Humanizer Skill

Edit AI-sounding prose while keeping your facts and voice.

Source: blader

OpenAI Codex · CompatibleClaude.ai · CompatibleClaude Code · CompatibleWriting & Content

At a glance

Best for
Editing a draft you already have
Works with
Your original text · Audience and tone · Facts to preserve
Outcome
A revised draft · An explanation of edits when requested
Source
blader/humanizer

01 / Understand the skill

About Humanizer Skill

Humanizer is a writing skill by blader for editing drafts that sound formulaic. Use it on product announcements, articles and documentation when stock phrases, inflated language or repetitive sentence patterns get in the way. Give your agent the draft, audience and a sample of your writing; review the revision against the original before publishing.

Documentation reviewed ·

Setup instructions are based on the upstream README, Humanizer 4.0.0 and platform documentation. We tested editing in Codex with a manually installed copy. The installer commands, Claude Code and Claude.ai have not been tested.

Requirements

  • An agent with skills support
  • A draft you can fact-check and a clear audience

Inputs

  • Your original text
  • Audience and tone
  • Facts to preserve

Outputs

  • A revised draft
  • An explanation of edits when requested

02 / Confirm the fit

Use Humanizer Skill when you need to

  • Make a product announcement specific and readable without adding claims.
  • Simplify documentation while preserving commands, limits and conditions.
  • Match a writing sample instead of applying a generic casual voice.
  • Review inflated language, repeated structures and unnecessary filler.

03 / See the workflow

How Humanizer Skill works

  1. 01Choose the installation instructions for your agent below, then start a fresh session if the skill is not listed.
  2. 02Provide the complete draft, intended reader and tone. List numbers, names, links and conditions that must not change.
  3. 03Ask the agent to use Humanizer. Request a revision and a brief explanation of the changes.
  4. 04Compare the result with the original. Check every factual claim and condition, then restore any deliberate wording the edit removed.

Check before installing

Compatible platforms

OpenAI Codex · Claude.ai · Claude Code

Best fit

Editing a draft you already have

Review the source instructions if the required output differs from the formats described above.

Install Humanizer for your platform

Choose the agent you use. The terminal commands require Node.js 22.20 or later and target the current project; Claude.ai uses a ZIP upload. A successful install still needs an activation check.

Claude CodeRun in your project terminal

This Skills CLI command installs the skill for Claude Code in the current project. Review the source first; the installer may ask you to confirm the target.

Installation command

npx skills add blader/humanizer --skill humanizer --agent claude-code
  1. Open Claude Code in the same project after installation.
  2. Use /humanizer with your draft, audience and constraints.

Check activation

Type / and look for humanizer. If it is missing, check that the project skill directory contains humanizer/SKILL.md, then reopen the session.

Claude Code skills documentation
OpenAI CodexRun in your project terminal

Install for Codex in the current project. This is separate from the Claude Code installation; leave off --global to keep the change scoped to this project.

Installation command

npx skills add blader/humanizer --skill humanizer --agent codex
  1. Start a new Codex session in the project after installation.
  2. Ask Codex to use the humanizer skill on your draft, or select it from the available skills.

Check activation

Confirm Humanizer appears among the available skills and ask the agent which skill it loaded. A rewritten draft alone does not prove the skill was used.

Codex skills documentation
Claude.aiUpload in Claude.ai · no terminal command

Download the repository ZIP from GitHub. Claude.ai uses an uploaded skill rather than the terminal commands above. Skill access depends on your account and organization settings.

  1. Open the source repository, choose Code, then Download ZIP.
  2. In Claude.ai, open Customize → Skills and upload the skill ZIP. Enable code execution and file creation if the interface requires it; an organization admin may need to enable skills.
  3. Enable Humanizer, open a conversation, and ask Claude to use it on the draft. If the upload rejects the archive, check that it contains a skill folder with SKILL.md at its root.

Check activation

Confirm the uploaded skill is enabled and inspect Claude’s skill-use indication. Check the current upload requirements in the linked help article if the interface differs.

Claude.ai upload and account requirements

Review the repository and MIT license before installing. Source review alone does not verify compatibility or constitute a security audit.

Give it a draft and clear constraints

After activating the skill, adapt this prompt to your reader and paste the complete draft. Add a short writing sample when matching your voice matters.

Starter prompt

Use the humanizer skill to edit the draft below.
Audience: existing customers.
Voice: direct, calm and specific.
Keep every product name, number, link and condition unchanged. Do not add features or promises.
Return the revised draft and a short explanation of the edits.

Draft:
[paste your complete draft here]

Before and after: two editing examples

Two fictional drafts edited with Humanizer in one Codex session on 15 November 2026. The revisions below are recorded outputs; your results may differ.

How these examples were tested

OpenAI Codex CLI 0.166.0 · gpt-6.1-sol · Humanizer 4.0.0.

We manually installed the skill in an isolated project and explicitly asked Codex to load it. We then checked both revisions against the original facts, numbers and conditions.

