Tone Of Voice — Development skill for Claude Code


**English** · [Русский](README.ru.md) · [Українська](README.uk.md)

   

**Tone of Voice** is a pair of [Claude Code](https://code.claude.com/docs/en/overview) skills that reverse-engineer how a person (or a brand) writes — and then write new texts that sound like them.

Feed it posts, emails, articles, or speech transcripts. Get back a structured **voice profile**: what makes the voice recognizable, which words it lives on, which it would never touch, how it changes when it sells versus when it apologizes — plus a compact *machine block* you can paste into any LLM as a system prompt.

The two skills

Skill	What it does
tone-of-voice-analyzer	Corpus in → metrics-grounded 8-level linguistic analysis → profile.md + blind validation test
tone-of-voice-writer	Profile + brief in → a post/email/article in the profiled voice, verified against the profile before delivery

How it works

flowchart LR
    A["Corpus
posts · emails · transcripts"] --> B["Segmentation
by context & language"]
    B --> C["metrics.py
objective numbers"]
    B --> D["8-level analysis
per segment"]
    C --> D
    D --> E["Synthesis
Voice vs Tone"]
    E --> F["profile.md
incl. machine block"]
    F --> G{"Blind test:
real vs generated?"}
    G -- "grader tells them apart" --> D
    G -- "coin flip" --> H["Profile ready"]
    H --> I["tone-of-voice-writer
new texts in the voice"]

Three design decisions do most of the work:

1. Numbers before adjectives. A bundled stdlib-only script (`metrics.py`) measures what a model should never eyeball: sen