Our take
Activeregular commits and releases- Activity
- 61
- Community
- 52
- Issues
- 64
- Pull requests
- 39
+7,569 stars in the last 22 days
- Slower to merge pull requests than 89% of the projects we track
Humanizer is a focused editing skill with clear guardrails, and its recent activity and broad contributor base make it a viable option for agent-based writing workflows.
Good fit if
- You use Claude Code, Codex, or another supported agent and want to revise AI-written prose for human readers.
- You need an editing workflow that checks claims against the source and can preserve non-prose parts of a file.
- You want an MIT-licensed project with regular commits and releases, and issue reports that are being handled.
Look elsewhere if
- Your goal is to evade AI detectors, which the project explicitly does not target.
- You need a standalone editor rather than a skill for an agent that supports skills.
- Your workflow depends on a detector-passing guarantee or other behavior the README does not promise.
All health signals
| Last commit | 2026-10-16 (2 days ago) |
|---|---|
| Commits, last 90 days | 62 |
| Releases, last 12 months | 6 (latest v3.1.0, 2026-09-28) |
| Contributors | 22 (top contributor: 65% of commits) |
| Issues closed, last 90 days | 34+ (typically closed in 10 days) |
| Pull requests merged, last 90 days | 30 (typically merged in 5 days) |
| Project age | 9 months |
Checked on 2026-10-19 with the GitHub API.
Overview
Humanizer is an agent skill built on Wikipedia’s “Signs of AI writing” guide. It helps people using Claude Code, Codex, or another agent that supports skills remove common AI-writing tells while keeping the original meaning and factual claims intact.
It marks patterns in the source, drafts a rewrite, checks the result against the original claims, and presents a final version. It is intended to improve prose for human readers, not to bypass AI detectors; the README says detectors still flag most of its output.
Key Features
- Checks for 26 patterns, including inflated significance, staged introductions, forced triads, chatbot residue, and decorative formatting.
- Shows the first rewrite, a critique of remaining artificial-sounding passages, and a final version when text is pasted.
- Can rewrite prose in a file while leaving code, data, frontmatter, and link targets alone.
- Preserves names, numbers, dates, quotes, citations, and other factual details from the source, and asks when a needed detail is missing.
- Supports voice matching from a sample of the writer’s own prose.
- Keeps personal writing’s opinions and quirks, while aiming for neutral, plain language in technical and reference prose.
Use Cases
- A writer wants to revise an AI-generated announcement or article before publishing, without changing its claims.
- A developer wants an agent to edit prose in a documentation file without altering code or metadata.
- A team using Claude Code or Codex wants a repeatable editing skill available in its agent workflow.
- A writer wants to adapt an AI draft to their own rhythm and word choice by providing a writing sample.
What you need
Detected in the repository
- Python (main language on GitHub)
- Automated checks on GitHub Actions
License in plain words
MITpermissive
- Commercial use: yes
- Modify and redistribute: yes
- You must keep: the copyright and license notice
- Share your changes: no
A summary, not legal advice: the LICENSE file is what applies.
Getting Started
For Codex, install the skill globally with:
npx skills add blader/humanizer --global --agent codex
See the README for Claude Code, Claude.ai, and other agent installation options, plus usage details.
Considerations
- The project is a relatively new skill, though it is actively maintained, has regular releases, and has contributions from multiple people. Its recent issue and pull request activity indicates ongoing maintenance, not a guarantee of future support.
- Rewrites still require review: the project describes safeguards against inventing details, but does not claim that every edit will preserve meaning perfectly.
- Humanizer is designed for improving text for people, not for passing AI detectors. The README says detectors still flag most of its output.
- The README lists no runtime or container requirements, but use depends on an agent that supports skills or on manually placing
SKILL.mdin an agent’s skill folder.
Found this useful?
Share it with someone who would like humanizer.