Skip to content
Back to search

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

Safety scan

A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.

Checked today
Badge for your README
findsafeskills safety rating

Markdown

[![findsafeskills safety rating](/api/listings/nulightjens--humanizer-stack/badge.svg)](/listings/nulightjens--humanizer-stack)

HTML

<a href="/listings/nulightjens--humanizer-stack"><img src="/api/listings/nulightjens--humanizer-stack/badge.svg" alt="findsafeskills safety rating" /></a>

Install

i

Install instructions from the README

As written by the repo's author. findsafeskills has not verified these steps.

git clone https://github.com/NulightJens/humanizer-stack.git
cd humanizer-stack
./install.sh

This symlinks both skills into `~/.claude/skills/`, so updates land with a `git pull`. Pass `--copy` if you would rather have independent copies than symlinks.

To install manually, copy `skills/humanizer` and `skills/structural-humanizer` into `~/.claude/skills/` (user-level) or `.claude/skills/` (project-level).

What was scanned

The files the safety scan read, as written by the repository's author and shown as plain text. Lines with a finding are marked.

README.md · README · 157 lines

View this file on GitHub (read at the repository's current version: the scan did not record which commit it read)

  1. # humanizer-stack
  2. A two-pass pipeline for removing the signs of AI writing from outward-facing text,
  3. packaged as [Claude Code Skills](https://code.claude.com/docs/en/skills).
  4. Most humanizers only fix words. That is the easy half, and it is the half that is
  5. decaying fastest. This repo pairs a surface pass with a structural pass, because the
  6. research says structure is where the durable fingerprint lives.
  7. > **Free community.** I build tools like this in the open inside the
  8. > [Jens AI Community](https://www.skool.com/jens-ai-community-1306), a free Skool group
  9. > for putting AI to work in your business. If this repo is useful to you, come join us:
  10. > https://www.skool.com/jens-ai-community-1306
  11. ## Why two passes
  12. The [StoryScope study](https://github.com/jenna-russell/storyscope) (Russell et al.,
  13. 2026) classified 61,608 stories from humans and five LLMs using only discourse-level
  14. features, with every style feature withheld. It detected AI text at **93.2% F1**.
  15. Then the authors ran AI text through LAMP, a professional span-level rewriting system
  16. that strips cliche, purple prose, and redundant exposition. Functionally, a very good
  17. surface humanizer.
  18. Detection dropped **1.6 points**.
  19. Meanwhile the surface layer is eroding on its own. GPT 5.4 already cut its em-dash
  20. usage sharply, and fine-tuning drops stylistic detection from 97% to 3%. Word-level
  21. tells are a moving target. Structural tells require structural rewrites.
  22. So: pass 1 fixes the words. Pass 2 fixes the shape. Run them in that order.
  23. ## What is in here
  24. ```
  25. skills/
  26. humanizer/ Pass 1: words and phrasing
  27. SKILL.md
  28. references/copy-tells.md Copy-specific tells (em dash, hype vocab, antithesis)
  29. structural-humanizer/ Pass 2: discourse structure
  30. SKILL.md
  31. references/
  32. storyscope-findings.md The study distilled: 30 core features with rates
  33. genre-calibration.md Which audits apply per genre
  34. scripts/structural_scan.py Deterministic scanner for grep-able structural tells
  35. scripts/
  36. copy_scan.py Deterministic scanner for mechanical copy tells
  37. docs/
  38. PIPELINE.md How the passes chain, and what each one owns
  39. ```
  40. ### Pass 1: `humanizer`
  41. Vocabulary, punctuation, and phrasing. Inflated symbolism, promotional language,
  42. superficial "-ing" analyses, vague attributions, em dash overuse, rule of three, AI
  43. vocabulary, negative parallelism. Built from Wikipedia's
  44. [Signs of AI writing](https://en.wikipedia.org/wiki/Wikipedia:Signs_of_AI_writing).
  45. The `copy-tells.md` reference adds the tells that show up specifically in public copy,
  46. ranked by a 3.2M-post Reddit analysis of what people actually flag.
  47. ### Pass 2: `structural-humanizer`
