humanizer-stack
Two-pass pipeline for removing AI writing tells from outward-facing text: a surface pass plus a structural pass grounded in the StoryScope study. Packaged as Claude Code Skills. Free community: skool.com/jens-ai-community-1306
What is it?
What it is
A two-pass pipeline for removing signs of AI writing from text, packaged as Claude Code Skills.
Why it exists
Surface-level AI tells decay fast; durable fingerprint is in discourse structure, requiring both passes to avoid creating a new detectable cluster.
Who should use it
Content creators using LLMsAI prompt engineersCopywritersClaude Code usersTechnical writers aiming for human-like prose
Who should avoid it
Users seeking 100% undetectability guaranteesShort-form nonfiction writers (due to study limitations)Those looking for a simple word-replacement tool
How it works
A quick walkthrough in plain English
How humanizer-stack works
Step 1 of 3
You interact with it
Open humanizer-stack, send a request, or connect it to your stack.
Features
Advantages
- Addresses both surface-level and structural AI writing patterns
- Backed by empirical research (StoryScope study)
- Deterministic tools enable reliable integration
- Focus on structural elements improves detectability resistance
- Permissive MIT license with clear attribution requirements
- Customizable interventions per piece to avoid detection clusters
- Scanners provide machine-readable output for automation
- Community support and documentation (Skool group)
Disadvantages
- Limited validation on nonfiction/text types outside the study's scope
- Scanners only detect ~50% of issues (human judgment still required)
- Structural audits may conflict with traditional writing conventions
- CC BY-SA 4.0 components impose share-alike obligations
- Requires technical setup (Python scripts, Git workflow)
- Two-pass process may increase workflow complexity
- Not designed for complete undetectability (aims for human-like appearance)
Installation
FAQ
What is the difference between the two passes in the humanizer-stack pipeline?
Pass 1 (humanizer) focuses on surface-level changes like vocabulary, punctuation, and phrasing to remove 'copy tells.' Pass 2 (structural-humanizer) focuses on discourse structure and 'hape' to address the deeper structural fingerprints that AI detectors use to identify LLM-generated text.
How do I install these skills for use in Claude Code?
You can install them by cloning the repository and running `./install.sh`, which symlinks the skills into your `~/.claude/skills/` directory. Alternatively, you can manually copy the `humanizer` and `structural-humanizer` folders into your user-level or project-level Claude skills directory.
How do I trigger the humanizer skills within Claude Code?
The skills are intent-based. You can trigger them by using natural language commands such as 'humanize this post', 'de-slop this lesson', or 'run the structural pass on draft.md'.
Can I use the scanners to check my text for AI patterns?
Yes. You can use `scripts/copy_scan.py` for mechanical copy tells and `skills/structural-humanizer/scripts/structural_scan.py` for structural tells. Both scripts support standard input and can output results in JSON format for machine readability.
Does using this tool guarantee that my text will be undetectable by AI detectors?
No. The goal of the tool is to produce writing that reflects a specific human point of view rather than achieving absolute undetectability. The scanners only catch pattern-matchable elements; cadence and formulaic shape often require human judgment to fully address.
Featured in Videos
YouTube tutorials and walkthroughs for humanizer-stack
Alternatives
Similar projects ranked by category, topics, and text overlap.
