Human Voice Plugin

A Claude Code plugin that detects AI-generated writing patterns and builds voice profiles through adaptive interviews and computational stylistics.

Experimental. This project is a research prototype. The scoring pipeline, dimension mapping, and NLP analysis have not been validated against external benchmarks or peer-reviewed psychometric standards. The voice profiles it produces are plausible but unproven. Treat the output as a starting point for editorial guidance, not as a validated instrument. The question bank, scoring weights, and dimension definitions will change as the system matures. Use it, break it, report what does not work.

Features

• Multi-tier pattern detection: Character, language, structural, and voice analysis
• Automated character fixes: Auto-fix em dashes, smart quotes, emojis
• Proactive review: Agent triggers after content creation/editing
• Interactive setup: Configuration wizard for project-specific settings
• Configurable: Customize file types, directories, and detection tiers

Installation

From GitHub

claude plugin install zircote/human-voice

Manual Installation

Clone and add to Claude Code:

git clone https://github.com/zircote/human-voice.git
claude --plugin-dir /path/to/human-voice

Or copy to your project's .claude-plugin/ directory.

Prerequisites

• Claude Code CLI
• Node.js 18+ (for validation scripts)

Components

Component	Name	Purpose
Skill	human-voice	Core detection patterns and writing guidelines
Command	/human-voice:voice-setup	Interactive configuration wizard
Command	/human-voice:voice-review [path]	Analyze content for AI patterns
Command	/human-voice:voice-fix [path]	Auto-fix character-level issues
Agent	voice-reviewer	Proactive content review after edits

Usage

Quick Start

# Set up configuration for your project
/human-voice:voice-setup

# Review content for AI patterns
/human-voice:voice-review docs

# Auto-fix character issues
/human-voice:voice-fix docs --dry-run

Skill Triggers

The skill loads automatically when you say:

• "review for AI patterns"
• "make this sound human"
• "check for AI writing"
• "ai slop detection"
• "fix AI voice"
• "improve writing voice"

Commands

Set up configuration:

/human-voice:voice-setup

Detects project structure, content directories, and creates config.json with your preferences.

Review content for AI patterns:

/human-voice:voice-review docs           # review specific directory
/human-voice:voice-review content/blog   # review specific path
/human-voice:voice-review                # auto-detects content directories

Auto-fix character issues:

/human-voice:voice-fix docs              # apply fixes to directory
/human-voice:voice-fix --dry-run docs    # preview changes first
/human-voice:voice-fix                   # auto-detect and fix

Agent

The voice-reviewer agent triggers:

• Proactively: After Write/Edit operations on .md/.mdx files
• On request: When you ask to review content voice

Detection Tiers

Tier 1: Character Patterns (Automated)

Character	Unicode	Replacement
Em dash (--)	U+2014	Period, comma, colon
En dash (-)	U+2013	Hyphen
Smart quotes	U+201C/D, U+2018/9	Straight quotes
Ellipsis (...)	U+2026	Three periods
Emojis	Various	Remove

Tier 2: Language Patterns (Manual)

• Buzzwords: delve, realm, pivotal, harness, revolutionize, seamlessly
• Hedging: "it's worth noting", "generally speaking", "arguably"
• Filler: "in order to", "due to the fact", "at this point in time"

Tier 3: Structural Patterns

• List addiction (everything as bullets)
• Rule of three overuse
• "From X to Y" constructions
• Monotonous sentence structure

Tier 4: Voice Patterns

• Passive voice overuse
• Generic analogies
• Meta-commentary ("In this article...")
• Perfect grammar with shallow insights

Configuration

Run /human-voice:voice-setup for interactive configuration, or edit config.json directly.

Configuration is stored at $CLAUDE_PLUGIN_DATA/config.json (defaults to ~/.human-voice/config.json in standalone mode). Use python -m lib.config show to view the effective config, or python -m lib.config reset to write defaults.

Memory Integration (Optional)

When Subcog MCP server is available, the plugin can leverage persistent memory:

• Recall project-specific voice decisions before analysis
• Capture findings and patterns for future sessions
• Track configuration preferences across sessions

All features work without Subcog. Memory integration is additive and never blocks core functionality.

File Structure

human-voice/
├── .claude-plugin/
│   └── plugin.json
├── agents/
│   └── voice-reviewer.md
├── commands/
│   ├── voice-fix.md
│   ├── voice-review.md
│   └── voice-setup.md
├── skills/
│   └── human-voice/
│       ├── SKILL.md
│       ├── scripts/
│       │   ├── fix-character-restrictions.js
│       │   └── validate-character-restrictions.js
│       ├── references/
│       │   ├── character-patterns.md
│       │   ├── language-patterns.md
│       │   ├── structural-patterns.md
│       │   └── voice-patterns.md
│       └── examples/
│           └── before-after.md
├── templates/
│   └── observer-protocol.md
├── LICENSE
├── CHANGELOG.md
└── README.md

Voice Elicitation (Voice)

Voice is an experimental voice elicitation system that captures a writer's voice through a 67-question adaptive interview, computational NLP analysis of writing samples, and automated profile synthesis. It produces two independent profiles per writer: a self-reported profile (what the writer believes about their voice) and a computationally observed profile (what their writing exhibits). A calibration layer identifies where these profiles agree and where they diverge.

Status: The scoring pipeline produces numeric dimension scores but these scores have not been validated against external psychometric instruments. The NLP analysis uses standard stylometric measures (type-token ratio, Flesch-Kincaid, hedge density, etc.) but the mapping from NLP metrics to voice dimensions is hand-authored and unvalidated. The question bank is based on published findings in voice elicitation research but the specific item-to-dimension mappings are untested for reliability (Cronbach alpha) across a population. This is a functional prototype, not a finished measurement tool.

CLI Tools

Tool	Purpose
voice-session	Session lifecycle: create, load, list, pause, resume
voice-scoring	Score a completed session and produce dimension profiles
voice-nlp	Run the stylometric NLP analysis pipeline on writing samples
voice-branching	Evaluate interview routing and module sequencing
voice-sequencer	Determine the next question based on session state
voice-quality	Detect satisficing and response quality issues

Getting Started

See the Getting Started tutorial for a complete walkthrough of running your first voice elicitation session.

See the CLI Reference for detailed documentation of all commands, options and output formats.

Research Sources

Pattern detection based on:

• Measuring AI "Slop" in Text
• The Field Guide to AI Slop
• Common AI Words - Grammarly

Contributing

1. Fork the repository
2. Create a feature branch
3. Make your changes
4. Submit a pull request

License

MIT