Academic Style Rewriter

中文说明

Section-aware academic writing skill for Claude Code, Codex, Gemini CLI, Cursor, and similar agent tools.

Why This Exists

Many researchers can already draft the content of a paper section, but the prose still reads like notes, lab talk, or a manual. The harder part is not having ideas. The harder part is rewriting each section in the rhetorical style reviewers expect.

This repository treats academic writing as a section-aware rewriting problem. Instead of using one generic prompt for an entire paper, it organizes rewriting logic around the writing patterns of specific sections.

Who It Is For

• Researchers writing ML, AI, systems, and related papers
• Authors who already have a rough draft for a section
• Users who want help turning spoken or note-like prose into reviewer-facing academic writing
• People using agent or CLI tools such as Claude Code, Codex, Gemini CLI, and Cursor

What Makes It Different

• Section-aware instead of one-size-fits-all
• Focused on rewriting, not on inventing new claims
• Preserves technical faithfulness, citations, equations, and LaTeX commands
• Improves causal logic and academic tone without turning the text into generic "AI academic style"

Current Support

• Methodology

The current module rewrites methodology drafts into publication-ready prose while keeping the original technical content intact.

Workflow

1. Write a rough draft for one paper section.
2. Pass that draft to the skill.
3. Let the skill choose the matching section logic.
4. Review the rewritten result, back-translation, and modification log.

When used in chat, the default output format is:

1. ## Part 1 [LaTeX]
2. ## Part 2 [Translation]
3. ## Part 3 [Modification Log]

Example

Input draft:

First, we retrieve a set of demonstrations. Then the planner uses them to generate a figure description. Finally, the renderer converts the description into the final output.

Typical rewrite target:

To produce a figure-ready layout from an underspecified method description, the system first retrieves a small set of demonstrations that ground the downstream planning stage. The planner then converts these demonstrations and the source description into a structured figure specification, which is finally rendered into the output artifact.

The goal is not decorative wording. The goal is to make the design logic visible.

Installation

This repository ships the core prompt asset and reference material:

• SKILL.md is the primary skill prompt
• references/ contains section-specific rewrite guidance
• agents/ contains tool-facing metadata

Different tools manage skills differently. For Claude Code, Codex, Gemini CLI, Cursor, or other agent environments, use this repository as the source asset and register or adapt SKILL.md according to your local skill or prompt workflow.

If your tool supports repository-based skills, you can point it at this repository directly. If it does not, copy the prompt and references into that tool's preferred local skill location.

Repository Contents

Academic-Style-Rewriter/
├── SKILL.md
├── agents/
├── evals/
└── references/

Roadmap

• Methodology
• Abstract
• Introduction
• Related Work
• Experiments
• Discussion
• Conclusion
• Other section-specific modules

Contributing

Issues and pull requests are welcome, especially for:

• new section modules
• clearer rewrite heuristics
• evaluation examples
• tool-specific integration notes

License

MIT. See LICENSE.