I built the original Claude Watermark checker to separate real clipboard artifacts from unverifiable claims about statistical watermarking. The thing users kept asking for was: can my agent just clean the text I'm already looking at? This launch is that workflow. Install one portable Agent Skill, then ask Codex, Claude Code or Cursor to remove the Claude watermark from the last answer. The bundled CLI handles browser sign-in on first use, uses the same account and credits as the website, and retries after top-up without making you copy-paste the text again. The honest boundary is unchanged: deterministic cleanup removes verifiable HTML classes, zero-width characters, exotic spaces and punctuation artifacts. A rewrite resamples wording, but nobody outside Anthropic can verify a secret statistical watermark because no public detector exists. Would love feedback on the install flow and which agent you want supported next.

Claude Watermark Skill

Remove Claude watermarks inline in Codex, Cursor & Claude

Remove Claude watermarks inline in Codex, Cursor & Claude