  48. Six audits run one at a time, because aspect-based checking found 95% of issues in the
  49. study's own pipeline against 68% for a single combined pass:
  50. 1. **Theme explicitness.** AI states its lesson. Narrator explains the theme 77% of the
  51. time against 52% for humans.
  52. 2. **Structural tidiness.** Single-track, everything resolved. Humans digress and leave
  53. threads open.
  54. 3. **Emotion mode.** The largest gap in the study. AI performs emotion through the body
  55. 81% of the time against 38% for humans. Humans just name the feeling.
  56. 4. **Reference specificity.** Humans name real things (47% against 24%). AI stays at
  57. vague allusion.
  58. 5. **Reader engagement.** Humans acknowledge the reader. AI writes as though no one is
  59. watching.
  60. 6. **Shape convergence.** Does this piece have the same skeleton as your last three?
  61. ## The trap
  62. Do not trade one default for another. The study's deepest finding is convergence: all
  63. five models occupy one tight region of structural space while humans are dispersed.
  64. Rarity is the human signal.
  65. If every piece now opens mid-scene, names three feelings, and ends unresolved, you have
  66. built a new detectable cluster. Pick one or two interventions per piece, vary them
  67. across pieces, and be able to say why this piece got this shape.
  68. ## Install
  69. ```bash
  70. git clone https://github.com/NulightJens/humanizer-stack.git
  71. cd humanizer-stack
  72. ./install.sh
  73. ```
  74. This symlinks both skills into `~/.claude/skills/`, so updates land with a `git pull`.
  75. Pass `--copy` if you would rather have independent copies than symlinks.
  76. To install manually, copy `skills/humanizer` and `skills/structural-humanizer` into
  77. `~/.claude/skills/` (user-level) or `.claude/skills/` (project-level).
  78. ## Use
  79. In Claude Code, the skills trigger on intent:
  80. ```
  81. humanize this post
  82. de-slop this lesson
  83. run the structural pass on draft.md
  84. ```
  85. Run the surface pass first, then the structural pass. `docs/PIPELINE.md` covers the
  86. order and what each layer owns.
  87. ### Scanners
  88. Both scanners are deterministic and hook-friendly. They catch the pattern-matchable
  89. slice only, and neither replaces the judgment work in the skills.
  90. ```bash
  91. python3 scripts/copy_scan.py draft.md
  92. python3 skills/structural-humanizer/scripts/structural_scan.py draft.md
  93. python3 scripts/copy_scan.py --json draft.md # machine-readable
  94. python3 scripts/copy_scan.py --strict draft.md # exit 1 on any hit
  95. cat draft.md | python3 scripts/copy_scan.py - # stdin
  96. ```
  97. Mark a line `copy-ignore` to suppress an intentional usage.
  98. ## Honest limits
  99. - **StoryScope studied roughly 5,000-word fiction.** Applying it to short nonfiction is
  100. an inference, not a result the paper establishes. The subset that transfers most
  101. cleanly is audits 1, 3, 4, and 6.
  102. - **Nothing here makes text undetectable, and that is not the goal.** The goal is
  103. writing that reads as though a person with a specific point of view wrote it, because
  104. a person did.
  105. - **The scanners catch maybe half.** Cadence, formulaic shape, and polished-but-empty
  106. filler are only visible to a human reader.
  107. - **Audit 3 contradicts standard writing advice.** "Show, don't tell" is now a machine
  108. signature. That is what the data says, and it is worth sitting with before applying.
  109. ## Attribution
  110. Built on work by [@blader](https://github.com/blader/humanizer) (MIT),
  111. [jcarterjohnson](https://github.com/jcarterjohnson/vibecoded-design-tells) (MIT), and
  112. Wikipedia's WikiProject AI Cleanup (CC BY-SA 4.0). Grounded in Russell et al. 2026.
  113. Full breakdown with license obligations: [ATTRIBUTION.md](ATTRIBUTION.md).
  114. ## License
  115. MIT for this repository's own work. Portions carry upstream terms, including CC BY-SA
  116. 4.0 material with share-alike obligations. See [ATTRIBUTION.md](ATTRIBUTION.md) before
  117. redistributing.

Ratings and comments

Sign in to rate this skill.

Sign in

Free to use. If it’s helped, pass it on.