Changelog

All notable changes to humanizer-workbench are documented here.

Format follows Keep a Changelog. This project uses semantic versioning.

[Unreleased]

SKILL.md: Claude Code skill interface for interactive humanization in Claude Code sessions
Updated README to document both CLI and Claude Code skill usage
examples/sample_ai_text.txt: realistic sample for testing the tool

Multi-stage humanization pipeline: REWRITE → REFINE → AUDIT
Six writing style presets: casual, professional, technical, founder, academic, storytelling
Three intensity levels: light (1 stage), medium (2 stages), aggressive (3 stages)
Heuristic AI-likeness scorer with five components:
- AI vocabulary density (up to 30 points)
- Filler phrase density (up to 25 points)
- Sentence length uniformity (up to 20 points)
- Structural pattern density (up to 15 points)
- AI sentence opener patterns (up to 10 points)
Lexical detector with 50+ AI vocabulary words and 30+ filler phrases
Structural detector for sentence uniformity, em dash overuse, rule-of-three patterns
CLI with --style, --intensity, --diff, --score, --explain, --output flags
humanizer detect subcommand for pattern analysis without transformation
Python API: HumanizerEngine.humanize() returns a HumanizerResult
Unit tests covering detectors, scorer, styles, engine, and pipeline
Integration tests marked with @pytest.mark.integration