write-like-a-human

name	write-like-a-human
description	Strip the AI-isms out of any text — em-dash overuse, "It's not just X, it's Y", inflated significance, sycophantic openers, rule-of-three padding. Use when the user says "humanize this", "make this less AI", "tighten this", or pastes obviously-AI prose and asks for a rewrite.
license	MIT

Write Like a Human

Rewrite text so it stops sounding like a language model wrote it. Wraps the open-source humanizer skill, which catalogs ~30 specific tells (em-dash overuse, "stands as a testament", -ing analyses, rule-of-three padding, inline-bold lists, and so on).

When to use

• The user pastes a chunk of writing and asks to "humanize", "de-AI", "tighten", or "make this sound like me".
• The user asks for editing on something they wrote with an LLM.
• A blog post / launch email / README draft has tells the user wants gone.

Steps

1. Load the reference list of patterns the first time this skill activates in a session. On Spencer's machine, read the canonical local copy (newer than upstream):

   cat ~/.agents/skills/humanizer/SKILL.md
   

   Otherwise pull it from upstream:

   curl -fsSL https://raw.githubusercontent.com/blader/humanizer/main/SKILL.md
   

   There are ~30 patterns; you don't need to memorize them, just keep them visible while editing.

2. Ask for a voice sample if you don't already have one. One paragraph the user wrote themselves is enough. Match cadence, vocabulary, and how they handle transitions. For content written as Spencer, the voice layer is the spencers-writing-style skill in the stack repo (.agents/skills/spencers-writing-style/) — use it instead of asking.

3. Do a first pass removing the highest-signal tells in this order:

   1. Sycophantic openers ("Great question!", "Absolutely!").
   2. Em-dashes that should be commas, periods, or parens.
   3. "It's not just X, it's Y" / "Not only … but …" parallelisms.
   4. Inflated significance words: testament, pivotal, landscape, vibrant, tapestry, underscore.
   5. Rule-of-three lists that pad ideas into groups of three.
   6. Inline-bold vertical lists (- **Speed:** …).

4. Do a second "audit" pass. Re-read your draft and ask out loud: "What still makes this sound AI-generated?" Answer briefly, then revise once more. This step is the difference between "cleaned up" and "actually sounds human".

5. Return three things:

   • A draft rewrite.
   • A short list of remaining tells you noticed during the audit.
   • A final rewrite with those fixed.

Don't

• Don't replace AI mush with new AI mush. If the original said "stands as a testament to" and you change it to "represents a key milestone in", you haven't fixed anything.
• Don't sand off the user's voice. Quirks, opinions, and the occasional sentence fragment are the point.
• Don't add em-dashes back in. If you must use a punctuation break, prefer a period.

Reference

Patterns and examples: https://github.com/blader/humanizer