avoid-ai-writing

name	avoid-ai-writing
description	Audit and rewrite content to remove AI writing patterns ("AI-isms"). Use this skill when asked to "remove AI-isms," "clean up AI writing," "edit writing for AI patterns," "audit writing for AI tells," or "make this sound less like AI." Supports detect-only, rewrite (default), and edit-in-place modes, voice profiles, and iterative convergence.
category	quality
license	MIT
version	3.10.0

Avoid AI Writing — Audit & Rewrite

You are editing content to remove AI writing patterns ("AI-isms") that make text sound machine-generated.

When to Use

• When asked to "humanize" or "clean up" AI-generated text.
• Before publishing agent-generated READMEs, changelogs, or documentation.
• When auditing content for statistical AI tells.

Core Modes

• rewrite (default): Flag AI-isms and rewrite text to fix them. Returns Issues, Rewritten text, What changed, and a Second-pass audit.
• detect: Flag AI-isms only. No rewriting. Useful for CI/CD gates or audits.
• edit: Edit a file in-place with minimal, targeted changes using the Edit tool.

Voice Profiles (Optional)

Independent of audience context; sets the persona.

• casual: Contractions, short sentences (≤14 words), low jargon.
• professional: Active voice, concrete claims, low hedging.
• technical: Plain copulatives (is/has), one idea per sentence, imperative mood.
• warm: Direct address ("you"), strong verbs, medium cadence (15-20 words).
• blunt: Short declaratives, no padding, no hedges.

Context Profiles

Adjusts rule strictness for specific audiences.

• linkedin: Social-first; relaxed on formatting, strict on P0 credibility killers.
• blog: Default; all rules full strength.
• technical-blog: Technical terms (robust, leverage, etc.) exempt; strict prose rules.
• investor-email: Extra strict on promotional language and significance inflation.
• docs: Clarity over voice; relaxed on hedging and copula avoidance.
• casual: Slack/DM style; P0 credibility killers only.

Instructions

1. Audit: Identify AI-isms using the references/patterns.md catalog.
2. Apply Profile: Adjust strictness based on context and tone based on voice.
3. Execute Mode:
   • rewrite: Return audit, full rewrite, change summary, and second-pass audit.
   • detect: Return severity-grouped issues (P0/P1/P2) and assessment.
   • edit: Apply minimal Edit tool changes to target file; return edit report.
4. Convergence: If --iterate 2 is passed, repeat until no patterns remain (max 2 passes).

Severity Tiers

• P0 — Credibility killers: Cutoff disclaimers, chatbot artifacts, vague attributions.
• P1 — Obvious AI smell: Word-list violations, template phrases, "Let's" openers.
• P2 — Stylistic polish: Generic conclusions, uniform length, transition overuse.

Rationalizations

Rationalization	Reality
"AI writing is grammatically perfect, so it's fine."	Perfection is a tell; natural human writing has varied rhythm and personality.
"I can just ask the LLM to 'sound human'."	One-shot "sound human" prompts are vibes-based; this skill uses a structured audit.
"This will flag non-native speakers."	Patterns are signals, not proof. Use judgment and pair with context.

Red Flags

• Sanding away all personality in pursuit of "clean" prose (creating new uniformity).
• Rewriting quoted material or code blocks.
• Ignoring the Second-pass audit (patterns often survive the first edit).

References

• references/patterns.md — Exhaustive catalog of AI patterns and vocabulary.

See Also

• ui-ux-optimize — For optimizing UI/UX specific copy and layouts.
• readme-best-practices — For documentation structure and discovery.

Voice & Context

• Default: professional + blog
• Reference: voice-profiles skill for definitions and auto-detection.