← ClaudeAtlas

untelllisted

Humanize AI-generated text via a closed-loop, detector-feedback rewrite (the "untell" skill). Use when the user wants to humanize text, make writing sound more human or less like AI, reduce AI-detection scores, bypass or beat an AI detector (GPTZero, Turnitin, Originality, ZeroGPT), or lower the "AI probability" of a passage or file. Research/defensive tool — preserves meaning, citations, and facts.
ssamba1/untell · ★ 26 · AI & Automation · score 75
Install: claude install-skill ssamba1/untell
# untell Rewrite text so a local ensemble of AI-text detectors stops flagging it, while semantic meaning and all citations/numbers/quotes stay intact. The core technique is a **closed loop**: score → rewrite using the per-detector scores as feedback → re-score, repeating until the hardest detector is under threshold or the iteration cap is hit. You (Claude) are the rewriter; the local scripts only score and protect text. > **Research / defensive use.** Detectors are noisy proxies (non-native writers are falsely > flagged at high rates); these local detectors are *signals*, not ground truth. Do not present > output as guaranteed undetectable by any commercial system. State this if the user implies a > high-stakes deceptive use. ## When to run Trigger on requests like "humanize this", "make this sound less like AI", "reduce the AI score", "help this pass an AI detector". Input is either pasted text or a file path. ## Setup (once per session) Run all commands **from the skill directory** (the folder that contains this `SKILL.md`); the `python scripts/<name>.py` paths below are relative to it. The scripts work with **zero dependencies** (lite tier) and self-resolve their own package, so no `pip install` or `PYTHONPATH` is needed. For stronger detector signal, the user may `pip install -e ".[full]"` — the scripts auto-detect and use it; you don't change anything. (If `pip install`ed, the `untell-score` / `untell-sentences` / `untell-verify` console commands also work from an