crimeacs

AI Tells Validator — AI skill for Claude Code

AI community

Detect and rewrite AI-sounding text.

How to install AI Tells Validator

This entry records only its repository, not the path inside it, so there is no exact command to give. Open crimeacs/ai-tells-validator and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What AI Tells Validator does

Detect and rewrite AI-sounding text. Grounded in the live Wikipedia 'Signs of AI writing' catalog.

Alternatives in AI

Humanizer — A Claude Code skill that removes signs of AI-generated writing from text, making it sound more natural and hum 11.3k ★
Academic Humanizer — Strip AI-writing tells from papers and grant proposals (NSF/NIH), while keeping scholarly voice and tying clai 1.1k ★
Humanizer Academic — A Claude Code skill that removes signs of AI-generated writing from academic medical papers, making them sound 169 ★

README

ai-tells-validator

A Claude Code skill (and standalone CLI) that **flags and rewrites AI-sounding text** — grounded in the live Wikipedia [`Signs of AI writing`](https://en.wikipedia.org/wiki/Wikipedia:Signs_of_AI_writing) catalog.

It refetches that page on every run, parses the bulleted lists into a structured rule set, augments with a hardcoded set of SaaS-pitch tells the article doesn't yet name, and enforces every rule with deterministic regex. Output is JSON for `validate` and clean text for `rewrite`.

If you're shipping cold-outbound, marketing copy, LinkedIn DMs, or any prose where "this reads as ChatGPT" is a trust-killer, this is your last-mile gate.

Install

As a Claude Code skill:

mkdir -p ~/.claude/skills
git clone https://github.com/crimeacs/ai-tells-validator ~/.claude/skills/ai-tells-validator

Claude Code auto-discovers it on next session.

As a standalone CLI:

git clone https://github.com/crimeacs/ai-tells-validator
cd ai-tells-validator
npm link   # optional — gives you `ai-tells-validate` and `ai-tells-rewrite` on PATH

No `npm install` step needed. The skill has zero runtime dependencies (Node 20+, native fetch, native test runner).

Quick start

# Find tells in a draft
echo "We delve into the intricate landscape of fraud detection." | node scripts/validate.mjs
# → exit 1, JSON report of every violation

# Rewrite a draft until it passes
cat draft.txt | node scripts/rewrite.mjs > clean.txt
# → exit 0 on success; stderr has the rewrite report

What it catches

The validator merges two sources of rules:

Live Wikipedia catalog — bulleted items pulled from sections like "Overused words", "Common phrases", "AI vocabulary", "Promotional language". Refreshed on every run with a 1-hour mtime cache.
Hardcoded augmentation — modern SaaS-pitch vocabulary and canned email tropes the Wikipedia article doesn't yet name explicitly (leverage, unlock, seamless, value proposition,