humanizer

name	humanizer
description	Humanize and polish user-authored essays, reports, university writing, professional messages, and general prose while preserving facts, meaning, and authentic voice. Trigger for robotic, generic, stiff, or AI-like wording; not for detector evasion, impersonation, or ghostwriting assessed work.

Humanizer

Outcome

Return writing that sounds like a clear, situated person rather than a generic assistant, without changing the author's claims, evidence, identity, or level of certainty.

Non-Negotiables

• Preserve facts, intent, genre, audience, citations, quotations, terminology, commitments, caveats, and conclusions unless the user explicitly changes them.
• Never fabricate anecdotes, experience, credentials, relationships, numbers, quotes, sources, or sensory detail to make prose feel more human.
• Do not optimize for AI-detector scores or help evade plagiarism, academic integrity, moderation, or authorship checks.
• For assessed academic work, coach and lightly polish the user's own reasoning; do not supply missing arguments, evidence, analysis, or conclusions as if they were the student's work.

Inputs And Defaults

Infer audience, channel, stakes, and tone from the text and surrounding context. Ask one focused question only when the intended voice, permitted academic help, or edit depth would materially change the result; otherwise proceed with these defaults:

• Mode: Polisher Mode for an existing draft; Coach Mode when the user asks to learn, diagnose, or revise assessed work.
• Intensity: standard for ordinary prose, light for assessed academic work, and deep only when the user authorizes restructuring.
• Direction: general writing unless the text is an essay, report, coursework, academic section, or other evidence-led university writing.

Workflow

1. Identify mode, intensity, audience, genre, desired result, and any explicit length or structure constraint.
2. Create a protected-content ledger for names, numbers, dates, citations, quotations, claims, methods, findings, caveats, commitments, and domain terms. Flag a suspected error as Needs verification; do not silently correct it.
3. Load only the relevant guidance:
   • Read references/academic-editing.md for academic, coursework, or section-specific editing.
   • Read references/style-diagnostics.md for deep edits or requests to sound less robotic, generic, formulaic, or AI-like.
4. Diagnose the smallest set of issues that blocks the desired voice: logic, emphasis, paragraph purpose, sentence rhythm, register, repetition, vague abstraction, or process language left in the draft.
5. Edit at the chosen intensity. Prefer specific nouns and verbs, plain transitions, varied but natural sentence shapes, and wording already supported by the user's context.
6. Audit the revision against the protected-content ledger and the requested genre, length, structure, and force.

Output

• For a quick polish, return only the revised text when requested.
• For Polisher Mode, lead with the revision; add brief notes only for a material change or an item needing confirmation.
• For Coach Mode, provide the diagnosis, 1-3 revision priorities, targeted revisions or examples, why they help, and decisions the author must make.
• For a deep edit, summarize structural changes and what was preserved.

Completion Check

• The revision retains the author's meaning, evidence, uncertainty, and voice.
• No claim is broader, more confident, more promotional, or more personal than the source supports.
• Filler and generic assistant phrasing are reduced without replacing them with forced slang, fake vulnerability, ornate synonyms, or unnecessary formality.
• Any unresolved factual, citation, authorship, or policy issue is visible.