas (tables/table-schema.md), and a figure manifest (figures/data-manifest.md). Upfront planning is mandatory — you cannot skip structure and jump straight into prose.
Step 3: Figure Generation
Data plots are generated preferentially via Python scripts for reproducibility. Flowcharts, architecture diagrams, and mechanism diagrams are handled by the figures-diagram module, which produces prompts for image-generation tools like Gemini. Technical and conceptual diagrams are separated, never mixed.
Step 4: Pre-submission Self-review
The peer-review module checks for logic gaps, data consistency, and the boundaries of conclusions.
Step 5: Delivery
Deliverables are Markdown or LaTeX, not Word. Word can be produced manually or via Pandoc. The rationale: research collaboration needs trackable, reusable text assets.
De-AI-flavoring: Preserve Information Density, Adjust Only Expression
The skill's understanding of "removing AI flavor" is precise: don't shorten the text — keep all facts, data, qualifiers, and explanatory sentences, and adjust only the expression.
Unlike most "AI-rate reduction" tools that compress text and delete key qualifiers — turning rigor into vagueness — this skill's stance is: prefer slightly verbose over incomplete. That's the essence of academic writing.
Discipline-specific Routing
Generic tools apply one-size-fits-all logic. This skill ships discipline modules:
This is not just swapping a prompt template — it's a systematic adjustment of argumentation logic, citation formats, and chapter weighting.
Multi-platform Support
The directory-based design adapts to:
No vendor lock-in: write the body with Claude Code, do figures in Cursor, generate flowcharts with Gemini CLI — all artifacts managed in one project directory.
20+ Skill Modules
| Scenario | Skill module | |------|---------| | Entry and routing | using-research-writing | | Medium/full-paper orchestration | paper-orchestration | | Brainstorming | brainstorming-research | | Literature-driven intro/related work | evidence-driven-writing | | Chapter writing | writing-chapters | | Experiment and results planning | experiment-results-planning | | LaTeX output | latex-output | | Literature review | literature-review | | Translation / polishing / de-AI | prompts-collection | | Pre-submission self-review | peer-review | | Statistical analysis | statistical-analysis | | Python data figures | figures-python | | Flowchart / architecture diagrams | figures-diagram | | Environment setup and troubleshooting | environment-setup |
Why Markdown by Default Instead of Word
Word suits final delivery; Markdown suits iterative work. The skill puts each in its right place.
LaTeX Support: Template Parsing and Auto-compilation
If your school or journal provides LaTeX templates, place the .cls/.sty/.tex files in the latex-templates/ directory. The skill parses the template structure and generates corresponding chapter .tex files, producing a directly compilable LaTeX project. It understands your template and generates content to fit it.
Conclusion: Research Writing Needs Industrialization
The core value of research-writing-skill is not "let AI write your paper" but "turn paper writing from a craft workshop into an assembly line." Its engineering workflow, de-AI principles, discipline routing, multi-platform support, and deliverable design all serve one goal: let researchers spend time on real research content instead of endless format and wording edits.
For undergraduates, graduate students, and early-career researchers, this is more practical than any "one-click paper generator" because it promises no shortcuts — only a sustainable, reusable, iterable writing system.
> "Paper writing is not chat. It's engineering."
References