voice-extractor

The voice-extractor skill allows you to transform raw writing samples into a structured, executable voice profile. It is designed for marketers, founders, and content leads who struggle with generic AI-generated content and want to ensure their automated systems reflect their unique 'communication DNA.' By analyzing casual emails, Slack threads, podcast transcripts, and polished articles, the skill identifies the specific rhythm, vocabulary, and structural patterns that define an individual's authentic voice.

• Performs multi-stage extraction: Analyzes role, energy, recurring themes, phraseology, and confidence mapping.

• Includes a systematic phrase extraction engine: Captures specific transition phrases, emphasis markers, and closing statements.

• Multi-mode execution: Supports 'quick' (snapshot), 'standard' (full documentation), and 'deep' (full guide plus sample rewrites and AI training examples) modes.

• Confidence zone mapping: Distinguishes between areas of expertise and active exploration to ensure tone remains authentic across different topics.

• Anti-pattern detection: Explicitly documents words, phrases, and tones the user naturally avoids to prevent 'robotic' or 'over-formal' AI output.

• Validation testing: Requires a validation step where the model generates contrasting versions of text to confirm the extracted profile is accurate.

• To use effectively, provide at least 500 words of writing samples; raw, unedited text like Slack messages or transcripts is prioritized over polished website copy.

• The skill acts as a gatekeeper: it will pause and request more data if the sample size is insufficient, ensuring high-quality output.

• Use the output as a system prompt for Claude Code or other LLMs to ensure future content generation aligns with your specific style guide.

• It is highly effective for ghostwriting onboarding, brand voice systems, and scaling content teams where consistent voice is mandatory.

• The framework relies on linguistic pattern matching to identify 'authentic' vs 'performative' writing, helping to eliminate AI-typical hedging and buzzwords.