Literary Style Skill

Section 1: Analyze Mode

Given a text sample, extract a comprehensive style profile. The profile is structured so it can be used directly as a target for Transform mode or as an input to Combine mode.

Minimum Sample Guidance

State the confidence level based on sample size. Never claim high confidence from a short sample.

The 8-Step Extraction Protocol

Step 1: First Impression

Read the full sample. Write 2-3 sentences capturing the overall feel without technical analysis. This anchors the detailed work and serves as a sanity check on the final profile. What does the prose feel like? What is the dominant quality a reader would notice?

Step 2: Statistical Features

Estimate (do NOT claim exact counts — describe qualitatively anchored to ranges):

• Sentence length tendency: predominantly short (5-12 words), medium (12-25), long (25+), or mixed
• Sentence length variation: uniform, moderately varied, or highly varied
• Sentence type mix: declarative-heavy, question-rich, imperative, exclamatory, or balanced
• Paragraph length tendency: short (1-3 sentences), medium (4-7), long (8+), or varied
• Pronoun profile: first-person dominant, second-person (direct address), third-person, or impersonal
• Active vs. passive tendency: predominantly active, mixed, or passive-heavy
• Contraction usage: frequent, occasional, or absent

Step 3: Dimensional Rating

Rate each dimension 1-5 with a one-sentence justification citing specific textual evidence. Use the dimension definitions from references/dimensions.md. The 17 dimensions:

Flag any dimension where the sample provides insufficient evidence.

Step 4: Device Inventory

Scan for rhetorical and literary devices (see references/devices.md for the full catalog). For each device found, note:

• Device name
• Frequency: rare / occasional / regular / pervasive
• Example: one instance from the text
• Effect: what it contributes to the style

Group by family: sound devices, repetition devices, figurative language, syntactic devices, narrative devices, tonal devices.

Step 5: Characteristic Patterns

Identify 3-7 specific, concrete habits that distinguish this style — the "signature moves" that go beyond dimensional ratings:

• Characteristic sentence openers (e.g., "tends to open with temporal clauses")
• Paragraph-ending patterns (e.g., "ends paragraphs with a single-word sentence")
• Punctuation habits (em-dash heavy, semicolon-free, ellipsis-laden)
• Transitional strategies (explicit connectives, white space, associative leaps)
• Showing-to-telling ratio
• Distinctive word-class preferences (favors gerunds, avoids adverbs, heavy nominalization)
• Structural tics (always begins with a question, alternates long/short paragraphs)

These are what make a profile specific rather than generic.

Step 6: Negative Space

List 3-5 things the style conspicuously avoids. What devices, constructions, vocabulary, or patterns are absent despite being common in the genre? Negative space is essential for generation — knowing what NOT to do is as important as knowing what to do.

Step 7: Exemplar Selection

Choose 2-3 passages (1-3 sentences each) that are maximally representative of the style. These serve as anchors for generation. Select passages that exhibit the most characteristic features simultaneously.

Step 8: Synthesis — The Style Profile

Compile everything into this structured format:

## Style Profile: [Name or "Extracted from [source]"]

### First Impression
[2-3 sentences from Step 1]

### Dimensional Profile
| Dimension    | Rating | Justification                |
| ------------ | ------ | ---------------------------- |
| Register     | N      | [one sentence with evidence] |
| Tone         | N      | ...                          |
| [... all 17] |        |                              |

### Statistical Features
- Sentence length: [tendency + variation]
- Sentence types: [distribution]
- Paragraph length: [tendency]
- Pronouns: [profile]
- Voice: [active/passive tendency]
- Contractions: [frequency]

### Device Inventory
| Device | Frequency | Example   | Effect         |
| ------ | --------- | --------- | -------------- |
| [name] | [freq]    | "[quote]" | [what it does] |

### Characteristic Patterns
1. [pattern with example]
2. [pattern with example]
...

### Negative Space (What This Style Avoids)
1. [avoidance]
2. [avoidance]
...

### Exemplar Passages
> [passage 1]

> [passage 2]

### Generation Guidance
To write in this style:
- [prescriptive instruction derived from the profile]
- [prescriptive instruction]
- [constraint — what to avoid]
...

### Confidence: [high/medium/low]
[Note sample size and any caveats]

The Generation Guidance section is the bridge from analysis to production. Translate every descriptive finding into a prescriptive instruction. Key translations:

Section 2: Transform Mode

Transform input text into a target style. The target can be:

1. A named style from the style index (references/style-index.md)
2. An extracted profile produced by Analyze mode (Section 1)
3. A blended profile produced by Combine mode (Section 3)

Transformation Algorithm

Step 1: Identify the Target

• Named style: Read the target entry from references/style-index.md. Each entry has a dimensional profile, constraints, and pitfalls.
• Extracted profile: Use the profile directly — the Generation Guidance section is your primary instruction set.
• Blended profile: Use the resolved blend profile from Section 3.

Step 2: Analyze the Input

Run a lightweight version of the Analyze protocol (Steps 1, 3, and 4 — first impression, dimensional rating, and device inventory) on the input text. This tells you where the input currently sits.

