The Five Core Patterns Humanizer Identifies in AI Text

Humanizer categorizes AI‑style artefacts into five core patterns—Staging, Rhythm by rule, Inflation, Formatting by rule, and Leftovers—to detect and rewrite mechanical prose according to the rules defined in SKILL.md.

The blader/humanizer repository is a lightweight skill that rewrites AI‑sounding prose by targeting specific rhetorical habits common to large language models. These habits are documented in SKILL.md and represent the default "write‑what‑most‑people‑would‑say" behavior that produces detectable artefacts. Understanding these five core patterns helps developers and writers identify why machine‑generated text feels artificial and how to restore natural cadence.

1. Staging

Staging occurs when text signals importance without adding factual content. In SKILL.md lines 21‑25, this pattern is defined by sentences that "stage" a point rather than stating it directly, such as "It’s not just X; it’s Y" or "Let’s dive in…".

This pattern inflates prose and distracts the reader. Human writers typically state facts directly, whereas AI models default to transitional phrases that add no informational value. Humanizer flags these staging constructions and removes them to tighten the narrative.

2. Rhythm by Rule

Rhythm by rule describes repeated structural habits imposed by the model regardless of semantic meaning. The specification identifies forced triads, pervasive em‑dashes, and other rhythmic devices that appear mechanically in SKILL.md.

This creates a cadence that sounds artificial because it ignores context. A human author varies sentence length and pacing naturally, while AI text applies uniform rhythmic templates. Humanizer detects these rule‑based repetitions and breaks them to restore organic flow.

3. Inflation

Inflation dresses ordinary facts as pivotal or expert‑backed statements. According to the "Inflation" bullet in SKILL.md, this includes over‑use of grandiose adjectives like "key", "pivotal", "robust", or "groundbreaking" without substantive backing.

Inflated language adds no new information and makes text sound promotional rather than informative. Humanizer identifies these intensifiers and either removes them or replaces them with specific, concrete descriptions that carry actual meaning.

4. Formatting by Rule

Formatting by rule refers to decorative formatting that adds no semantic value. The pattern includes bold labels, title‑case headings, decorative emojis, and uniform list styling that signal preset templates rather than writer intent, as documented in SKILL.md.

Such formatting serves as a visual tell that the content follows a model's default style guide. Humanizer strips these decorative elements or normalizes them to match the surrounding prose, eliminating the visual signature of AI generation.

5. Leftovers

Leftovers are remnants of chat‑style drafting that were never intended for the final reader. These include conversational wrappers like "I hope this helps", "Feel free to ask", or other drafting artefacts defined in the "Leftovers" section of SKILL.md.

These artefacts betray the text’s origin as a model‑generated draft rather than a polished human piece. Humanizer scans for these conversational signatures and removes them to produce final‑copy quality text.

How to Invoke Humanizer

Humanizer operates as a skill compatible with Claude and OpenAI‑compatible clients. It evaluates each sentence against the five core patterns, flags the strongest (most frequent) violations, and rewrites them while preserving the original information.

To invoke Humanizer from a Claude‑compatible client, send a JSON payload specifying the humanizer model:

{
  "model": "humanizer",
  "messages": [
    {
      "role": "user",
      "content": "It's not just about the beat riding under the vocals; it's part of the aggression and atmosphere. The new policy — announced without warning — affects thousands of workers."
    }
  ]
}

The skill automatically detects Staging ("not‑X‑but‑Y" constructions) and Formatting by rule (em‑dash usage), then returns a human‑style rewrite:

The heavy beat adds to the aggressive tone. The new policy, announced without warning, affects thousands of workers.

For OpenAI‑compatible integrations, use the descriptor in agents/openai.yaml:

model: humanizer
messages:
  - role: user
    content: |
      It's not just about the beat riding under the vocals; it's part of the aggression and atmosphere. The new policy — announced without warning — affects thousands of workers.

Key Source Files

The implementation of the five core patterns spans several files in the repository:

• SKILL.md – Core skill definition containing the pattern specifications and rewrite rules (lines 21‑25 define the initial pattern set).
• agents/openai.yaml – OpenAI‑compatible descriptor that exposes the skill to standard chat completions APIs.
• .claude-plugin/plugin.json – Claude plugin manifest that loads SKILL.md as a native skill.
• scripts/validate-package.py – Helper script that validates the skill’s internal consistency and pattern rule syntax.
• README.md – User‑facing documentation covering installation and advanced usage.

Summary

• Humanizer identifies five core patterns in AI text: Staging, Rhythm by rule, Inflation, Formatting by rule, and Leftovers.
• These patterns are defined in SKILL.md and represent mechanical habits like forced transitions, rhythmic templates, grandiose adjectives, decorative formatting, and chat‑style remnants.
• The skill can be invoked via Claude‑compatible or OpenAI‑compatible APIs using the descriptors in .claude-plugin/ and agents/.
• Humanizer preserves factual content while removing the rhetorical signatures that mark text as AI‑generated.

Frequently Asked Questions

What are the five core patterns Humanizer identifies in AI text?

Humanizer identifies Staging, Rhythm by rule, Inflation, Formatting by rule, and Leftovers. These patterns capture the most common AI‑style artefacts, from empty transitional phrases to decorative formatting, as specified in SKILL.md.

How does Humanizer detect these patterns in practice?

Humanizer evaluates each input sentence against the rules defined in SKILL.md, flags the strongest violations (those appearing most frequently), and applies targeted rewrites. The detection logic focuses on surface‑level rhetorical habits rather than semantic content, allowing it to strip mechanical style while preserving meaning.

Can I use Humanizer with standard OpenAI SDK clients?

Yes. The repository includes agents/openai.yaml, an OpenAI‑compatible descriptor that allows you to call Humanizer through standard chat completions endpoints. You reference model: humanizer in your request payload, and the skill processes the text according to the five core patterns.

Where are the rewrite rules for these patterns defined?

All rewrite rules and pattern definitions reside in SKILL.md at the repository root. This file contains the canonical specifications for Staging, Rhythm by rule, Inflation, Formatting by rule, and Leftovers, along with guidance on how to transform each pattern into natural prose.

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