Humanize skill
Two skills for English prose: humanize rewrites a model draft in a natural voice, and ai-check flags machine-writing tells
- 587 ★ on the repository
- Last commit
- MIT
Minimal access
Why this level
- The skills are text rules and run nothing
- The install.sh script creates links or copies of folders in your agent's skills directories
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add skills/humanizeDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewMinimal accessThe tool runs nothing and makes no network calls.
- The skills are text rules and run nothing
- The install.sh script creates links or copies of folders in your agent's skills directories
This is third-party code. Review the repository files before installing.
What it does
Humanize is a plugin with two skills for English text. The humanize skill rewrites a model draft so it reads as written by a person, or writes new text in that register from the start. Its rules rest on nine levers. The skill varies sentence length and picks less predictable words, strips hedges and template lists, and adds concrete detail and the author's voice. SKILL.md opens with seven hard rules for the finished text. No more than one em dash per 300 words, no semicolons, straight quotes only. Words like delve and leverage are banned, and so is "not just X" negation framing. The reply is the rewritten text alone, with no preamble or changelog. To match your own style, paste a sample of your writing, two or three paragraphs, before the text. The ai-check skill works the other way. It scores text in nine signal categories on a 0 to 3 scale and quotes evidence for every category that fires. The reply holds a verdict from Human to AI, a confidence level and an estimate of how much of the text was AI-edited. Both skills are static: no models and no outside calls, only a rulebook for your agent.
Who it is for. For writers, marketers and developers who draft English text with Claude Code, Codex and other agents and want the draft to read naturally.
Good fit when
- A draft email, post or release note reads like a chatbot reply, full of delve and furthermore, same-length sentences and dashes
- You need to audit your own or someone else's text and learn which spots give away machine writing, with quotes from the text
- You write in your own voice and give the agent two or three sample paragraphs so the rewrite follows your style
- You write across genres, from a tech blog and a client email to a LinkedIn post
- You want to score a paragraph with ai-check first and then hand it to humanize to fix the flagged tells
Not a fit when
- You need Russian text: the rules are tuned for English business and technical prose
- You are editing fiction, a strict academic paper with fixed notation or a legal document
- The text lacks your own experience and verified facts: the skill fixes form and invents no content
Example request
Run ai-check on this paragraph and show the score, then rewrite it with the humanize skill in my voice using the sample belowLimitations
- The skill improves the prose and does not guarantee that commercial detectors will accept the text as human. The GPTZero, Grammarly and Pangram classifiers still flag the result as AI because they read the model-training fingerprint, which surface rewriting leaves unchanged.
- The rules fit general business and technical English prose. For fiction, academic papers with fixed notation, legal language and other languages the author warns that different logic applies.
- Confident inventions, a lack of personal experience and implausible detail stay in the text: fixing them is the author's job.
- Quality of the rewrite depends on your agent's model. The model may break the hard rules at the top of SKILL.md, so check the finished text against them.
How to disable. Remove the plugin with claude plugin uninstall humanize@humanize, or in Codex with codex plugin remove humanize@humanize. If you installed through install.sh, delete the links to the humanize and ai-check folders in the skills directory, for example in ~/.claude/skills.
FAQ
How do I install Humanize in Claude Code?
Run `claude plugin marketplace add harshaneel/humanize`, then `claude plugin install humanize@humanize` and restart Claude Code. Both skills come in one plugin.
How do I invoke humanize and ai-check?
Through the plugin use `/humanize:humanize` and `/humanize:ai-check` with the text below. Installed from a clone the commands are shorter: `/humanize` and `/ai-check`. You can also just ask the agent: "Please humanize this text" or "Does this sound AI?".
How is humanize different from ai-check?
humanize rewrites text so it reads as human. ai-check rewrites nothing: it scores text in nine categories, quotes evidence and returns a verdict with a confidence level.
How do I install Humanize in ChatGPT, Cursor or Claude.ai?
In Claude.ai and Claude Desktop upload both SKILL.md files under Settings, Capabilities, Skills. ChatGPT desktop reads `~/.agents/skills`, which `./install.sh chatgpt` fills. In other chats, Cursor included, paste the SKILL.md contents at the start of the conversation.
