---

name: voice-extractor description: Analyze existing content to extract voice patterns for consistent brand messaging allowed-tools:

• Read
• WebFetch
• Task
• Glob
• Grep invocation: user

---

Voice Extractor

Use this skill to analyze a user's existing content (GitHub repos, blog posts, social profiles) and extract their unique voice patterns. This ensures generated content matches their authentic style.

Why Voice Matters

Generic AI copy sounds like... generic AI copy. To create content that resonates:

1. Match the creator's existing voice
2. Maintain consistency across platforms
3. Sound authentic, not robotic

What to Analyze

Source Types (in priority order)

1. GitHub README files - Technical communication style
2. Blog posts/articles - Long-form voice
3. Twitter/X posts - Casual voice, hooks
4. LinkedIn posts - Professional voice
5. About pages - Self-description patterns
6. Commit messages - Micro-communication style

Voice Dimensions to Extract

1. Tone

• Formal ←→ Casual
• Serious ←→ Playful
• Authoritative ←→ Approachable
• Technical ←→ Accessible

2. Vocabulary

• Jargon level (heavy technical vs simplified)
• Industry-specific terms used
• Avoided words/phrases
• Signature phrases

3. Sentence Structure

• Short punchy ←→ Long flowing
• Simple ←→ Complex
• Active ←→ Passive
• Questions ←→ Statements

4. Personality Traits

• Humor style (if any)
• Use of emojis
• Use of first person (I/we)
• Directness level

5. Patterns

• How they open pieces
• How they structure arguments
• How they use examples
• How they close/CTA

Extraction Process

Step 1: Gather Content

Collect 3-5 pieces of existing content from the user.

Step 2: Analyze Each Dimension

Rate each dimension and provide specific examples.

Step 3: Identify Patterns

Find recurring elements across samples.

Step 4: Create Voice Profile

Compile into a reusable voice guide.

Output Format

Copy## Voice Profile

### Overview
**Voice in 3 words**: [Word 1], [Word 2], [Word 3]
**Overall tone**: [Description]

### Tone Spectrum
- Formality: [Rating 1-5, with 1 being very casual, 5 being very formal]
- Technical depth: [Rating 1-5]
- Humor level: [Rating 1-5]
- Directness: [Rating 1-5]

### Vocabulary Patterns

**Signature phrases** (use these):
- "[Phrase 1]"
- "[Phrase 2]"
- "[Phrase 3]"

**Technical terms used**:
- [Term 1] - [how they explain it]
- [Term 2] - [how they explain it]

**Words to avoid**:
- [Word 1] - they never use this
- [Word 2] - they avoid this

### Sentence Structure

**Typical opening patterns**:
- "[Pattern 1]"
- "[Pattern 2]"

**Typical sentence length**: [Short/Medium/Long]
**Paragraph style**: [Description]

### Personality Elements

- **First person**: [I/We/Both/Neither]
- **Emojis**: [Heavy/Light/None]
- **Questions**: [Frequent/Occasional/Rare]
- **Humor style**: [Description]

### Platform Adaptations

| Platform | Tone Adjustment | Length | Special Notes |
|----------|-----------------|--------|---------------|
| GitHub | [Adjustment] | [Length] | [Notes] |
| Twitter/X | [Adjustment] | [Length] | [Notes] |
| LinkedIn | [Adjustment] | [Length] | [Notes] |
| Blog | [Adjustment] | [Length] | [Notes] |

### Example Voice Samples

**GitHub README style**:
> [Example sentence in their GitHub voice]

**Social media style**:
> [Example sentence in their social voice]

**Professional/LinkedIn style**:
> [Example sentence in their professional voice]

### Generation Guidelines

When generating content for this user:
1. [Specific instruction 1]
2. [Specific instruction 2]
3. [Specific instruction 3]
4. [Specific instruction 4]

Example Analysis

Input: Developer's GitHub README

Sample text:

"This tool does one thing well: it makes deploys boring. No drama, no 3am pages. Just push and forget. Built by someone who's been woken up by PagerDuty way too many times."

Extracted Voice Profile

Voice in 3 words: Direct, Practical, Relatable

Tone:

• Formality: 2/5 (casual)
• Technical depth: 3/5 (moderate)
• Humor level: 3/5 (dry humor)
• Directness: 5/5 (very direct)

Signature patterns:

• Short, punchy sentences
• Personal anecdotes ("Built by someone who...")
• Focus on pain points
• Understatement humor ("makes deploys boring")

Generation guidelines:

1. Keep sentences short and punchy
2. Reference real developer pain points
3. Use dry humor, avoid try-hard jokes
4. Stay practical, not aspirational
5. Use first person sparingly but authentically

Handling Multiple Sources

When analyzing multiple content sources:

1. Look for consistency - What's the same across all sources?
2. Note platform variations - How do they adapt for different contexts?
3. Weight by recency - Newer content shows current voice
4. Weight by quality - Their best content shows intentional voice

Common Voice Archetypes