What Happens If Humanizer Needs Information Not Present in the Source Text?

Humanizer never fabricates facts—when it encounters missing information, it stops and asks the user rather than inventing a substitute.

The blader/humanizer repository implements a strict "do not make anything up" architecture. This design is critical for content workflows where factual accuracy matters, such as technical documentation, legal writing, or published journalism. When the rewrite engine identifies a gap between the desired output and available source material, it follows a disciplined fallback protocol instead of hallucinating plausible-sounding details.

How Humanizer Detects Missing Information

The detection happens during pattern matching. In SKILL.md, section §23 (Knowledge‑limit disclaimers and guesses) specifies that the engine checks every claim against the original source before proceeding with any rewrite.

According to the source code at lines 25‑33 of SKILL.md, the pattern-matching logic flags sentences as incomplete when they require names, numbers, dates, quotes, citations, or other factual elements absent from the input. This prevents the engine from proceeding with unverified assumptions embedded in the text.

The Four-Step Fallback Workflow

Humanizer's response to information gaps follows a predictable sequence:

1. Detect the Gap – The rewrite engine validates each claim against the original source. Missing details trigger the pattern-matching logic defined in §23 of SKILL.md.

2. Prompt the User – Rather than inserting placeholder text or educated guesses, Humanizer generates a specific question requesting the needed information. The README.md at line 73 explicitly documents this behavior: "If a sentence needs a detail that is missing, Humanizer asks instead of inventing one."

3. Await a Response – The skill pauses execution until the user supplies the missing data. No rewrite proceeds with fabricated content.

4. Continue the Rewrite – Once the user provides the verified detail, the engine incorporates it and resumes pattern checking against remaining constraints.

Practical Examples of Humanizer's Missing-Information Handling

Example 1: Missing Date

User input:

/humanizer

The project was launched in <missing‑date>.

Humanizer response:

I need the launch date to complete the rewrite. Please provide the exact date.

After user supplies January 15 2024:

The project was launched on January 15 2024.

Example 2: Missing Citation

User input:

/humanizer

According to recent studies, the trend is increasing【?】.

Humanizer response:

I don't have a source for that claim. Could you share the citation or remove the statement?

These examples demonstrate how Humanizer maintains factual integrity through explicit user collaboration rather than automated speculation.

Configuration Reference

Three files govern this behavior in the blader/humanizer repository:

• SKILL.md – Contains the complete pattern list and §23's rule for knowledge‑limit disclaimers and guesses
• README.md – Documents the "ask instead of invent" workflow at line 73
• AGENTS.md – Provides agent compatibility metadata that integrates the skill into broader automation systems

Summary

• Humanizer's core principle prohibits fabrication – no hallucinated facts enter the output stream
• Gap detection occurs through pattern matching against §23 in SKILL.md
• User prompting replaces automated guessing – explicit questions replace placeholder insertion
• Workflow pauses for verification – execution resumes only after user-supplied confirmation
• Design prioritizes safety for high-stakes content – technical, legal, and journalistic use cases benefit from guaranteed accuracy

Frequently Asked Questions

Does Humanizer ever guess when information is missing?

No. The blader/humanizer codebase contains no fallback mechanism for automated guessing. The README.md explicitly states the skill "asks instead of inventing," and SKILL.md §23 codifies this as a hard constraint. Every missing detail triggers a user prompt.

What types of information gaps trigger Humanizer's questioning?

Any factual element not present in the source text—including names, dates, numerical data, direct quotes, and citations—activates the detection logic. The pattern-matching engine in SKILL.md evaluates semantic categories of claims rather than simple string matching.

Can Humanizer complete a rewrite with partial information?

The pause-and-resume design means rewrites proceed incrementally. Humanizer incorporates verified details as users supply them, then continues evaluating remaining patterns against the updated context. No section finalizes with unverified content.

Is this behavior configurable or can it be overridden?

The source files show no configuration toggle for disabling the "ask don't invent" protocol. The architecture treats factual integrity as non-negotiable, making Humanizer suitable for environments where hallucination risks outweigh convenience benefits.

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