Step 3: Compute the Delta

Compare input dimensions to target dimensions. Identify:

• Dimensions that need to shift — and by how much (a shift of 1 is subtle; 3+ is a fundamental rewrite)
• Dimensions that already match — preserve these; don't break what works
• Devices to add or remove
• Constraints to apply (from the target style's entry)

Step 4: Transform

Rewrite the text. Apply changes in this order (structural before decorative):

1. Paragraph structure — reorganize if the target demands a different structure (e.g., inverted pyramid for journalism, IMRaD for academic)
2. Sentence structure — restructure for target complexity and variety
3. Register and vocabulary — shift word choice, formality, contractions
4. Voice and perspective — change pronoun usage, active/passive balance
5. Devices — add rhetorical devices characteristic of the target; remove devices that don't belong
6. Rhythm and sound — adjust sentence length variation, add sound patterns if the target demands them
7. Imagery and emotional valence — adjust figurative language density and emotional intensity
8. Characteristic patterns — apply the target's signature moves

Step 5: Verify

Check the output against:

• The target style's dimensional profile — does each dimension land in the right range?
• The target's common pitfalls — are you falling into any?
• The target's constraints — are hard constraints satisfied?
• Factual accuracy — has the meaning been preserved?
• Negative space — have you avoided what the target style avoids?

Key Dimension Reference

Domain-Specific Constraints

Quick Reference: Style by Purpose

Section 3: Combine Mode

Blend two or more style profiles into a coherent combined target. Use when the user requests hybrid styles: "write this like a legal document but make it funny", "academic tone but accessible", "70% Hemingway, 30% Didion", "formal structure with poetic vocabulary."

Step 1: Identify the Source Styles

Each source can be:

• A named style from references/style-index.md
• An extracted profile (from Analyze mode)
• A previously combined profile

Step 2: Choose the Blending Strategy

Four strategies, each suited to different requests:

Default: Primary/Modifier. It produces the most coherent results. Use Proportional only for subtle shifts between nearby styles. Use Layer-Based when the user explicitly assigns styles to different text aspects. Use Dominant Trait when the user names specific qualities to borrow.

Step 3: Resolve Dimension Conflicts

When source styles demand opposite values on a dimension, resolve using this priority hierarchy (structural wins over decorative):

Structural dimensions (always from primary style):

1. Paragraph & structural patterns — macro-architecture must be coherent
2. Register — most perceptible signal; inconsistency here is immediately jarring
3. Sentence structure — tightly coupled with rhythm
4. Information density — determines reader effort
5. Vocabulary — must be consistent with register

Blendable dimensions (interpolate or assign per layer): 6. Audience relationship — can soften primary's stance 7. Syntactic patterns — can borrow specific devices 8. Specificity — can serve either style

Decorative dimensions (freely borrowable from any source): 9. Emotional valence — independent; easy to borrow 10. Sound patterns — independent; easy to overlay 11. Imagery — decorative layer 12. Tone sub-dimensions (warmth, playfulness) — combinable with nearly any structure 13. Rhetorical devices — can add selectively

Dimension Independence Reference

Step 4: Build the Combined Profile

Produce a resolved style profile in the same format as Analyze mode output. For each dimension:

• State the resolved value (1-5)
• State which source style it came from and why
• Note any deliberate tension (where the mismatch IS the point)

Include a Generation Guidance section that synthesizes instructions from all sources, with conflict resolutions already applied.

Step 5: Validate the Blend

After writing, check for incoherence markers — signs the blend failed:

Combination Recipe Format

For complex blends, produce a recipe before writing:

## Combination Recipe: [Name]

### Strategy: [primary_modifier / layer_based / proportional / dominant_trait]

### Sources
- Primary: [style name or profile reference]
- Modifier(s): [style name or profile, with specific features to borrow]

### Resolved Dimensional Profile
| Dimension    | Value | Source   | Notes    |
| ------------ | ----- | -------- | -------- |
| Register     | N     | Primary  | [reason] |
| Tone         | N     | Modifier | [reason] |
| [... all 17] |       |          |          |

### Conflict Resolutions
- [Dimension]: [How resolved and why]

### Generation Guidance
- [Merged instructions from all sources]
- [Specific constraints]
- [What to avoid from each source]

### Devices to Include
- [From primary: ...]
- [From modifier: ...]

### Devices to Avoid
- [Incompatible devices]

Section 4: Reference Materials

The following reference files provide the detailed knowledge backing this skill:

How to Use References

• For Analyze mode: Read dimensions.md for the rating framework and devices.md for the device catalog
• For Transform mode: Read the target style entry in style-index.md, then dimensions.md for the shift delta
• For Combine mode: Read dimensions-full.md for the co-variance matrix (which dimensions couple), then both source entries in style-index.md

Named Style Index Domains

Common Mistakes (All Modes)

When NOT to Transform or Combine

• Factual accuracy — never sacrifice accuracy for style
• Technical precision — in legal/technical/scientific styles, precision outranks elegance
• Attribution — do not fabricate quotes, citations, or sources
• Cultural sensitivity — when writing in culturally-rooted styles (e.g., blues, spoken word, ghazal), respect the tradition
• Meaning preservation — a style transform should not change WHAT is said, only HOW