---
name: voiceprint
description: Build a portable Voiceprint - a writer's voice profile captured from real writing samples - covering tone, sentence rhythm, structure, vocabulary, and signature rhetorical moves, then use it to generate or rewrite content that sounds like that person. Use this whenever the user wants to capture someone's writing voice, build a style/voice profile, says "voiceprint," "capture my voice," "build a voice profile," uploads writing samples to be analyzed for tone and style, or wants new content generated or rewritten to match a specific person's voice. Also trigger on BEGIN, VOICEPRINT, or UPDATE commands mid-workflow, and whenever the user references an existing VOICEPRINT.md file.
---

# Voiceprint Instructions

Let's take a Voiceprint.

## Human Operator Workflow

### Upload Your Writing Samples

Upload your writing samples.

You can paste them one at a time, or all at once.

### Uploading Samples One at a Time

If you paste them one at a time:

- Agent will reply only with CONTINUE after each sample.
- When done uploading, write VOICEPRINT.
- Then I will analyze your writing and create your Voiceprint.

## Agent Workflow

### Initial Response

After reading this prompt, respond with:

Ready to make your Voiceprint. Provide writing samples, one at a time.

## Voiceprint Objective

### Goal

Understand how the writer thinks, and communicates, from analysing their samples. Use this insight to replicate content in their voice.

### Patterns to Match

Match their:

- Tone and relationship with the reader
- Sentence rhythm and paragraph structure
- Vocabulary and level of complexity
- Openings, transitions, and conclusions
- Use of stories, examples, metaphors, and evidence
- Recurring rhetorical patterns

Focus on patterns, not exact wording.

## Sample Collection

### Sample Requirements

Require at least three representative writing samples.

Prefer samples written by the user that genuinely reflect their voice. Treat heavily edited, ghostwritten, or AI-generated material as weaker evidence.

### One-at-a-Time Submission Workflow

If the user says BEGIN and submits samples one at a time:

- Reply only CONTINUE after each sample
- Do not critique, summarize, edit, or analyze
- Begin analysis when the user says VOICEPRINT

### All-at-Once Submission Workflow

If all samples are provided at once, proceed directly to analysis.

If you receive multiple samples, do not assume that all samples have been received. Offer a choice of a) More? or b) Voiceprint?

## Evidence Analysis

### Analysis Method

Look for consistent patterns across all samples. Avoid overgeneralizing from a single passage.

### Patterns to Identify

Identify:

1. Overall voice and tone
2. Sentence length, variation, and cadence
3. Paragraph length, density, and pacing
4. Typical openings and conclusions
5. Point of view and reader relationship
6. Vocabulary, jargon, humour, and repeated expressions
7. Use of stories, examples, metaphors, frameworks, and evidence
8. Formatting habits
9. Signature rhetorical patterns

Describe observable behaviour, not vague traits.

### Description Standard

#### Prefer:

Opens with a direct claim, develops it through an example, and ends with a practical takeaway.

#### Avoid:

Professional, engaging, authentic.

## Clarifying Gaps

### Question Limits

Ask no more than four questions, and only when needed.

### Suggested Questions

Use these when relevant:

1. Who do you usually write for, and what should they think, feel, or do?
2. Which sample sounds most like you, and which sounds least like you?
3. What should writing in your voice never sound like?
4. Which format should be the default: post, email, article, script, report, or another format?

Do not delay the process for minor uncertainty.

## Voiceprint Creation

### Output Requirements

Keep the final Voiceprint under 8000 characters.

Use this structure:

# Voiceprint

## Voice Summary

[One concise paragraph.]

## Signature Traits

- **[Trait]:** [Observable behaviour.]
- **[Trait]:** [Observable behaviour.]
- **[Trait]:** [Observable behaviour.]
- **[Trait]:** [Observable behaviour.]

## Rhythm and Structure

- **Sentences:** [...]
- **Paragraphs:** [...]
- **Openings:** [...]
- **Conclusions:** [...]
- **Formatting:** [...]

## Signature Moves

1. **[Pattern]:** [Sequence and purpose.]
2. **[Pattern]:** [Sequence and purpose.]
3. **[Pattern]:** [Sequence and purpose.]

## Language

- **Use:** [...]
- **Avoid:** [...]
- **Jargon:** [...]
- **Humour:** [...]

## Reader Relationship

[How the writer establishes authority, addresses the reader, and expresses opinions.]

## Format Notes

- **Posts:** [...]
- **Emails:** [...]
- **Long-form:** [...]
- **Scripts or reports:** [...]

## Confidence

- **Strong evidence:** [...]
- **Tentative inference:** [...]
- **User-stated preference:** [...]

### Section Selection

Include only relevant sections.

### File Handling

If file storage is available, save as VOICEPRINT.md. Otherwise, return it in a copy-ready block.

## Voice Calibration

### First Writing Task

For the first writing task, create three versions:

1. Closest Match - faithful to the Voiceprint
2. More Restrained - slightly less distinctive
3. More Amplified - slightly more distinctive without exaggeration

### Calibration Question

Ask:

Which version sounds most like you, and what would make it closer?

Use the answer to refine the Voiceprint.

### Post-Calibration Default

After calibration, default to one polished version unless asked otherwise.

## Content Generation

### Generation Rules

When creating content:

1. Follow the Voiceprint
2. Adapt to audience and format
3. Preserve accuracy and intent
4. Reuse structural patterns without repetition
5. Maintain natural variation
6. Avoid copying phrases from samples
7. Do not force stylistic elements
8. Prioritize clarity over imitation

## Content Rewriting

### Rewriting Rules

When rewriting:

- Preserve meaning, facts, and intent
- Improve voice consistency, structure, and clarity
- Do not introduce unsupported claims
- Do not distort meaning for stylistic effect

## Voiceprint Updates

### Update Method

Treat feedback and new samples as additional evidence.

When the user says UPDATE:

1. Compare new input with the current profile
2. Identify what it confirms, adds, or contradicts
3. Revise only affected sections
4. Separate global vs format-specific changes
5. Avoid overcorrecting from a single example

## Final Quality Check

### Delivery Checklist

Before delivering content, confirm:

- Structure matches the writer's patterns
- Tone fits the audience and format
- Signature moves are used selectively
- No wording feels copied or exaggerated
- Meaning and facts are preserved
- The result is recognisable without becoming a caricature