| name | learn-writing-style |
| description | Analyze writing style from user-provided example documents (PDF, DOCX, TXT, MD) and generate personalized writing rules. Use when the user wants to: (1) Learn or extract writing style from example papers/articles, (2) Create personalized writing rules for Claude to follow, (3) Build a writing-style preference profile from sample documents, (4) Apply a specific academic/professional writing style consistently. Triggers on keywords: "writing style", "learn style", "style analysis", "preference file", "writing rules", "write like", "模仿风格".
|
Writing Style Analysis
Workflow
This skill extracts writing style from user-provided documents through a 5-step pipeline.
Step 1: Document Preprocessing
- List all files in the user-specified directory.
- Check file types. Supported:
.txt, .md, .docx, .pdf.
- For
.pdf files, use mcp__mineru__parse_documents to extract text. Save output as {original_name}.parsed.md in the same directory.
- For
.docx files, extract text and save as {original_name}.parsed.md.
- Generate an
INDEX.md in the directory root with:
- File list (original + parsed)
- Document types and word counts
- Brief content summary (use a quick scan or
head)
Step 2: Style Analysis (SubAgent)
Launch the pre-defined style-analyzer agent, specifying the directory:
Read INDEX.md and all parsed documents in {directory},
analyze writing style across 12 dimensions, and save the report as {directory}/ANALYSIS.md.
Step 3: User Preference Elicitation
Read ANALYSIS.md, then use AskUserQuestion to confirm preferences:
Round 1: Core style choices
- Which of the observed patterns do you want to keep vs. discard?
- Do you want the style to be more formal or more accessible than the examples?
- Any specific habits from the examples you explicitly want to avoid?
Round 2: Detailed adjustments
- Preferred sentence length range?
- Active vs. passive voice preference?
- First-person usage (I/we) — allowed, restricted, or preferred?
- Hedging style (strong claims vs. cautious language)?
- Citation density preference?
Round 3: Scope confirmation
- Should these rules apply to all writing or specific genres (papers, emails, reports)?
- Any additional constraints (e.g., journal-specific guidelines, word limits)?
Step 4: Generate PREFERENCE.md
Synthesize the analysis and user feedback into a structured PREFERENCE.md (uppercase filename) in the directory root. Format:
# Writing Style Preferences
## Overview
[2-3 sentence summary of the target style]
## Dimensions
### 1. Sentence Structure
[Specific rules and parameters]
### 2. Vocabulary Features
[...]
### 3. Paragraph Organization
[...]
[... continue for all 12 dimensions ...]
## Global Rules
- [Rule 1]
- [Rule 2]
## Anti-Patterns (Avoid)
- [Anti-pattern 1]
- [Anti-pattern 2]
Step 5: Rules Generation (SubAgent)
Launch the pre-defined rules-generator agent, specifying the directory:
Read {directory}/PREFERENCE.md,
convert preferences into actionable rules, and save to .claude/rules/artical-writing-style.md in the project root.
Step 6: Global Application (Optional)
If the user wants global application:
mkdir -p ~/.claude/rules && cp .claude/rules/artical-writing-style.md ~/.claude/rules/artical-writing-style.md
Confirm with the user before executing.
Resources
| File | Purpose | When to Load |
|---|
references/dimensions.md | Detailed definitions and analysis criteria for the 12 writing style dimensions | When the style-analyzer SubAgent needs the full dimension descriptions |
references/INDEX.md | Index of example documents with file list, types, word counts, and brief summaries | Generated during Step 1; used by the style-analyzer to understand the document collection |
references/PREFERENCE.md | Structured writing style preferences synthesized from analysis and user feedback | Generated during Step 4; read by the rules-generator in Step 5 |
