humaniserlisted
Install: claude install-skill kvokov/oh-my-ai
# Humaniser
Strip generic "model voice" from text: fewer stock phrases and signposts, more direct sentences, and (when requested) a closer match to the user's own writing. The aim is **human-sounding text without new false specifics**.
## Non-negotiables (do not skip)
1. **Run interactive intake first** using [interactive-intake](./rules/interactive-intake.md). Do not rewrite until goal, register, preservation rules, and uncertainty handling are settled. Every AskUserQuestion set must include an **uncertainty** option (e.g. "I'm not sure — help me decide").
2. **Preserve truth and stance.** Do not invent statistics, quotes, citations, names, dates, or product facts. Do not reverse or soften the author's position unless explicitly asked.
3. **Apply the hunt list** from [ai-tells-and-patterns](./rules/ai-tells-and-patterns.md) during the first rewrite and again in the audit pass.
4. **Voice calibration:** when the user supplies a writing sample, match its rhythm and diction; do not replace it with a generic "clean" voice.
5. **Do a second pass** after the first rewrite: re-read only for lingering AI tells (signposting, em dash clusters, promotional adjectives, chatbot wrappers). Fix what still sounds synthetic; if the first pass already reads human, keep edits minimal.
6. **Deliver clearly.** Default output is the rewritten text. If the user asks how it changed, add a **short** bullet log (max 5–8 bullets) of categories touched — not a line-by-line diff unless requested.
7.