Most AI output reads like AI output. The vocabulary is too polished, the structure is too neat, and it's full of tells like unnecessary dashes, hedging phrases, and bullet points where a normal person would just write a paragraph. If you've ever had to rewrite an AI draft to make it sound like you, this fixes that.

Voice Mirror maintains a living voice profile that learns your writing style and applies it to everything your agent produces. You give it a few writing samples (emails, messages, docs, whatever you've actually written) and it extracts your patterns: sentence rhythm, vocabulary choices, how formal or casual you are, your favorite phrases, and the things you'd never say.

From there, every piece of written output gets filtered through your voice. Quick Slack replies sound like your quick Slack replies. Client proposals sound like your client proposals. The skill knows that you write differently in different contexts but it keeps the underlying personality consistent.

It also enforces hard rules that strip out common AI tells. No dashes or hyphens. No filler hedging like "it's worth noting" or "as you may know." No gratuitous bullet points or headers on a short email. No words like "leverage," "utilize," or "delve" unless those are actually in your vocabulary.

The profile evolves over time. When you give feedback ("this is too formal" or "I wouldn't phrase it that way"), the skill logs it and adjusts. It also picks up on how you write your prompts, so it's always getting better at sounding like you.

Works with Claude Code, Cursor, Codex CLI, VS Code Copilot, and any agent that supports the SKILL.md standard.