How Humanizer Handles Inflated Claims in Text: Pattern Detection and Rewrite Flow

Humanizer treats inflated claims about importance and legacy as Pattern #1, scanning text for specific cue words like "pivotal" or "marks a shift" to strip grandiose framing while preserving factual core content.

The blader/humanizer repository provides a specialized skill for identifying and neutralizing AI-generated rhetorical excess. When processing prose, the tool specifically targets how inflated claims in text are handled through systematic pattern matching defined in SKILL.md. This approach ensures that exaggerated significance is removed without sacrificing accurate dates, names, or sourced details.

The Inflated Claims Detection Mechanism

According to the source code in blader/humanizer, Pattern #1 identifies prose that exaggerates the importance or legacy of events and entities. The skill maintains a specific word list of cue terms that trigger this classification.

In SKILL.md (lines 50-58), the pattern definition includes trigger phrases such as:

• "stands as" or "serves as"
• "pivotal" or "key role"
• "marks a shift" or "marking a pivotal moment"

When the engine detects any of these linguistic markers, it flags the sentence for transformation under the inflated claims protocol.

The Three-Step Rewrite Process

The humanization workflow follows a strict pipeline defined in the skill documentation:

Step 1: Pattern Identification

The engine performs regex-based scanning against the cue word list defined in SKILL.md §1. Upon matching terms like "pivotal moment" or "key role," the system categorizes the input as requiring Pattern #1 processing.

Step 2: Claim Preservation

Despite removing embellishments, Humanizer enforces the "Keep every claim" rule documented in SKILL.md (lines 23-25). The system extracts and retains all factual data—including specific dates, proper nouns, and verifiable statistics—ensuring no substantive information is lost during de-inflation.

Step 3: Fact-Only Output

The "Do not invent facts" rule (SKILL.md, lines 23-26) strictly prohibits hallucination. The rewritten sentence conveys identical factual meaning using neutral language, eliminating grandiose framing without introducing new assertions or speculative context.

Practical Example: Before and After

The following transformation demonstrates Pattern #1 execution on a sentence containing inflated significance:

Input (Inflated):

The Statistical Institute of Catalonia was officially established in 1989, marking a pivotal moment in the evolution of regional statistics in Spain.

Output (Humanized):

The Statistical Institute of Catalonia was established in 1989, part of a wider decentralization of administrative functions in Spain.

This example from SKILL.md (lines 55-57) shows the removal of "marking a pivotal moment" while preserving the establishment date and institutional name.

Invoking the Skill

To process text via the Claude-compatible interface, use the following markdown syntax:

/humanizer

The Statistical Institute of Catalonia was officially established in 1989, marking a pivotal moment in the evolution of regional statistics in Spain.

For programmatic access, structure the request as JSON:

{
  "skill": "humanizer",
  "input": "The Statistical Institute of Catalonia was officially established in 1989, marking a pivotal moment in the evolution of regional statistics in Spain."
}

Implementation Context

The inflated claims pattern operates as one of 35 core content patterns identified in the repository. According to README.md (lines 61-63), these patterns collectively target specific AI-writing artifacts, with inflated claims representing a fundamental category of rhetorical overstatement commonly found in generated text.

The pattern matching relies on static analysis of cue words rather than neural classification, making the detection deterministic and explainable. This architectural choice ensures consistent handling of exaggerated language across different input domains.

Summary

• Pattern #1 specifically targets inflated claims about importance and legacy using a defined cue word list in SKILL.md (lines 50-58).
• The engine preserves all factual claims including dates and names while removing grandiose language, following the "Keep every claim" rule.
• "Do not invent facts" enforcement prevents hallucination during the rewrite process.
• Both markdown and JSON interfaces support invoking the skill for programmatic text processing.
• The repository defines 35 total patterns for comprehensive AI text humanization.

Frequently Asked Questions

What specific phrases trigger the inflated claims pattern?

Humanizer scans for explicit cue words defined in SKILL.md including "pivotal," "key role," "stands as," "serves as," and "marks a shift." The presence of any these terms flags the sentence for Pattern #1 processing.

How does Humanizer ensure factual accuracy when removing inflated language?

The skill follows two binding rules from SKILL.md (lines 23-26): "Keep every claim" preserves all factual data including dates and proper nouns, while "Do not invent facts" prevents adding unsourced information during the rewrite.

Where are the pattern definitions stored in the repository?

All detection patterns, including the inflated claims word list and transformation rules, are documented in SKILL.md at the repository root. The README.md (lines 61-63) provides high-level context about the 35-pattern architecture.

Can Humanizer handle multiple inflated claims in a single text?

Yes. The engine processes text iteratively, applying Pattern #1 wherever cue words appear. Each sentence undergoes independent analysis, allowing simultaneous handling of multiple instances of rhetorical exaggeration throughout the document.

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