ai-watermark-removerlisted

# AI Text Watermark Remover Skill (Claude, Gemini, SynthID & GPT) This skill provides an end-to-end operational protocol to detect and eliminate all forms of AI text watermarking across **Claude (Anthropic)**, **Gemini & SynthID-Text (Google DeepMind)**, and **GPT (OpenAI)**. It targets three distinct watermarking layers: 1. **Invisible Steganography** (Zero-width characters, homoglyphs, invisible separators). 2. **Statistical / $n$-gram Logit Watermarks** (**Google SynthID-Text**, Kirchenbauer green/red list, Aaronson Gumbel watermarks). 3. **Model-Specific Stylistic Fingerprints** (Claude-isms, Gemini tropes, predictable cadence, and low-entropy formatting). --- ## 1. The Anatomy of AI Text Watermarks Modern AI text generators (including Claude, GPT, and Gemini) leave traces across three distinct vectors: | Watermark Type | Mechanism | Detection Method | Removal Strategy | | :--- | :--- | :--- | :--- | | **Steganographic / Unicode** | Invisible zero-width spaces (`\u200B`), non-joiners (`\u200C`), directional marks, or homoglyphs injected during copying/generation. | Byte-level scanning, Unicode inspect. | Deterministic stripping regex and NFKC normalization. | | **Statistical / $n$-gram Logits** | Pseudo-random token selection biasing $n$-gram transitions into a "green list" (SynthID, Kirchenbauer). | $z$-score hypothesis testing across token sequences. | Syntactic permutation, synonym swapping, clause shifting, breaking $n$-gram hash chains. | | **Stylistic / Entrop