This test does not cover the installer commands, automatic skill discovery or other platforms.

A SaaS product announcement

Fictional product: Harbor. Audience: existing customers. Goal: announce a CSV export without adding a promise.

Before

We are thrilled to unveil a game-changing addition to Harbor that empowers teams to unlock the full potential of their data. Starting today, workspace admins can export up to 5,000 tasks as a CSV file. Export links remain available for 2 hours. This marks a pivotal step in our commitment to seamless workflows.

Recorded revision

Starting today, workspace admins can export up to 5,000 tasks from Harbor as a CSV file. Export links remain available for 2 hours.

The edit removes the ceremonial opening and unsupported praise. The feature arrives in the first sentence.

Facts and conditions to check

  • Harbor is still the product name.
  • Only workspace admins are named.
  • The 5,000-task limit, CSV format and 2-hour link window remain.
Try this input in your agent

A SaaS product announcement prompt

Use the humanizer skill to edit this announcement for existing customers. Keep Harbor, workspace admins, CSV, 5,000 tasks, 2 hours and the launch timing unchanged. Add no new promises. Return the revision and explain each edit.

Draft:
We are thrilled to unveil a game-changing addition to Harbor that empowers teams to unlock the full potential of their data. Starting today, workspace admins can export up to 5,000 tasks as a CSV file. Export links remain available for 2 hours. This marks a pivotal step in our commitment to seamless workflows.

API export instructions

Fictional API documentation. Audience: developers. Goal: simplify prose without weakening a required step.

Before

It is important to note that, in order to successfully retrieve your export, you must first send a GET request to /exports/{id}. The status must be ready before you download the file. Download links expire after 24 hours. If the response is 429, wait 60 seconds before retrying. These steps ensure a seamless and robust experience.

Recorded revision

To retrieve your export, first send a GET request to /exports/{id}. The status must be ready before you download the file. Download links expire after 24 hours. If the response is 429, wait 60 seconds before retrying.

The edit replaces the staged introduction with a direct instruction and removes the unsubstantiated closing claim.

Facts and conditions to check

  • GET and /exports/{id} are unchanged.
  • The ready status is still a prerequisite.
  • The 24-hour download-link expiry, 429 condition and 60-second wait remain.
Try this input in your agent

API export instructions prompt

Use the humanizer skill to edit this developer instruction. Preserve GET, /exports/{id}, ready, 24 hours, 429 and 60 seconds exactly. Keep the ready-status requirement and retry condition. Do not change API behavior.

Draft:
It is important to note that, in order to successfully retrieve your export, you must first send a GET request to /exports/{id}. The status must be ready before you download the file. Download links expire after 24 hours. If the response is 429, wait 60 seconds before retrying. These steps ensure a seamless and robust experience.

What to review before publishing

  • A smoother sentence can still be wrong. Check facts, quotations, citations and technical behavior yourself.
  • Keep deliberate style choices. A repeated phrase, dash or formal tone is not automatically a defect.
  • Use a normal editing prompt for a short one-off change. Humanizer is useful when you want a reusable review process across longer drafts.
  • This guide does not promise an AI-detector score or an undetectable result. Judge the writing by clarity, accuracy and fit for its reader.

Troubleshooting Humanizer

The skill does not appear after installation

Check which agent and project you installed for. A Claude Code installation does not automatically install for Codex. Verify that the skill folder contains SKILL.md, then start a fresh session in that project. Follow your agent’s linked skills documentation if discovery still fails.

Claude.ai rejects the ZIP or has no upload option

Check the archive structure and the current Claude.ai upload requirements. The skill folder must contain SKILL.md. Confirm code execution and file creation are enabled, and check organization policy if the Skills page or upload control is unavailable.

The output changes a fact or removes an important condition

Reject that edit. Supply a list of facts and conditions to preserve, then request a smaller revision. Compare names, numbers, links, commands and requirements with the original. Editing instructions cannot guarantee factual accuracy.

The result sounds too casual or loses my voice

Add a short sample of your own writing and identify the audience. Explicitly preserve terms or stylistic choices you want to keep. Ask for a conservative edit and an explanation of each change.

How can I tell whether the agent actually used Humanizer?

Explicitly select or name the skill, then inspect the agent’s skill-use indication or ask which file it loaded. A polished answer is not evidence of activation. If the skill was not loaded, check discovery before judging its results.

Continue exploring

Similar skills

Write internal communications — status reports, 3P updates, leadership notes, newsletters, FAQs, incident reports — in your company's formats.

Reviews docs and prose against Vercel's Writing Guidelines by fetching the latest rules and reporting findings in a terse file:line format.

Works withClaude CodeCursor+1

Generates, formats, and validates technical documentation — including docstrings, OpenAPI/Swagger specs, JSDoc annotations, doc portals, and user guides.

Works withClaude Code

Generate beautiful CHANGELOG.md from git commits following Keep a Changelog format

Works withClaude Code