writing-dna-skill git:20260627.88b9941A

Immutable. This exact content never changes and is served at /api/v1/blob/917beb6ccbe696b5.

---
name: writing-dna-skill
description: Distill reusable writing style rules from a corpus of complete articles. Use when the user wants to analyze 30+ complete articles and extract language DNA, article structures, topic logic, material strategy, cognitive frame, visual style, and a Writing-DNA.md document for style-consistent writing.
---

# 写作蒸馏器

Use this skill to distill a reusable Writing DNA from a corpus of complete articles. Do not summarize the articles; extract the rules that make the writing style repeatable.

## Corpus Requirements

Require enough source material before claiming a reliable style:

- Use at least 30 complete articles.
- Prefer 80-100 complete articles for stronger results.
- Prefer articles longer than 1000 Chinese characters or words each.
- Use `.md` or `.txt` files in `raw/` or `raw-corpus/`.
- Do not place unauthorized source articles in a public repository.

If the corpus is small, explain that the output is only a format or workflow demo, not a reliable style distillation.

## Workflow

1. Inspect the corpus and create `_meta/` records for each article: title, date, author/account, article type, topic tags, hook type, structure pattern, source types, word count, and notable traits.
2. Extract L1 surface language: frequent nouns, verbs, adverbs, sentence length, short/long sentence ratio, paragraph rhythm, punctuation, title style, and mixed-language habits. Write `语言DNA.md`.
3. Extract L2 article structure: opening hook, first turn, body architecture, section rhythm, transition style, and ending pattern. Write `文章结构模板.md`.
4. Extract L3-L5 thinking rules: topic selection logic, timing, angles, material strategy, preferred authority or evidence, recurring assumptions, values, and core propositions. Write `写作视角与认知框架.md`.
5. Extract L6 visual style when the source contains images or formatting: image type, placement rhythm, screenshot/evidence role, typography hierarchy, emphasis, color, layout density, and text-image collaboration. Write `视觉风格指南.md`.
6. Integrate the outputs into `Writing-DNA.md`, keeping it concise enough for an agent to reread before writing.

For image-heavy articles, open and inspect images. Screenshots, comments, charts, and conversation images may carry evidence that is not present in the text.

## Output Files

Produce these files in the corpus directory:

```text
_meta/
语言DNA.md
文章结构模板.md
写作视角与认知框架.md
视觉风格指南.md
Writing-DNA.md
```

Use `templates/author-corpus/` as the recommended directory template. Read `写作蒸馏器.skill.md` for the full public-facing workflow when detailed instructions are needed.

## Writing With A Distilled Style

Before writing in a distilled style, reread all style documents, especially `Writing-DNA.md`, `语言DNA.md`, `文章结构模板.md`, `写作视角与认知框架.md`, and `视觉风格指南.md`. Then follow the target style while avoiding misleading impersonation.