How Humanizer Handles Quoted Text and Proper Names in Source Material

Humanizer treats quoted passages and proper names as immutable literals, preserving them exactly as they appear while rewriting surrounding AI-styled prose.

The blader/humanizer repository provides a specialized rewriting system designed to de-AI-ify text while maintaining strict fidelity to source material. When processing content containing quoted text or proper names, the tool implements specific safeguards to ensure these elements remain untouched. Understanding how Humanizer handles quoted text or proper names is essential for anyone using the system to rewrite technical documentation, academic papers, or branded content.

Core Preservation Rules in SKILL.md

The behavioral constraints governing quotation and proper name handling are defined in the SKILL.md file at the repository root. These rules establish immutable boundaries for the rewriting process.

The Secondhand Text Clause (§14)

According to lines 14-15 of SKILL.md, Humanizer implements a "Secondhand text" clause that explicitly prohibits rewriting within specific boundaries. The directive instructs the model to skip any rewrite inside:

• Quotations and cited passages
• Proper names and titles
• Brand names and product identifiers
• Examples being discussed rather than used

This ensures that names like React-Native or titles like "The Great Gatsby" survive the humanization process as exact copies of the source text.

Quote Handling and Style Corrections (§19)

While the system may flag stylistic choices related to quotation marks, the content within those quotes remains sacrosanct. Lines 27-33 of SKILL.md specify that curly quotes are identified as a style issue only for the purpose of standardizing the quotation mark characters themselves—replacing curly quotes with straight quotes if the source uses straight quotes. However, the system does not rewrite the quoted string itself, preserving the original wording, terminology, and phrasing exactly as authored.

Post-Rewrite Verification Process

After each rewrite iteration, Humanizer executes a mandatory fact-preservation check defined in lines 45-47 of SKILL.md. The verification step poses a specific question to the model: whether any fact, name, number, date, quote, citation, or ranking was added or removed during the rewrite process. If the system detects that a quoted segment was unintentionally altered, the model must immediately revert the change. This creates a secondary safeguard ensuring that proper nouns and direct quotations remain immutable even if initial processing attempted modification.

Practical Code Examples

The following examples demonstrate how Humanizer processes content containing protected text elements.

Example 1: Direct Quotation Preservation

Input:

She said “AI will change the world” in her keynote.

Output:

She said “AI will change the world” in her keynote.

In this case, the quoted phrase AI will change the world remains identical between source and output, while surrounding prose may be adjusted for style.

Example 2: Proper Name and Product Title Protection

Input:

The new framework React-Native 2.0 introduces several performance improvements.

Output:

The new framework React-Native 2.0 introduces several performance improvements.

Here, the product name React-Native 2.0 is preserved exactly, including version numbering and hyphenation, ensuring brand consistency and technical accuracy.

Summary

• Immutable literals: Humanizer treats quoted text and proper names as untouchable elements within the rewriting pipeline.
• SKILL.md governance: Lines 14-15 establish the "Secondhand text" clause protecting quotations, titles, and proper nouns from modification.
• Character-level precision: While curly quotes may be standardized to straight quotes (lines 27-33), the content within quotation marks never changes.
• Verification safeguard: A post-processing check (lines 45-47) requires confirmation that no facts, names, dates, or quotes were altered during rewriting.
• Brand protection: Product names, technical terms, and cited works survive the humanization process as exact replicas of the source material.

Frequently Asked Questions

Does Humanizer ever change the wording inside quotation marks?

No. According to the SKILL.md specifications at lines 27-33, Humanizer explicitly preserves the content within quotation marks. While the system may standardize curly quotes to straight quotes for stylistic consistency, the actual text between the quotation marks remains an immutable literal that survives the rewrite process unchanged.

How does Humanizer handle technical product names like programming frameworks?

Technical product names, brand identifiers, and versioned titles are protected under the "Secondhand text" clause in lines 14-15 of SKILL.md. The system recognizes these as proper nouns and skips rewriting within these boundaries, ensuring that names like React-Native 2.0 or Python 3.11 remain exactly as they appear in the source material.

What happens if Humanizer accidentally modifies a proper name during rewriting?

If a proper name or quotation is unintentionally altered, the fact-preservation verification step (lines 45-47 of SKILL.md) catches the error. This post-rewrite check asks whether any fact, name, number, date, quote, or citation was added or removed. Upon detecting a discrepancy, the model must revert the change to maintain source fidelity.

Are academic citations and source references protected during humanization?

Yes. Academic citations, bibliographic references, and quoted excerpts fall under the protected categories defined in SKILL.md. The system treats these as "secondhand text" and excludes them from the rewriting process, ensuring that citations, author names, publication titles, and quoted passages remain accurate for scholarly integrity.

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