CLAUDE.md

Human Voice Plugin

A Claude Code plugin for detecting and preventing AI-generated writing patterns.

Project Context

This plugin provides:

• Skill: human-voice - Core detection patterns and writing guidelines
• Commands: /human-voice:setup, /human-voice:review, /human-voice:fix
• Agent: voice-reviewer - Proactive content review after edits

Development Guidelines

• Follow Claude Code plugin standards
• Test commands locally with claude --plugin-dir .
• Update CHANGELOG.md for user-facing changes
• Keep backward compatibility (all features work without optional dependencies)

File Structure

skills/human-voice/SKILL.md     # Core skill with detection patterns
commands/*.md                   # Slash commands
agents/voice-reviewer.md        # Proactive review agent
scripts/                        # Node.js validation/fix scripts
references/                     # Pattern documentation

Subcog Memory Integration (Optional)

If Subcog MCP server is available, this plugin can leverage persistent memory for:

• Recalling project-specific voice decisions
• Storing learned patterns and exceptions
• Tracking configuration preferences across sessions

Relevant Namespaces

• decisions: Voice style choices (e.g., "project allows emojis in README")
• patterns: Recurring voice fixes (e.g., "replace em dashes with colons")
• learnings: Project-specific gotchas discovered during review
• config: Per-project voice configuration settings

Memory-Aware Workflow (When Available)

Before analysis:

subcog_recall: query="voice patterns OR voice decisions", filter="ns:decisions ns:patterns"

After significant findings:

subcog_capture: namespace=learnings, content="[finding]", tags=[human-voice, voice-pattern]

Graceful Degradation

All plugin functionality works without Subcog:

• Commands execute normally without memory context
• Agent performs full analysis without recall
• Configuration via .claude/human-voice.local.md always works

Memory integration is additive - it enhances but never blocks core functionality.