cosmos-makers/writer-persona
Code reviewこのページは AI により自動翻訳されています。技術的な詳細は原文をご確認ください。 原文を見る
A Claude Code skill that extracts your writing persona from real messages, measures its accuracy, and writes in your voice.
概要
A Claude Code plugin that extracts your writing persona from real messages, measures its accuracy, and writes in your voice. You want AI to write as you — in emails, Slack, code reviews. But how close is "close enough"? Without measurement, you're guessing. creates a feedback loop: extract, measure, improve, write. The AI never sees your actual response when generating drafts — it only gets the conversation context. This information isolation ensures honest evaluation. : When installed as a plugin, commands use the plugin-name:skill-name format. If installed as a local skill (in .claude/skills/), use /writer-persona directly.
README
以下は原文 README の抜粋です。原文を見る。
writer-persona
A Claude Code plugin that extracts your writing persona from real messages, measures its accuracy, and writes in your voice.
The Problem
You want AI to write as you — in emails, Slack, code reviews. But how close is “close enough”? Without measurement, you’re guessing.
The Solution
writer-persona creates a feedback loop: extract, measure, improve, write.
Bootstrap (analyze 100+ real messages)
↓
Backtest (AI writes drafts → compare with what you actually wrote)
↓
Score (8-axis linguistic evaluation)
↓
Auto-correct (fix the biggest gaps)
↓
Write (generate messages in your voice)
The AI never sees your actual response when generating drafts — it only gets the conversation context. This information isolation ensures honest evaluation.
Commands
| Command | What it does |
|---|---|
/writer-persona:writer-persona |
Show persona status and score |
/writer-persona:writer-persona --bootstrap |
Extract persona from 100+ real messages |
/writer-persona:writer-persona --backtest |
Measure accuracy (last 24h by default) |
/writer-persona:writer-persona --backtest --days 7 |
Measure over last 7 days |
/writer-persona:writer-persona --calibrate |
Manual tuning with your feedback |
/writer-persona:writer-persona --write "context" |
Write as you in a given situation |
/writer-persona:writer-persona --write |
Interactive: describe the situation, get a draft |
Note: When installed as a plugin, commands use the
plugin-name:skill-nameformat. If installed as a local skill (in.claude/skills/), use/writer-personadirectly.
Key Features
- 8-axis evaluation grounded in computational stylometry research
- Self-similarity ceiling — your own writing varies; the AI is scored against your natural range, not perfection
- Data source agnostic — Slack, email, Discord, text files, or any MCP tool
- Language adaptive — English, Korean, Japanese, or any language
- Auto-convergence with overfitting guards (max 2 axes per run, stall detection)
- Write mode — the payoff: generate messages that sound like you
The 8 Axes
| Axis | Weight | What it measures |
|---|---|---|
| Semantic Fidelity | 25% | Does the draft say the right thing? |
| Tone / Sentiment | 20% | Does it feel like you? |
| Formality / Register | 15% | Right level of formal/casual? |
| Discourse Patterns | 15% | How you structure responses |
| Lexical Choice | 10% | Your characteristic vocabulary |
| Function Words | 5% | Unconscious grammatical habits |
| Sentence Structure | 5% | Length, complexity patterns |
| Formatting | 5% | Emoji, punctuation, layout |
See references/evaluation-rubric.md for the full framework and academic citations.
Installation
Via Plugin Marketplace (Recommended)
推奨ツール
別のキーワードを試すか、フィルタを外してください。
インストール
npx skillfish add cosmos-makers/writer-persona