# humanizer (agent instructions)

This repository is the `humanizer` skill: a pure-prompt instruction set that
rewrites AI-sounding prose so it reads as genuinely human, and rewrites in a
specific writer's voice when a sample or profile is available. No scripts, no
dependencies, no network access. It is the entry point for any AI coding tool
that reads `AGENTS.md` (Codex, OpenCode, Antigravity, Pi Coder, and others).

## When to apply this skill

Apply it whenever the user wants to humanize, de-slop, de-AI, or de-robotify
text; to fix writing that sounds like an LLM, corporate, salesy, generic, or
"off"; to make a draft sound like them or like a named author; or to edit
prose for authentic voice and rhythm. Apply it even when they do not say the
word "humanize" and do not name this skill.

## How to run it

Read `SKILL.md` in this repository and follow it exactly. The full method
lives there and in `references/`; do not improvise a shortcut. In brief:

1. **Step 0** discover the voice (pasted sample, named author, or a
   `VOICE.md` / `STYLE-GUIDE.md` in the working tree). No voice found means
   generic mode; never invent a persona.
2. **Step 0b** stance mode is off by default. Turn it on only if the user
   explicitly asks for more voice, edge, or opinion.
3. **Step 0c** density pre-check: skim for dead-giveaway tells and pick a
   light, standard, or full pass so human-first text is not over-edited.
4. **Multi-pass:** voice injection (if a voice exists), then tell removal
   against `references/tell-patterns.md` (32 patterns, six families), then a
   self-audit against `references/do-not-flag.md`.
5. Emit the exact output contract from `SKILL.md`: Humanized draft / What
   changed / Deliberately left alone / Meaning check / Next step.

## Hard rule (faithfulness over liveliness)

Never add a fact, number, date, name, quote, cause, mechanism, or example the
source did not contain. A fluent fabrication is worse than the fog it
replaced. When the source is thin, say less, do not invent. Preserve specific
details and genuine human quirks; restraint on already-human text is success,
not failure.

## Scope

This skill improves prose quality and authentic voice. It is not for defeating
plagiarism or AI-detection systems, and names no detector. Reframe such
requests toward genuine quality and voice (see `SKILL.md` "Scope and intended
use").