Create personal voice profile skill for AI writing artifacts

[kergoth]

Summary

Build a Claude Code skill that captures my personal writing voice and applies it when generating prose artifacts (emails, blog posts, proposals, design documents).

Context

Evaluated 7 community skills for AI writing quality (humanizer, avoid-ai-writing, anti-ai-slop, skill-stack/anti-ai-writing, claude-vibes/ai-writing-detection, journalism/ai-writing-detox, melodic/ai-writing-assistant). All are subtractive (remove AI patterns) but none solve additive voice matching (sound like me).

The skill-stack voice-matching-wizard has the best framework: 6-dimension extraction (sentence architecture, word choice, tone/attitude, structural moves, distinctive techniques, avoidances) with spectrum ratings. Use it as a starting point.

Approach

1. Gather writing samples across genres (old pre-AI artifacts, session JSONL conversational input, blog posts, technical docs, emails)
2. Analyze patterns using the voice-matching-wizard framework or a refined version
3. Produce a persistent voice-kergoth skill with per-genre variants
4. Auto-load on narrower triggers (emails, blog posts, proposals) via skill description

Key design decisions

• Persistent voice profile, not session-scoped
• Per-genre variants (technical doc voice ≠ email voice ≠ blog voice)
• Whitelist patterns I actually use that overlap with AI markers
• Complement the subtractive layer (avoid-ai-writing), not replace it

References

• blader/humanizer — voice calibration from samples
• cdeistopened/skill-stack — voice-matching-wizard + anti-ai-writing
• conorbronsdon/avoid-ai-writing — best subtractive skill (tiered vocab, context profiles)
• melodic-software/claude-code-plugins — voice preservation framework concepts
• Wikipedia: Signs of AI writing