What Is the Final Step in the Humanizer Workflow?

The final step in the Humanizer workflow is "Write the final version," where the editor produces a polished, human-sounding document by naturally stating each point, rewriting awkward sentences within their surrounding context, and varying sentence length to eliminate AI-style tells.

The blader/humanizer repository provides a structured editing skill designed to strip AI-generated patterns from text while preserving factual accuracy. Defined in SKILL.md, the workflow follows a rigorous four-stage process that transforms mechanical prose into authentic human writing. Understanding this culminating phase ensures users can effectively leverage the OpenAI-compatible plugin defined in agents/openai.yaml to achieve publication-ready results.

The Four-Stage Editing Process

According to the source code in SKILL.md, the Humanizer skill implements a sequential editing process before reaching its conclusion. The workflow progresses through distinct phases: marking AI "tells" in the source text, drafting a comprehensive rewrite, checking the draft against the original facts, and finally executing the final step in the Humanizer workflow (lines 31-39). This structured approach ensures that content undergoes thorough refinement rather than superficial word substitution.

The Final Step: Write the Final Version

The terminal phase documented at lines 31-39 of SKILL.md is explicitly labeled "Write the final version." During this stage, the editor performs three critical actions to ensure the output reads authentically human:

State Each Point Naturally

Rather than patching individual flagged phrases in isolation, the editor reconstructs the entire point using natural language. This holistic approach prevents the "band-aid" effect that often characterizes AI-edited text, ensuring ideas flow logically rather than reading as disconnected fixes.

Rewrite Awkward Sentences by Reshaping Context

The editor examines sentences flagged as awkward and rewrites them by reshaping the surrounding paragraph. This technique addresses syntactic stiffness at the structural level, allowing the text to breathe organically rather than forcing rigid AI constructions into slightly different rigid constructions.

Vary Sentence Length

To mimic authentic human rhythm, the editor deliberately varies sentence length throughout the document. This variation breaks the monotonous cadence typical of AI generation, creating dynamic pacing that reflects natural human speech patterns and writing styles.

Implementing the Humanizer Workflow

You invoke the final step through the OpenAI-compatible plugin defined in agents/openai.yaml. The following examples demonstrate how to trigger the complete workflow, including the final version generation:

Using the OpenAI Plugin Prompt

**Prompt to Humanizer (OpenAI plugin)**
Use $humanizer to rewrite this text in my voice without changing its facts.

> AI‑generated text:
> “It’s not just about the UI; it’s about the experience. The real win is better engagement. Let that sink in.”

**Humanizer’s output (final version)**
“The UI improves the overall experience, leading to higher engagement.”

Programmatic Implementation

# Example using the OpenAI-compatible YAML definition

import openai

prompt = """Use $humanizer to rewrite this text in my voice without changing its facts.

> It's not just about the UI; it's about the experience. The real win is better engagement. Let that sink in.
"""

response = openai.ChatCompletion.create(
    model="gpt-4o-mini",
    messages=[[{"role": "user", "content": prompt}},
)

print(response.choices[0].message.content)

# → “The UI improves the overall experience, leading to higher engagement.”

These implementations call the skill defined in SKILL.md, automatically progressing through all four stages to return only the final polished version.

Summary

• The final step in the Humanizer workflow is explicitly defined as "Write the final version" in SKILL.md (lines 31-39).
• This stage requires stating points naturally rather than patching phrases, rewriting awkward sentences by reshaping surrounding paragraphs, and varying sentence length to mimic human rhythm.
• The workflow is implemented through the OpenAI-compatible plugin configuration in agents/openai.yaml.
• Successful execution produces a factually accurate document that eliminates AI-style "tells" through structural editing rather than superficial word replacement.

Frequently Asked Questions

What file contains the definition of the Humanizer workflow?

The core workflow definition resides in SKILL.md at the root of the blader/humanizer repository. This file specifies the four-stage editing process, including the detailed instructions for the final step at lines 31-39.

How does the final step differ from earlier stages in the workflow?

While earlier stages focus on identifying AI tells and drafting preliminary rewrites, the final step executes holistic revision. Instead of editing individual phrases, it reshapes entire paragraphs and varies sentence structure to create organic flow, producing the definitive human-sounding version.

Can I use the Humanizer workflow with any OpenAI-compatible model?

Yes, the repository provides agents/openai.yaml, an OpenAI-compatible plugin definition that enables integration with models like GPT-4o-mini. As shown in the Python example, you can invoke the $humanizer skill through standard OpenAI API calls to trigger the complete workflow including the final step.

What specific AI writing patterns does the final step target?

According to SKILL.md, the final step specifically targets mechanical sentence rhythms, isolated phrase-level edits, and awkward syntactic structures. By reshaping surrounding context and varying sentence length, it eliminates "tells" such as the phrase "Let that sink in" and other formulaic AI constructions.

{
  "mcpServers": {
    "instagit": {
      "command": "npx",
      "args": ["-y", "instagit@latest"]
    }
  }
}