| name | voice-creator |
| description | Create a custom voice/interaction style template. Use when the user says "create a voice", "make a style", "new personality", "custom voice", "I want Claude to talk like", or provides examples of a writing style they want to replicate. |
| argument-hint | [style name or description] |
| disable-model-invocation | true |
| allowed-tools | Read, Write, Glob |
Voice Style Creator
Goal
Produce a saved voice style file at .claude/skills/voice-style/styles/<name>.md that captures a distinct personality with enough specificity (rules + examples) that Claude reliably adopts it when /voice-style <name> is invoked.
Dependencies
- Tools: Read, Write, Glob
- Existing styles: Check
.claude/skills/voice-style/styles/ to avoid naming collisions and to use as format reference
Context
- Review one existing style file in
.claude/skills/voice-style/styles/ before writing to match the established format
- No external APIs or MCP servers required
Process
Step 1: Gather Input
Ask the user for ONE of the following (in order of preference):
Option A - Text Examples (best results):
"Paste 2-3 examples of the style you want. These can be your own writing, excerpts from someone you admire, chat messages, tweets, anything that captures the vibe."
Option B - Description:
"Describe the personality, tone, and energy you want. Think about: How formal? How much personality? Any specific words/phrases? Who does this voice sound like?"
Option C - Reference Person/Character:
"Name a person, character, or archetype and I'll build a style based on their communication patterns."
Step 2: Analyze the Style
From the input, extract:
- Tone: Formal/casual/playful/serious/warm/sharp
- Energy level: Calm/moderate/high/chaotic
- Vocabulary patterns: Specific words, slang, jargon, phrases
- Sentence structure: Short and punchy? Long and flowing? Questions? Fragments?
- Emoji usage: None/minimal/moderate/generous
- Humor style: None/dry/sarcastic/goofy/witty
- Signature moves: Catchphrases, opening patterns, closing patterns
- What they NEVER do: Equally important as what they do
CHECKPOINT: Present the style analysis to the user before writing the template. Ask: "Here's what I've extracted from your input. Does this capture the voice correctly, or should I adjust any of these attributes before writing the template?"
Step 3: Write the Template
Create a style file following this structure:
[Style Name] Mode - [One-line description of the vibe]
## Rules
1. **[Core trait]:** [Detailed description with examples]
2. **[Language pattern]:** [Specific vocabulary, phrases, structures]
3. **[Energy/tone]:** [How to modulate intensity]
4. **[Emoji/formatting]:** [Visual style rules]
5. **[Supportive behavior]:** [How to respond to user emotions]
6. **[Boundaries]:** [What this style NEVER does]
7. **Still helpful:** [Reminder that accuracy isn't compromised]
## Examples
**[Common scenario 1]:**
"[Example response in this style]"
**[Common scenario 2]:**
"[Example response in this style]"
**[Common scenario 3]:**
"[Example response in this style]"
CHECKPOINT: Show the completed template to the user before saving. Ask: "Here's the full voice template. Should I save this as-is, or would you like to tweak anything first?"
Step 4: Save and Test
- Save to
.claude/skills/voice-style/styles/<name>.md
- Demonstrate the style with a sample response
- Ask user for feedback and iterate
Key Principles
- 6-8 rules maximum. More than that and the style gets diluted.
- Specific over vague. "Use words like X, Y, Z" beats "be casual."
- Include examples. The model learns more from examples than rules.
- Include anti-examples. What does this voice NEVER sound like?
- Test with technical content. Make sure the style works for code explanations, not just conversation.
Output
- Save location:
.claude/skills/voice-style/styles/<name>.md
- Format: Markdown with a Rules section (6-8 numbered rules) and an Examples section (3+ scenarios)
- Activation: User can invoke with
/voice-style <name> after saving
