agentsclimarketplace

Context aware humanizer

Skill M1w234/context-aware-humanizer/skills/context-aware-humanizer

A context-aware writing skill that humanizes outward-facing prose without flattening voice or changing meaning.

Install
npx -y skills add M1w234/context-aware-humanizer --skill context-aware-humanizer

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

3 things to look at

  • 14 days oldThe repository was created 14 days ago. New is not bad, but a brand new repository carrying a familiar-sounding name is the shape a typosquat arrives in, and there has been no time for anyone else to find a problem with it.
  • no licenseNo license file was found in the repository. Code published without one is not open source by default, so using it at work is a question for whoever answers licensing questions where you are.
  • 0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its author says it does

Copied from the file, not written here

Make final outward-facing prose natural, specific, and appropriate for its real reader without flattening the author's voice or changing meaning by accident. Use implicitly only when producing or revising email, SMS, DMs, replies, outreach, professional updates, final blogs, articles, newsletters, captions, announcements, or marketing copy. Also use when asked to humanize, polish, rewrite, tighten, soften, or make prose less AI-sounding. Do not trigger merely to read, summarize, critique, discuss, brainstorm, research, analyze, or work with private notes, internal analysis, internal documents, code, or data unless the requested result is final prose for another person or public audience. Creative, academic, legal, contractual, regulated, and technical writing is explicit only or handled by its specialized workflow.

SKILL.md

5.1 KB, 900 tokens by cl100k_base, as published. Nobody here has run it

Context-Aware Humanizer

Edit for the reader and purpose, not for an AI-detector score. Preserve the author's recognizable voice. If the prose already works, make a small edit or no edit.

Core pass

  1. Write for the actual person or audience. Match the relationship, channel, context, purpose, and desired response.
  2. Get to the real point without ceremonial setup. Keep necessary context, but remove explanations the reader does not need.
  3. Preserve specific details, useful texture, real opinions, and genuine uncertainty. Do not round them into generic claims or artificially tidy conclusions.
  4. Remove model habits that make the passage feel canned, especially when several appear together. Do not ban an isolated word or construction that fits the author and context.
  5. Read the result in the intended voice. Vary rhythm naturally without adding fake slang, typos, anecdotes, vulnerability, or quirky punctuation.
  6. During a style-only edit, keep the intended meaning intact. Change facts, asks, commitments, approvals, authorization, or other substance when the user explicitly requests that change.

Use voice and workflow context

  • Match approved samples and confirmed preferences already in scope. If references/voice-profile.md exists, follow it. Do not silently search private communications for samples or infer a complete voice from rough dictation.
  • Let a specialized writing or domain workflow lead. Act as its final prose pass only when that workflow invokes this skill.

Match the requested depth

  • Polish or humanize: Keep the structure and meaning. Make the smallest edit that solves the problem.
  • Rewrite or improve: Restructure when it materially improves clarity, pacing, tone, or usefulness.
  • Draft from notes: Compose from supplied material without inventing facts, experiences, feelings, positions, or sources.
  • Exact or verbatim: Do not rewrite protected text.

Do not turn a light edit into a wholesale rewrite.

Do not overcorrect

  • Treat model habits as clusters and context clues, not banned vocabulary.
  • Do not remove personality, dialect, humor, fragments, contractions, emojis, repetition, passive voice, unusual syntax, or genre conventions merely because they can appear in AI writing.
  • Do not make every message shorter, friendlier, firmer, more persuasive, more confident, or more casual.
  • Do not infer a decision from a question. When the user actually makes or changes a decision, state it normally.
  • Preserve quotations, citations, links, code, commands, identifiers, and exact language unless the user asks to edit them.

Handle specialized contexts carefully

When the skill is explicitly invoked for a specialized form:

  • For creative writing, preserve point of view, character voice, tense, scene facts, imagery, ambiguity, and intentional rhythm unless the requested edit targets them.
  • For memoir or personal narrative, invent no memory, dialogue, motive, sensory detail, or emotional closure.
  • For academic, legal, contractual, regulated, or technical prose, default to surface editing. Preserve citations, defined terms, obligations, qualifiers, commands, identifiers, units, and normative language. Flag a potentially meaning-changing edit instead of silently making it.

Load optional references

Read references/model-habits.md when:

  • A draft still feels noticeably generated or generic after the core pass.
  • The user asks what makes the writing sound AI-generated.
  • Editing long-form public copy where several patterns may interact.

Do not load the catalog for every short message.

Read references/editorial-principles.md only when auditing or updating this skill, or when the user asks why it differs from detector-oriented humanizers.

assets/voice-profile-template.md is an inactive template. Do not treat its prompts or examples as user preferences.

Return the result

  • When applied implicitly or embedded in another workflow, return only the final prose. Do not announce the edit or narrate the analysis.
  • For an explicit edit, return one recommended version. Add a short note only when a real ambiguity or risk remains.
  • Offer alternatives only when the user genuinely needs to choose between different tones.

Keep looking

Skills are one crate of 328,083. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.