humanize

humanize is a skill for Claude Code from kalyvask/winning-writing. It costs 109 tokens per session (2,000 once invoked), scanned A, original, MIT.

A writing aid that makes an overly polished draft sound more like something a person typed, using uneven phrasing, contractions, and occasionally a harmless typo.

In plain words
What is it for?
Use it to lightly roughen emails and other technically correct drafts when you want a more natural, less AI-clean voice.
Why use it?
It reduces the overly clean or artificial tone that can make a draft feel machine-written. It is not intended for high-stakes writing.

Skill for Claude Code

Written for Claude Code: shipped in a Claude Code plugin.

Part of the winning-writing plugin — 32 skills shipped together

Install

Getting it into your agent

One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.

agentmods
npx agentmods add skills/kalyvask/winning-writing/humanize
Any agent
npx skills add kalyvask/winning-writing --skill humanize
Clone the repo
git clone --depth 1 https://github.com/kalyvask/winning-writing

Made for: Claude Code.

Or install winning-writing, the plugin that ships this one along with the rest of its 32 skills.

Wrote this? Show the measurements

A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.

agentmods badge for humanize

README.md
[![agentmods](https://agentmods.dev/badge/skills/kalyvask/winning-writing/humanize.svg)](https://agentmods.dev/skills/kalyvask/winning-writing/humanize)
Your own site
<a href="https://agentmods.dev/skills/kalyvask/winning-writing/humanize"><img src="https://agentmods.dev/badge/skills/kalyvask/winning-writing/humanize.svg" alt="Measured on agentmods" height="20"></a>
Per session 109 Skills are progressive disclosure: only the name and description are preloaded; the body loads when the skill is used.
When invoked 2,000 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. Scan, not verified.
Origin original No closer match found in the catalogue.
Token cost

What it costs to keep this loaded

Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.

ModelPer sessionOnce invoked
Fable 5.1 $0.00109 $0.02000
Opus 5 $0.00055 $0.01000
Sonnet 5 $0.00022 $0.00400
Haiku 4.5 $0.00011 $0.00200

Measured 6d ago against content hash 4a014d647633, method: parsed. Prices are Anthropic first-party input rates as of 2026-09-06, from the pricing page.

Security

Grade A, and why

humanize scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 6d ago.

A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.

Nothing flagged

None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.

skills/humanize/SKILL.md · 139 lines

How it starts

The opening of the file, as written. The whole thing — 139 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Humanize

Source: points/ai-writing-rules.md and the centaur-writer thesis. The point of this skill is the inverse of every other skill in this repo — most of them sharpen, this one de-sharpens deliberately.

Operator note (2026-05-07): Dialed back from the original aggressive setting (too many typos and missing words landed in finals), then re-tuned for a middle ground: a few safe roughening moves are still welcome even in short pieces. The bias is "fewer types of typos, but still some texture" — not "skip everything."

The premise

A perfect email is suspicious. Real people:

  • Use contractions inconsistently ("it's" once, "it is" once in the same email — humans aren't consistent)
  • Vary sentence punctuation (sometimes a period where a comma would be cleaner)
  • Use parentheses inconsistently
  • Repeat a word from the previous sentence sometimes (real attention drift)

A model output is too clean. This skill leaves a small amount of real-person residue in. Note: residue, not damage. Missing required words and unsafe typos are damage; do not introduce them.

Two modes

Mode 1 — Shorten + roughen (default)

Take the draft and:

  1. Cut 10–20% of the words
  2. Convert most "I am / it is / they are / cannot" to contractions, but mix in one full form somewhere for inconsistency
  3. Drop the subject pronoun in one casual opener if the draft has one (never drop articles)
  4. Vary one sentence's punctuation in a slightly imperfect way (a period instead of a comma; a sentence fragment)
  5. Optionally combine two short adjacent paragraphs into one (or keep an aside on the same line instead of breaking) — humans don't always hit return where AI does
  6. Apply at most ONE safe typo (see below). Pieces under ~150 words: max 1 typo. Pieces over 300 words: still max 2 typos total. Never accumulate.

Mode 2 — Roughen only (preserve length)

Same as above, no length cut.

Safe roughening moves (use freely, even in short pieces)

These are not typos — they are rhythm choices that read as a real person rather than a model. Apply 1–2 per piece without worry:

Read the full file on GitHub · 139 lines

Changes

What this file has done since we first saw it

Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.

  1. 6d ago First seen · 139 lines · 109 tokens per session scan A 4a014d647633

Subscribe to this mod's changes

humanize is a skill published in the GitHub repository kalyvask/winning-writing (13 stars, last pushed today), licensed MIT. It adds 109 tokens to every session and 2,000 once invoked, about $0.0005 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.

Related

Other skills, from other repositories

project-memory

Generate a project-specific context file from a brief so an AI assistant remembers your editorial constraints, voice, audience, and quality bar across sessions.

ur-grue/autopunk-media-skills · 31 tokens

project-retrospective

Generate a LESSONS.md from a finished project: what worked, what didn't, what to reuse, what to retire — formatted for next-project carry-over.

ur-grue/autopunk-media-skills · 37 tokens

template-selector

Recommend the right skill bundle, agent, and workflow sequence for a new project — so media professionals can start producing instead of browsing a 394-skill library.

ur-grue/autopunk-media-skills · 35 tokens

jargon-flagger

Scans a draft and flags every instance of technical jargon, unexplained acronyms, and insider language that a general-audience reader would not understand — with a plain-language alternative for each.

ur-grue/autopunk-media-skills · 44 tokens

multi-author-harmonizer

Reviews a text written or assembled by multiple authors and produces a detailed inconsistency report — flagging voice shifts, terminology mismatches, tonal clashes, and formatting discrepancies — with specific harmonisation recommendations for each.

ur-grue/autopunk-media-skills · 47 tokens

passive-voice-checker

Identifies every passive-voice construction in a draft, assesses whether each weakens or serves the prose, and offers active-voice alternatives for those that should be changed — while leaving justified passives alone.

ur-grue/autopunk-media-skills · 48 tokens