wechat-style-profiler

Wechat Style Profiler

Goal

Turn “whether it reads like this author wrote it” from a subjective feeling into an executable, reusable, and iterable explicit asset.

Input requirements

1. Author name (required)
2. Platform type (WeChat official account, required)
3. Reference articles (at least 3, 5-10 recommended)
4. Article paths (optional)

Default article path suggestion:风格参考/{作者}/公众号/

Workflow

1. Collect author information and sample paths.
2. Read the reference articles and build a sample list and statistics.
3. First run scripts/build_style_profile.py to obtain the quantitative baseline.
4. Perform an in-depth analysis according to the 14-dimensional framework (see references/style-14d-framework.md）。
5. Separately summarize 标点符号偏好, clarifying the frequency, function, and prohibited scenarios of symbols such as dashes, quotation marks, question marks, exclamation marks, and ellipses.
6. Separately summarize 分块习惯, clarifying major headings, subheadings, numbered sections, hierarchy depth, and chunking density.
7. Separately summarize 段落配方, clarifying paragraph lengths, sentence count ratios, and trigger conditions for one-sentence paragraphs in different scenarios.
8. Separately summarize 叙述方法体系, clarifying the writing methods for conflict, progression, turning points, and closure.
9. Separately summarize 内容推进方式, clarifying how the author generally pushes the content forward, and how they move from phenomena to viewpoints and from cases to judgments.
10. Produce a first draft of the style profile and enter the user calibration loop.
11. Extract 3-5 Few-shot examples from the reference articles.
12. Generate an emotional curve and trigger word list.
13. Summarize a template library for introduction, development, turn, and conclusion.
14. Output the final style profile to a fixed file, and provide calling instructions for the writing and title skills.

User calibration rules

• User confirmation must be requested after the first draft of the profile.
• After the user points out deviations, targeted corrections must be made rather than a full rewrite.
• Complete at least 1 round of calibration; finalize the final version only after the user explicitly says it is “usable.”

14-dimensional analysis framework

Use the following four layers and a total of 14 dimensions (see references/style-14d-framework.md）：

1. Surface features (4 dimensions)
2. Structural features (3 dimensions)
3. Deep features (4 dimensions)
4. Unique markers (3 dimensions)

Output contract

Output in the following order:

1. 样本概况
2. 14维风格分析
3. 标点符号偏好
4. 分块习惯
5. 段落配方
6. 叙述方法体系
7. 内容推进方式
8. Few-shot 示例段落
9. 情绪曲线分析
10. 起承转合模板
11. 可复用 Prompt Block
12. 风格画像文件路径

Default fixed file

• If the user wants to form a long-term asset, by default also maintain a fixed DNA file.
• The current default file can be placed at: references/style-dna-default-template.md

Supported formats

• .md
• .txt
• .docx(extract text first, then analyze)

Finalized file

By default, the final profile is saved to: 风格参考/{作者}/风格画像.md

If the user does not specify an author directory, save it as a fallback to: /tmp/{author_slug}-style-profile.md

Quality red lines

• Only doing “vocabulary imitation” is not allowed; structure, rhythm, and argumentation must be covered.
• Skipping punctuation preference analysis is not allowed.
• Skipping chunking habit analysis is not allowed.
• Skipping paragraph recipes and the narrative method system is not allowed.
• It is not allowed to skip the analysis of content progression methods.
• It is not allowed to skip user calibration and finalize directly.
• It is not allowed to output vague conclusions; examples, proportions, or frequencies must be provided.
• It is not allowed to generate style judgments inconsistent with the sample evidence.

Script

Quantitative analysis script:

python3 scripts/build_style_profile.py --input-dir /path/to/samples --output /path/to/style_profile.json

Delivery template

The final document structure uses: references/style-profile-template.md