writing-core

Core Writing Standards

This skill is used to control the quality of paper writing, with a focus on natural expression in Chinese academic papers, de-AI-ification rewriting, and submittable Markdown formatting.

1. De-AI-ification Language Standards

1.1 Prohibited Expressions

Type	Prohibited Words/Sentences
Mechanical transition words	First, second, finally, in addition, moreover, next, in summary
Empty emphatic sentences	It is worth noting that, it should be pointed out that, it is important that, it must be emphasized that, obviously
Empty modifiers	Very, extremely, quite, rather (when unsupported by data)
Subjective lead-in sentences	I think, I feel, my personal view is (prohibited in the main text of papers)

1.2 Recommended Expressions

1. Use semantic cohesion instead of template cohesion
2. Use data and facts instead of adjectives
3. Alternate long and short sentences; avoid consecutive sentences of equal length
4. Use objective subjects such as "this paper," "the experimental results show," and "Table X shows," but avoid repeating the same template consecutively
5. Retain necessary qualifications, transitions, and a small amount of repetition so that the paragraph reads like the Chinese main text of a normal researcher

1.3 Syntax and Information Density

1. When converting lists into paragraphs, supply the missing subjects, predicates, and connectives
2. Develop one main relationship per sentence; when necessary, retain conditions, scope, and explanations, and do not force it into short clauses
3. Retain methods, conditions, objects, and data; do not use vague expressions such as "many" or "a relatively large improvement"
4. De-AI-ification does not mean compression. If shortening causes the loss of data definitions, experimental boundaries, evaluation objects, or conclusion limitations, the necessary explanations should be restored

1.4 Principles of Natural Writing in Chinese Journal Papers

Chinese journal papers do not pursue extreme concision in every sentence. A more natural way of writing is usually to first explain the research object, data scope, or problem background, and then describe the processing method, result phenomena, and boundaries of judgment. Sentences can be slightly longer, and paragraphs may contain necessary transitions, qualifications, and a small amount of repetition, as long as the subject is clear, the relationships are clear, and readers can follow the research process.

The focus of de-AI-ification is to reduce templating, translationese, overgeneralization, and empty elevation, rather than deleting all modifiers. Components that can explain the object, time range, sample definition, methodological conditions, indicator meanings, experimental boundaries, and causal relationships should be retained. Do not turn paragraphs into mechanical sentence groups of "background-method-result-significance," and do not make every sentence present a functional pattern of "object-action-conclusion."

When users give feedback such as "too AI," "too concise," or "sounds like a translation," first check whether there is excessive compression, English word order, templated summarization, reviewer-response-style expression, or empty elevation. The order of handling is to restore the necessary information, adjust it to natural Chinese word order, and then restrain overly strong judgments. Do not continue to simply delete words.

1.5 Common AI-like Symptoms and Handling

Symptom	Handling Method
Sentence patterns sound like direct English translation	Change to common Chinese word order: first explain the object and phenomenon, then give the judgment
Paragraphs are overly templated	Remove the fixed "background-method-result-significance" rhythm and retain explanatory and transitional sentences
Information decreases after rewriting	Restore the research object, data scope, methodological conditions, indicator definitions, and conclusion boundaries
Reviewer-response tone	Change "this indicator reflects" and "cannot be understood as" into narrative body text, such as "this paper takes ... as a reference"
Vague elevation	Replace generalized words such as "significant", "key", and "important significance" with specific indicators, phenomena, or qualifications

II. Output Formatting Standards

2.1 Markdown Body Text Standards

1. By default, do not use bold or italics in the body text
2. There must be a blank line between paragraphs
3. Body paragraphs should prioritize continuous narration and should not stack viewpoints with bullet points
4. Try to maintain a single central viewpoint within the same paragraph
5. Avoid splitting a complete viewpoint into too many short paragraphs

2.2 Scenarios Where Lists Are Allowed

Lists may be used only in the following scenarios:

• Planning documents (plan/*.md）
• Checklists
• Parameter configuration
• Operating steps

By default, do not use lists in the body of a paper.

III. Paragraph Construction Rules

A standard paragraph contains:

1. Topic sentence: the core conclusion of this paragraph
2. Supporting sentence: basis, evidence, explanation
3. Concluding sentence: transition or summary

Recommended length:

• Chinese body text: 150-300 characters
• English body text: 100-200 words

IV. Rules for Converting Lists to Paragraphs

Incorrect writing

本研究贡献如下：
- 提出新方法
- 完成自动化流程
- 验证有效性

Recommended writing

本研究提出了一种新方法，并将其整合为可执行的自动化流程。
实验结果显示，该方法在目标任务上具有稳定增益，验证了其可行性与应用价值。

V. Citations and Facts

VI. Three Rounds of Quality Checks

First round: structural check

• Whether the body text has been turned into a list
• Whether paragraphs revolve around a single center
• Whether the logic of chapters and sections is continuous
• Whether the body text has been written as mechanical template paragraphs of "background-method-results-significance"

Second round: language check

• Whether prohibited transitional words appear
• Whether prohibited emphatic sentence patterns appear
• Whether there is an accumulation of uninformative adjectives
• Whether subjective expressions appear
• Whether there is translationese, English word order, or reviewer-response-style expression
• Whether the research object, data scope, method conditions, indicator definitions, or conclusion boundaries are lost due to excessive compression
• Whether judgments such as "improvement", "effective", and "stable" are supported by specific data and phenomena

Third round: formatting check

• Whether there is meaningless bolding
• Whether a blank line is uniformly left between paragraphs
• Whether punctuation and spacing between Chinese and English are consistent

VII. Style Self-Check Script

After writing is completed, run the style check script:

macOS/Linux：

bash research-writing-skill/scripts/style_check.sh <文件.md>

Windows PowerShell：

powershell -ExecutionPolicy Bypass -File research-writing-skill/scripts/style_check.ps1 -FilePath <文件.md>

VIII. Execution Constraints

1. Before starting a long task, you must read or create plan/
2. After each writing completion, update plan/progress.md and plan/notes.md
3. If the user explicitly requests preserving a specific style, the user's request takes priority