Style Extractor

Style Extractor

Extract reusable writing style guides from any source text. Produces four standardized deliverables that capture general patterns — not copied phrases — so the style can be replicated for any topic.

When to Use

• User asks to extract or analyze a writing style from a book, article, or document
• User wants to replicate an author's voice for new content
• User wants a style guide, voice card, or writing rubric from a source text
• User says "extract style from...", "analyze the writing style of...", or "capture this voice"

Prerequisites

MCP tools required:

• mcp__fuzzy-search — PDF page counting, outline extraction, and page-level reading
• mcp__plugin_sequential-thinking_think__sequentialthinking — Structured analysis across dimensions

Output location: writing-styles/<style-name>/ in the project root. See writing-styles/README.md for the collection structure and writing-styles/_template/ for deliverable templates.

Workflow

Phase 1: Source Analysis

Identify the source text and assess its structure before reading.

For PDFs:

1. Get page count: mcp__fuzzy-search__get_pdf_page_count
2. Get outline/TOC: mcp__fuzzy-search__get_pdf_outline
3. Understand document structure before sampling

Select 5–7 sampling points distributed across the text:

Read representative passages using mcp__fuzzy-search__extract_pdf_pages or the Read tool. Aim for ~50–80 pages total across all samples.

For non-PDF files: Use the Read tool directly. Apply the same sampling distribution across the document's length.

Phase 2: Dimension Extraction

Analyze 17 style dimensions across all sampled passages. Use mcp__plugin_sequential-thinking_think__sequentialthinking to organize observations systematically.

Focus on patterns, not content. A style dimension describes HOW the author writes, never WHAT they write about.

For detailed guidance on analyzing each dimension, see references/style-dimensions.md.

Phase 2b: Narrative Construction (Fiction Only)

When the source text is fiction or narrative-driven writing, analyze 10 narrative dimensions in addition to the 17 surface style dimensions. Surface style captures how sentences read; narrative construction captures how stories are built. These are orthogonal layers — capturing both produces a significantly more complete style profile.

Pay special attention to where the author's choices diverge from AI narrative defaults (e.g., nonlinear time, morally ambivalent protagonists, implicit themes). These divergences are the most valuable and hardest-to-replicate aspects of the style.

For detailed guidance, see references/narrative-dimensions.md.

Phase 3: Synthesis

Generate four deliverables in writing-styles/<style-name>/:

Every deliverable MUST include 2–3 examples of neutral text rewritten in the extracted style. These examples demonstrate the style applied to content unrelated to the source.

For fiction sources: Each deliverable should include a "Narrative Construction" section alongside the surface style analysis. The voice card should list the author's distinctive narrative patterns (e.g., "nonlinear time, morally ambivalent protagonists, implicit themes"). The do/don't list should include narrative-level items (e.g., "Do: use subplots that contrast the main arc" / "Don't: resolve through protagonist epiphany"). The rubric should include at least one narrative construction dimension.

Use writing-styles/_template/ as the structural starting point for each file.

Phase 4: Output

1. Create the folder structure:

   writing-styles/<style-name>/
   ├── style/
   │   ├── full-style-guide.md
   │   ├── voice-card.md
   │   └── do-dont.md
   └── evals/
       └── style-rubric.md
   

2. Write all four deliverables.

3. Verify the structure matches the template layout in writing-styles/_template/.

Naming Convention

Use kebab-case: <author-or-title>

Best Practices

• Sample broadly. Pull from beginning, middle, AND end — styles evolve across a text.
• Prioritize consistency. Features that appear across multiple samples define the real style; one-off features are noise.
• Track adaptation. Note how style shifts for different material types, difficulty levels, or audiences. This is dimension 17 and often the most valuable.
• Extract patterns, not phrases. Never copy distinctive phrases as style rules. Identify the general pattern the phrase exemplifies.
• Use agent teams for large texts. Assign different sampling regions to different researchers, then synthesize.
• Cross-validate. If a pattern appears in only one sample, it may be a local anomaly — flag it but don't center the guide on it.

Anti-Patterns

1. Extracting content as style. Topic-specific vocabulary is content, not style. "Uses the word 'eigenvalue'" is content; "introduces technical terms with an informal gloss" is style.
2. Projecting absent features. Don't claim humor if there is none. Don't invent warmth in clinical prose. Report what IS there.
3. Sampling from one section only. A single chapter gives a biased picture. The five-point sampling is non-negotiable.
4. Copying signature phrases. "Always use the phrase 'Let us consider...'" is wrong. "Opens explorations with an inclusive invitation" is right.
5. Domain-locked guides. A style guide that only works for the original domain has failed. The deliverables must transfer to any topic.

References

For detailed analysis guidance and deliverable templates:

• references/style-dimensions.md — Detailed analysis guide for each of the 17 surface style dimensions
• references/narrative-dimensions.md — Analysis guide for 10 narrative construction dimensions (fiction/creative writing only)
• references/deliverable-templates.md — Templates for all four deliverables with embedded guidance and worked examples