How do I update Humanize?
In Claude Code run `claude plugin update humanize`, and in Codex `codex plugin marketplace upgrade` refreshes the marketplace snapshot. If the skills were linked from a clone, a git pull in the repository folder is enough.
Discussion
No questions or reviews yet.
Related
SkillsA skill that finds AI writing tells, rewrites the text in a chosen voice, and optionally scores it from 0 to 100
A skill that edits English drafts: strips AI boilerplate, keeps the author's voice, and can also just list what it found without rewriting
im-not-ai
im-not-ai (humanize-korean)
A Claude Code plugin skill that rewrites AI-written Korean text into a natural human style without touching the meaning
Avoid AI Writing
avoid-ai-writing
A skill that finds typical machine-writing patterns in text and rewrites them, with detect-only and edit-in-place modes
An editor for AI-written Russian text: removes bureaucratese, cliches and 38 AI patterns while keeping facts intact
An editor skill and linter for Russian AI-written text: bureaucratese, empty openers, clichés and chatbot wrappers
Comparison with similar tools
Entries that share the main search query, and how they differ.
| Entry | Official | Authentication | Repository stars | Main difference |
|---|---|---|---|---|
| Humanize (this entry) | no | 587 | ||
| Stop Slop | no | 18,452 | a short set of strict rules against boilerplate, without text scoring or a voice sample mode | |
| MCP Giga Checker | yes | API key | 5 | an MCP server that checks text for neural generation signs through the GigaChat API |
README in short
The README moves from installation for different agents to invocation examples and a test of the skill on 25 texts of different genres. It then reviews research on how machine and human writing differ, explains the nine levers and lists the limitations.
README badge
Original SKILL.md
---
name: humanize
description: >
Use whenever the user asks to "humanize", "make this sound more human", "rewrite to
avoid AI detection", "make this less AI-sounding", "add a human voice", or "write like
a person". Also use when the user pastes text and asks why it reads as robotic, generic,
flat, or AI-like, or when generating new text in a register where AI tells (em dashes,
semicolons, hedges, banned vocabulary like "delve", "leverage", "robust") would damage
credibility.
---
# Humanize Text Skill
Transforms AI-generated or flat text into output that mirrors the statistical and stylistic
fingerprint of human writing. Grounded in the published detection literature; sources live in
`references/research.md` (background only, not needed during a rewrite).
---
## Hard rules (read first, enforce last)
These seven fail more often than everything else combined, because the model that wrote the
draft is the model checking it. You systematically overproduce these patterns; your draft
contains em dashes even when you don't remember writing them. Treat "my draft is probably
clean" as false by default.
1. **Em dashes:** at most one per 300 words of output. Under 300 words, zero.
2. **Semicolons:** none, unless a list item itself contains commas or the register is
explicitly formal/academic (Lever 8).
3. **Straight quotes and apostrophes only.** Never curly. Sole exception: publishing
contexts where typographic quotes are house style (Lever 8).
4. **Banned vocabulary** (full list at end of this skill). Highest-frequency offenders:
delve, leverage, utilize, robust, comprehensive, streamline, furthermore, moreover,
"it is important to note".
5. **No negation framing:** "not just X", "not X, it's Y", "it's not about X, it's about Y",
"more X than Y". Say what the thing IS. Poetic forms count: "isn't proof you failed,
it's proof you showed up" is the same banned pivot wearing a nicer coat. (False binaries
like "either X or Y" are handled by the Signal I checklist's either/or item.)
6. **Output shape:** the rewritten text only. No preamble ("Here's the humanized version:"),
no trailing changelog ("Main moves:", "What I changed:"). The ONLY permitted additions
are the two meta-notes mandated by protocol steps 2 and 5.6, appended after the rewrite.
If the user wants the diff explained, they'll ask.
7. **Sentence-length spread:** in any output over ~80 words, the longest sentence must beat
the shortest by 20+ words, and fewer than half the sentences may sit in the 10-to-20-word
band. Your uncorrected rhythm clusters at 10-20 words with ~6 words of deviation; that